xref: /openbmc/qemu/target/i386/tcg/translate.c (revision 5ac034b1)
1 /*
2  *  i386 translation
3  *
4  *  Copyright (c) 2003 Fabrice Bellard
5  *
6  * This library is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * This library is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18  */
19 #include "qemu/osdep.h"
20 
21 #include "qemu/host-utils.h"
22 #include "cpu.h"
23 #include "disas/disas.h"
24 #include "exec/exec-all.h"
25 #include "tcg/tcg-op.h"
26 #include "tcg/tcg-op-gvec.h"
27 #include "exec/cpu_ldst.h"
28 #include "exec/translator.h"
29 #include "fpu/softfloat.h"
30 
31 #include "exec/helper-proto.h"
32 #include "exec/helper-gen.h"
33 #include "helper-tcg.h"
34 
35 #include "exec/log.h"
36 
37 #define PREFIX_REPZ   0x01
38 #define PREFIX_REPNZ  0x02
39 #define PREFIX_LOCK   0x04
40 #define PREFIX_DATA   0x08
41 #define PREFIX_ADR    0x10
42 #define PREFIX_VEX    0x20
43 #define PREFIX_REX    0x40
44 
45 #ifdef TARGET_X86_64
46 # define ctztl  ctz64
47 # define clztl  clz64
48 #else
49 # define ctztl  ctz32
50 # define clztl  clz32
51 #endif
52 
53 /* For a switch indexed by MODRM, match all memory operands for a given OP.  */
54 #define CASE_MODRM_MEM_OP(OP) \
55     case (0 << 6) | (OP << 3) | 0 ... (0 << 6) | (OP << 3) | 7: \
56     case (1 << 6) | (OP << 3) | 0 ... (1 << 6) | (OP << 3) | 7: \
57     case (2 << 6) | (OP << 3) | 0 ... (2 << 6) | (OP << 3) | 7
58 
59 #define CASE_MODRM_OP(OP) \
60     case (0 << 6) | (OP << 3) | 0 ... (0 << 6) | (OP << 3) | 7: \
61     case (1 << 6) | (OP << 3) | 0 ... (1 << 6) | (OP << 3) | 7: \
62     case (2 << 6) | (OP << 3) | 0 ... (2 << 6) | (OP << 3) | 7: \
63     case (3 << 6) | (OP << 3) | 0 ... (3 << 6) | (OP << 3) | 7
64 
65 //#define MACRO_TEST   1
66 
67 /* global register indexes */
68 static TCGv cpu_cc_dst, cpu_cc_src, cpu_cc_src2;
69 static TCGv cpu_eip;
70 static TCGv_i32 cpu_cc_op;
71 static TCGv cpu_regs[CPU_NB_REGS];
72 static TCGv cpu_seg_base[6];
73 static TCGv_i64 cpu_bndl[4];
74 static TCGv_i64 cpu_bndu[4];
75 
76 #include "exec/gen-icount.h"
77 
78 typedef struct DisasContext {
79     DisasContextBase base;
80 
81     target_ulong pc;       /* pc = eip + cs_base */
82     target_ulong cs_base;  /* base of CS segment */
83     target_ulong pc_save;
84 
85     MemOp aflag;
86     MemOp dflag;
87 
88     int8_t override; /* -1 if no override, else R_CS, R_DS, etc */
89     uint8_t prefix;
90 
91     bool has_modrm;
92     uint8_t modrm;
93 
94 #ifndef CONFIG_USER_ONLY
95     uint8_t cpl;   /* code priv level */
96     uint8_t iopl;  /* i/o priv level */
97 #endif
98     uint8_t vex_l;  /* vex vector length */
99     uint8_t vex_v;  /* vex vvvv register, without 1's complement.  */
100     uint8_t popl_esp_hack; /* for correct popl with esp base handling */
101     uint8_t rip_offset; /* only used in x86_64, but left for simplicity */
102 
103 #ifdef TARGET_X86_64
104     uint8_t rex_r;
105     uint8_t rex_x;
106     uint8_t rex_b;
107 #endif
108     bool vex_w; /* used by AVX even on 32-bit processors */
109     bool jmp_opt; /* use direct block chaining for direct jumps */
110     bool repz_opt; /* optimize jumps within repz instructions */
111     bool cc_op_dirty;
112 
113     CCOp cc_op;  /* current CC operation */
114     int mem_index; /* select memory access functions */
115     uint32_t flags; /* all execution flags */
116     int cpuid_features;
117     int cpuid_ext_features;
118     int cpuid_ext2_features;
119     int cpuid_ext3_features;
120     int cpuid_7_0_ebx_features;
121     int cpuid_7_0_ecx_features;
122     int cpuid_xsave_features;
123 
124     /* TCG local temps */
125     TCGv cc_srcT;
126     TCGv A0;
127     TCGv T0;
128     TCGv T1;
129 
130     /* TCG local register indexes (only used inside old micro ops) */
131     TCGv tmp0;
132     TCGv tmp4;
133     TCGv_i32 tmp2_i32;
134     TCGv_i32 tmp3_i32;
135     TCGv_i64 tmp1_i64;
136 
137     sigjmp_buf jmpbuf;
138     TCGOp *prev_insn_end;
139 } DisasContext;
140 
141 #define DISAS_EOB_ONLY         DISAS_TARGET_0
142 #define DISAS_EOB_NEXT         DISAS_TARGET_1
143 #define DISAS_EOB_INHIBIT_IRQ  DISAS_TARGET_2
144 #define DISAS_JUMP             DISAS_TARGET_3
145 
146 /* The environment in which user-only runs is constrained. */
147 #ifdef CONFIG_USER_ONLY
148 #define PE(S)     true
149 #define CPL(S)    3
150 #define IOPL(S)   0
151 #define SVME(S)   false
152 #define GUEST(S)  false
153 #else
154 #define PE(S)     (((S)->flags & HF_PE_MASK) != 0)
155 #define CPL(S)    ((S)->cpl)
156 #define IOPL(S)   ((S)->iopl)
157 #define SVME(S)   (((S)->flags & HF_SVME_MASK) != 0)
158 #define GUEST(S)  (((S)->flags & HF_GUEST_MASK) != 0)
159 #endif
160 #if defined(CONFIG_USER_ONLY) && defined(TARGET_X86_64)
161 #define VM86(S)   false
162 #define CODE32(S) true
163 #define SS32(S)   true
164 #define ADDSEG(S) false
165 #else
166 #define VM86(S)   (((S)->flags & HF_VM_MASK) != 0)
167 #define CODE32(S) (((S)->flags & HF_CS32_MASK) != 0)
168 #define SS32(S)   (((S)->flags & HF_SS32_MASK) != 0)
169 #define ADDSEG(S) (((S)->flags & HF_ADDSEG_MASK) != 0)
170 #endif
171 #if !defined(TARGET_X86_64)
172 #define CODE64(S) false
173 #define LMA(S)    false
174 #elif defined(CONFIG_USER_ONLY)
175 #define CODE64(S) true
176 #define LMA(S)    true
177 #else
178 #define CODE64(S) (((S)->flags & HF_CS64_MASK) != 0)
179 #define LMA(S)    (((S)->flags & HF_LMA_MASK) != 0)
180 #endif
181 
182 #ifdef TARGET_X86_64
183 #define REX_PREFIX(S)  (((S)->prefix & PREFIX_REX) != 0)
184 #define REX_W(S)       ((S)->vex_w)
185 #define REX_R(S)       ((S)->rex_r + 0)
186 #define REX_X(S)       ((S)->rex_x + 0)
187 #define REX_B(S)       ((S)->rex_b + 0)
188 #else
189 #define REX_PREFIX(S)  false
190 #define REX_W(S)       false
191 #define REX_R(S)       0
192 #define REX_X(S)       0
193 #define REX_B(S)       0
194 #endif
195 
196 /*
197  * Many sysemu-only helpers are not reachable for user-only.
198  * Define stub generators here, so that we need not either sprinkle
199  * ifdefs through the translator, nor provide the helper function.
200  */
201 #define STUB_HELPER(NAME, ...) \
202     static inline void gen_helper_##NAME(__VA_ARGS__) \
203     { qemu_build_not_reached(); }
204 
205 #ifdef CONFIG_USER_ONLY
206 STUB_HELPER(clgi, TCGv_env env)
207 STUB_HELPER(flush_page, TCGv_env env, TCGv addr)
208 STUB_HELPER(hlt, TCGv_env env, TCGv_i32 pc_ofs)
209 STUB_HELPER(inb, TCGv ret, TCGv_env env, TCGv_i32 port)
210 STUB_HELPER(inw, TCGv ret, TCGv_env env, TCGv_i32 port)
211 STUB_HELPER(inl, TCGv ret, TCGv_env env, TCGv_i32 port)
212 STUB_HELPER(monitor, TCGv_env env, TCGv addr)
213 STUB_HELPER(mwait, TCGv_env env, TCGv_i32 pc_ofs)
214 STUB_HELPER(outb, TCGv_env env, TCGv_i32 port, TCGv_i32 val)
215 STUB_HELPER(outw, TCGv_env env, TCGv_i32 port, TCGv_i32 val)
216 STUB_HELPER(outl, TCGv_env env, TCGv_i32 port, TCGv_i32 val)
217 STUB_HELPER(rdmsr, TCGv_env env)
218 STUB_HELPER(read_crN, TCGv ret, TCGv_env env, TCGv_i32 reg)
219 STUB_HELPER(get_dr, TCGv ret, TCGv_env env, TCGv_i32 reg)
220 STUB_HELPER(set_dr, TCGv_env env, TCGv_i32 reg, TCGv val)
221 STUB_HELPER(stgi, TCGv_env env)
222 STUB_HELPER(svm_check_intercept, TCGv_env env, TCGv_i32 type)
223 STUB_HELPER(vmload, TCGv_env env, TCGv_i32 aflag)
224 STUB_HELPER(vmmcall, TCGv_env env)
225 STUB_HELPER(vmrun, TCGv_env env, TCGv_i32 aflag, TCGv_i32 pc_ofs)
226 STUB_HELPER(vmsave, TCGv_env env, TCGv_i32 aflag)
227 STUB_HELPER(write_crN, TCGv_env env, TCGv_i32 reg, TCGv val)
228 STUB_HELPER(wrmsr, TCGv_env env)
229 #endif
230 
231 static void gen_eob(DisasContext *s);
232 static void gen_jr(DisasContext *s);
233 static void gen_jmp_rel(DisasContext *s, MemOp ot, int diff, int tb_num);
234 static void gen_jmp_rel_csize(DisasContext *s, int diff, int tb_num);
235 static void gen_op(DisasContext *s1, int op, MemOp ot, int d);
236 static void gen_exception_gpf(DisasContext *s);
237 
238 /* i386 arith/logic operations */
239 enum {
240     OP_ADDL,
241     OP_ORL,
242     OP_ADCL,
243     OP_SBBL,
244     OP_ANDL,
245     OP_SUBL,
246     OP_XORL,
247     OP_CMPL,
248 };
249 
250 /* i386 shift ops */
251 enum {
252     OP_ROL,
253     OP_ROR,
254     OP_RCL,
255     OP_RCR,
256     OP_SHL,
257     OP_SHR,
258     OP_SHL1, /* undocumented */
259     OP_SAR = 7,
260 };
261 
262 enum {
263     JCC_O,
264     JCC_B,
265     JCC_Z,
266     JCC_BE,
267     JCC_S,
268     JCC_P,
269     JCC_L,
270     JCC_LE,
271 };
272 
273 enum {
274     /* I386 int registers */
275     OR_EAX,   /* MUST be even numbered */
276     OR_ECX,
277     OR_EDX,
278     OR_EBX,
279     OR_ESP,
280     OR_EBP,
281     OR_ESI,
282     OR_EDI,
283 
284     OR_TMP0 = 16,    /* temporary operand register */
285     OR_TMP1,
286     OR_A0, /* temporary register used when doing address evaluation */
287 };
288 
289 enum {
290     USES_CC_DST  = 1,
291     USES_CC_SRC  = 2,
292     USES_CC_SRC2 = 4,
293     USES_CC_SRCT = 8,
294 };
295 
296 /* Bit set if the global variable is live after setting CC_OP to X.  */
297 static const uint8_t cc_op_live[CC_OP_NB] = {
298     [CC_OP_DYNAMIC] = USES_CC_DST | USES_CC_SRC | USES_CC_SRC2,
299     [CC_OP_EFLAGS] = USES_CC_SRC,
300     [CC_OP_MULB ... CC_OP_MULQ] = USES_CC_DST | USES_CC_SRC,
301     [CC_OP_ADDB ... CC_OP_ADDQ] = USES_CC_DST | USES_CC_SRC,
302     [CC_OP_ADCB ... CC_OP_ADCQ] = USES_CC_DST | USES_CC_SRC | USES_CC_SRC2,
303     [CC_OP_SUBB ... CC_OP_SUBQ] = USES_CC_DST | USES_CC_SRC | USES_CC_SRCT,
304     [CC_OP_SBBB ... CC_OP_SBBQ] = USES_CC_DST | USES_CC_SRC | USES_CC_SRC2,
305     [CC_OP_LOGICB ... CC_OP_LOGICQ] = USES_CC_DST,
306     [CC_OP_INCB ... CC_OP_INCQ] = USES_CC_DST | USES_CC_SRC,
307     [CC_OP_DECB ... CC_OP_DECQ] = USES_CC_DST | USES_CC_SRC,
308     [CC_OP_SHLB ... CC_OP_SHLQ] = USES_CC_DST | USES_CC_SRC,
309     [CC_OP_SARB ... CC_OP_SARQ] = USES_CC_DST | USES_CC_SRC,
310     [CC_OP_BMILGB ... CC_OP_BMILGQ] = USES_CC_DST | USES_CC_SRC,
311     [CC_OP_ADCX] = USES_CC_DST | USES_CC_SRC,
312     [CC_OP_ADOX] = USES_CC_SRC | USES_CC_SRC2,
313     [CC_OP_ADCOX] = USES_CC_DST | USES_CC_SRC | USES_CC_SRC2,
314     [CC_OP_CLR] = 0,
315     [CC_OP_POPCNT] = USES_CC_SRC,
316 };
317 
318 static void set_cc_op(DisasContext *s, CCOp op)
319 {
320     int dead;
321 
322     if (s->cc_op == op) {
323         return;
324     }
325 
326     /* Discard CC computation that will no longer be used.  */
327     dead = cc_op_live[s->cc_op] & ~cc_op_live[op];
328     if (dead & USES_CC_DST) {
329         tcg_gen_discard_tl(cpu_cc_dst);
330     }
331     if (dead & USES_CC_SRC) {
332         tcg_gen_discard_tl(cpu_cc_src);
333     }
334     if (dead & USES_CC_SRC2) {
335         tcg_gen_discard_tl(cpu_cc_src2);
336     }
337     if (dead & USES_CC_SRCT) {
338         tcg_gen_discard_tl(s->cc_srcT);
339     }
340 
341     if (op == CC_OP_DYNAMIC) {
342         /* The DYNAMIC setting is translator only, and should never be
343            stored.  Thus we always consider it clean.  */
344         s->cc_op_dirty = false;
345     } else {
346         /* Discard any computed CC_OP value (see shifts).  */
347         if (s->cc_op == CC_OP_DYNAMIC) {
348             tcg_gen_discard_i32(cpu_cc_op);
349         }
350         s->cc_op_dirty = true;
351     }
352     s->cc_op = op;
353 }
354 
355 static void gen_update_cc_op(DisasContext *s)
356 {
357     if (s->cc_op_dirty) {
358         tcg_gen_movi_i32(cpu_cc_op, s->cc_op);
359         s->cc_op_dirty = false;
360     }
361 }
362 
363 #ifdef TARGET_X86_64
364 
365 #define NB_OP_SIZES 4
366 
367 #else /* !TARGET_X86_64 */
368 
369 #define NB_OP_SIZES 3
370 
371 #endif /* !TARGET_X86_64 */
372 
373 #if HOST_BIG_ENDIAN
374 #define REG_B_OFFSET (sizeof(target_ulong) - 1)
375 #define REG_H_OFFSET (sizeof(target_ulong) - 2)
376 #define REG_W_OFFSET (sizeof(target_ulong) - 2)
377 #define REG_L_OFFSET (sizeof(target_ulong) - 4)
378 #define REG_LH_OFFSET (sizeof(target_ulong) - 8)
379 #else
380 #define REG_B_OFFSET 0
381 #define REG_H_OFFSET 1
382 #define REG_W_OFFSET 0
383 #define REG_L_OFFSET 0
384 #define REG_LH_OFFSET 4
385 #endif
386 
387 /* In instruction encodings for byte register accesses the
388  * register number usually indicates "low 8 bits of register N";
389  * however there are some special cases where N 4..7 indicates
390  * [AH, CH, DH, BH], ie "bits 15..8 of register N-4". Return
391  * true for this special case, false otherwise.
392  */
393 static inline bool byte_reg_is_xH(DisasContext *s, int reg)
394 {
395     /* Any time the REX prefix is present, byte registers are uniform */
396     if (reg < 4 || REX_PREFIX(s)) {
397         return false;
398     }
399     return true;
400 }
401 
402 /* Select the size of a push/pop operation.  */
403 static inline MemOp mo_pushpop(DisasContext *s, MemOp ot)
404 {
405     if (CODE64(s)) {
406         return ot == MO_16 ? MO_16 : MO_64;
407     } else {
408         return ot;
409     }
410 }
411 
412 /* Select the size of the stack pointer.  */
413 static inline MemOp mo_stacksize(DisasContext *s)
414 {
415     return CODE64(s) ? MO_64 : SS32(s) ? MO_32 : MO_16;
416 }
417 
418 /* Select only size 64 else 32.  Used for SSE operand sizes.  */
419 static inline MemOp mo_64_32(MemOp ot)
420 {
421 #ifdef TARGET_X86_64
422     return ot == MO_64 ? MO_64 : MO_32;
423 #else
424     return MO_32;
425 #endif
426 }
427 
428 /* Select size 8 if lsb of B is clear, else OT.  Used for decoding
429    byte vs word opcodes.  */
430 static inline MemOp mo_b_d(int b, MemOp ot)
431 {
432     return b & 1 ? ot : MO_8;
433 }
434 
435 /* Select size 8 if lsb of B is clear, else OT capped at 32.
436    Used for decoding operand size of port opcodes.  */
437 static inline MemOp mo_b_d32(int b, MemOp ot)
438 {
439     return b & 1 ? (ot == MO_16 ? MO_16 : MO_32) : MO_8;
440 }
441 
442 /* Compute the result of writing t0 to the OT-sized register REG.
443  *
444  * If DEST is NULL, store the result into the register and return the
445  * register's TCGv.
446  *
447  * If DEST is not NULL, store the result into DEST and return the
448  * register's TCGv.
449  */
450 static TCGv gen_op_deposit_reg_v(DisasContext *s, MemOp ot, int reg, TCGv dest, TCGv t0)
451 {
452     switch(ot) {
453     case MO_8:
454         if (byte_reg_is_xH(s, reg)) {
455             dest = dest ? dest : cpu_regs[reg - 4];
456             tcg_gen_deposit_tl(dest, cpu_regs[reg - 4], t0, 8, 8);
457             return cpu_regs[reg - 4];
458         }
459         dest = dest ? dest : cpu_regs[reg];
460         tcg_gen_deposit_tl(dest, cpu_regs[reg], t0, 0, 8);
461         break;
462     case MO_16:
463         dest = dest ? dest : cpu_regs[reg];
464         tcg_gen_deposit_tl(dest, cpu_regs[reg], t0, 0, 16);
465         break;
466     case MO_32:
467         /* For x86_64, this sets the higher half of register to zero.
468            For i386, this is equivalent to a mov. */
469         dest = dest ? dest : cpu_regs[reg];
470         tcg_gen_ext32u_tl(dest, t0);
471         break;
472 #ifdef TARGET_X86_64
473     case MO_64:
474         dest = dest ? dest : cpu_regs[reg];
475         tcg_gen_mov_tl(dest, t0);
476         break;
477 #endif
478     default:
479         tcg_abort();
480     }
481     return cpu_regs[reg];
482 }
483 
484 static void gen_op_mov_reg_v(DisasContext *s, MemOp ot, int reg, TCGv t0)
485 {
486     gen_op_deposit_reg_v(s, ot, reg, NULL, t0);
487 }
488 
489 static inline
490 void gen_op_mov_v_reg(DisasContext *s, MemOp ot, TCGv t0, int reg)
491 {
492     if (ot == MO_8 && byte_reg_is_xH(s, reg)) {
493         tcg_gen_extract_tl(t0, cpu_regs[reg - 4], 8, 8);
494     } else {
495         tcg_gen_mov_tl(t0, cpu_regs[reg]);
496     }
497 }
498 
499 static void gen_add_A0_im(DisasContext *s, int val)
500 {
501     tcg_gen_addi_tl(s->A0, s->A0, val);
502     if (!CODE64(s)) {
503         tcg_gen_ext32u_tl(s->A0, s->A0);
504     }
505 }
506 
507 static inline void gen_op_jmp_v(DisasContext *s, TCGv dest)
508 {
509     tcg_gen_mov_tl(cpu_eip, dest);
510     s->pc_save = -1;
511 }
512 
513 static inline
514 void gen_op_add_reg_im(DisasContext *s, MemOp size, int reg, int32_t val)
515 {
516     tcg_gen_addi_tl(s->tmp0, cpu_regs[reg], val);
517     gen_op_mov_reg_v(s, size, reg, s->tmp0);
518 }
519 
520 static inline void gen_op_add_reg_T0(DisasContext *s, MemOp size, int reg)
521 {
522     tcg_gen_add_tl(s->tmp0, cpu_regs[reg], s->T0);
523     gen_op_mov_reg_v(s, size, reg, s->tmp0);
524 }
525 
526 static inline void gen_op_ld_v(DisasContext *s, int idx, TCGv t0, TCGv a0)
527 {
528     tcg_gen_qemu_ld_tl(t0, a0, s->mem_index, idx | MO_LE);
529 }
530 
531 static inline void gen_op_st_v(DisasContext *s, int idx, TCGv t0, TCGv a0)
532 {
533     tcg_gen_qemu_st_tl(t0, a0, s->mem_index, idx | MO_LE);
534 }
535 
536 static inline void gen_op_st_rm_T0_A0(DisasContext *s, int idx, int d)
537 {
538     if (d == OR_TMP0) {
539         gen_op_st_v(s, idx, s->T0, s->A0);
540     } else {
541         gen_op_mov_reg_v(s, idx, d, s->T0);
542     }
543 }
544 
545 static void gen_update_eip_cur(DisasContext *s)
546 {
547     assert(s->pc_save != -1);
548     if (TARGET_TB_PCREL) {
549         tcg_gen_addi_tl(cpu_eip, cpu_eip, s->base.pc_next - s->pc_save);
550     } else {
551         tcg_gen_movi_tl(cpu_eip, s->base.pc_next - s->cs_base);
552     }
553     s->pc_save = s->base.pc_next;
554 }
555 
556 static void gen_update_eip_next(DisasContext *s)
557 {
558     assert(s->pc_save != -1);
559     if (TARGET_TB_PCREL) {
560         tcg_gen_addi_tl(cpu_eip, cpu_eip, s->pc - s->pc_save);
561     } else {
562         tcg_gen_movi_tl(cpu_eip, s->pc - s->cs_base);
563     }
564     s->pc_save = s->pc;
565 }
566 
567 static int cur_insn_len(DisasContext *s)
568 {
569     return s->pc - s->base.pc_next;
570 }
571 
572 static TCGv_i32 cur_insn_len_i32(DisasContext *s)
573 {
574     return tcg_constant_i32(cur_insn_len(s));
575 }
576 
577 static TCGv_i32 eip_next_i32(DisasContext *s)
578 {
579     assert(s->pc_save != -1);
580     /*
581      * This function has two users: lcall_real (always 16-bit mode), and
582      * iret_protected (16, 32, or 64-bit mode).  IRET only uses the value
583      * when EFLAGS.NT is set, which is illegal in 64-bit mode, which is
584      * why passing a 32-bit value isn't broken.  To avoid using this where
585      * we shouldn't, return -1 in 64-bit mode so that execution goes into
586      * the weeds quickly.
587      */
588     if (CODE64(s)) {
589         return tcg_constant_i32(-1);
590     }
591     if (TARGET_TB_PCREL) {
592         TCGv_i32 ret = tcg_temp_new_i32();
593         tcg_gen_trunc_tl_i32(ret, cpu_eip);
594         tcg_gen_addi_i32(ret, ret, s->pc - s->pc_save);
595         return ret;
596     } else {
597         return tcg_constant_i32(s->pc - s->cs_base);
598     }
599 }
600 
601 static TCGv eip_next_tl(DisasContext *s)
602 {
603     assert(s->pc_save != -1);
604     if (TARGET_TB_PCREL) {
605         TCGv ret = tcg_temp_new();
606         tcg_gen_addi_tl(ret, cpu_eip, s->pc - s->pc_save);
607         return ret;
608     } else {
609         return tcg_constant_tl(s->pc - s->cs_base);
610     }
611 }
612 
613 static TCGv eip_cur_tl(DisasContext *s)
614 {
615     assert(s->pc_save != -1);
616     if (TARGET_TB_PCREL) {
617         TCGv ret = tcg_temp_new();
618         tcg_gen_addi_tl(ret, cpu_eip, s->base.pc_next - s->pc_save);
619         return ret;
620     } else {
621         return tcg_constant_tl(s->base.pc_next - s->cs_base);
622     }
623 }
624 
625 /* Compute SEG:REG into A0.  SEG is selected from the override segment
626    (OVR_SEG) and the default segment (DEF_SEG).  OVR_SEG may be -1 to
627    indicate no override.  */
628 static void gen_lea_v_seg(DisasContext *s, MemOp aflag, TCGv a0,
629                           int def_seg, int ovr_seg)
630 {
631     switch (aflag) {
632 #ifdef TARGET_X86_64
633     case MO_64:
634         if (ovr_seg < 0) {
635             tcg_gen_mov_tl(s->A0, a0);
636             return;
637         }
638         break;
639 #endif
640     case MO_32:
641         /* 32 bit address */
642         if (ovr_seg < 0 && ADDSEG(s)) {
643             ovr_seg = def_seg;
644         }
645         if (ovr_seg < 0) {
646             tcg_gen_ext32u_tl(s->A0, a0);
647             return;
648         }
649         break;
650     case MO_16:
651         /* 16 bit address */
652         tcg_gen_ext16u_tl(s->A0, a0);
653         a0 = s->A0;
654         if (ovr_seg < 0) {
655             if (ADDSEG(s)) {
656                 ovr_seg = def_seg;
657             } else {
658                 return;
659             }
660         }
661         break;
662     default:
663         tcg_abort();
664     }
665 
666     if (ovr_seg >= 0) {
667         TCGv seg = cpu_seg_base[ovr_seg];
668 
669         if (aflag == MO_64) {
670             tcg_gen_add_tl(s->A0, a0, seg);
671         } else if (CODE64(s)) {
672             tcg_gen_ext32u_tl(s->A0, a0);
673             tcg_gen_add_tl(s->A0, s->A0, seg);
674         } else {
675             tcg_gen_add_tl(s->A0, a0, seg);
676             tcg_gen_ext32u_tl(s->A0, s->A0);
677         }
678     }
679 }
680 
681 static inline void gen_string_movl_A0_ESI(DisasContext *s)
682 {
683     gen_lea_v_seg(s, s->aflag, cpu_regs[R_ESI], R_DS, s->override);
684 }
685 
686 static inline void gen_string_movl_A0_EDI(DisasContext *s)
687 {
688     gen_lea_v_seg(s, s->aflag, cpu_regs[R_EDI], R_ES, -1);
689 }
690 
691 static inline void gen_op_movl_T0_Dshift(DisasContext *s, MemOp ot)
692 {
693     tcg_gen_ld32s_tl(s->T0, cpu_env, offsetof(CPUX86State, df));
694     tcg_gen_shli_tl(s->T0, s->T0, ot);
695 };
696 
697 static TCGv gen_ext_tl(TCGv dst, TCGv src, MemOp size, bool sign)
698 {
699     switch (size) {
700     case MO_8:
701         if (sign) {
702             tcg_gen_ext8s_tl(dst, src);
703         } else {
704             tcg_gen_ext8u_tl(dst, src);
705         }
706         return dst;
707     case MO_16:
708         if (sign) {
709             tcg_gen_ext16s_tl(dst, src);
710         } else {
711             tcg_gen_ext16u_tl(dst, src);
712         }
713         return dst;
714 #ifdef TARGET_X86_64
715     case MO_32:
716         if (sign) {
717             tcg_gen_ext32s_tl(dst, src);
718         } else {
719             tcg_gen_ext32u_tl(dst, src);
720         }
721         return dst;
722 #endif
723     default:
724         return src;
725     }
726 }
727 
728 static void gen_extu(MemOp ot, TCGv reg)
729 {
730     gen_ext_tl(reg, reg, ot, false);
731 }
732 
733 static void gen_exts(MemOp ot, TCGv reg)
734 {
735     gen_ext_tl(reg, reg, ot, true);
736 }
737 
738 static void gen_op_j_ecx(DisasContext *s, TCGCond cond, TCGLabel *label1)
739 {
740     tcg_gen_mov_tl(s->tmp0, cpu_regs[R_ECX]);
741     gen_extu(s->aflag, s->tmp0);
742     tcg_gen_brcondi_tl(cond, s->tmp0, 0, label1);
743 }
744 
745 static inline void gen_op_jz_ecx(DisasContext *s, TCGLabel *label1)
746 {
747     gen_op_j_ecx(s, TCG_COND_EQ, label1);
748 }
749 
750 static inline void gen_op_jnz_ecx(DisasContext *s, TCGLabel *label1)
751 {
752     gen_op_j_ecx(s, TCG_COND_NE, label1);
753 }
754 
755 static void gen_helper_in_func(MemOp ot, TCGv v, TCGv_i32 n)
756 {
757     switch (ot) {
758     case MO_8:
759         gen_helper_inb(v, cpu_env, n);
760         break;
761     case MO_16:
762         gen_helper_inw(v, cpu_env, n);
763         break;
764     case MO_32:
765         gen_helper_inl(v, cpu_env, n);
766         break;
767     default:
768         tcg_abort();
769     }
770 }
771 
772 static void gen_helper_out_func(MemOp ot, TCGv_i32 v, TCGv_i32 n)
773 {
774     switch (ot) {
775     case MO_8:
776         gen_helper_outb(cpu_env, v, n);
777         break;
778     case MO_16:
779         gen_helper_outw(cpu_env, v, n);
780         break;
781     case MO_32:
782         gen_helper_outl(cpu_env, v, n);
783         break;
784     default:
785         tcg_abort();
786     }
787 }
788 
789 /*
790  * Validate that access to [port, port + 1<<ot) is allowed.
791  * Raise #GP, or VMM exit if not.
792  */
793 static bool gen_check_io(DisasContext *s, MemOp ot, TCGv_i32 port,
794                          uint32_t svm_flags)
795 {
796 #ifdef CONFIG_USER_ONLY
797     /*
798      * We do not implement the ioperm(2) syscall, so the TSS check
799      * will always fail.
800      */
801     gen_exception_gpf(s);
802     return false;
803 #else
804     if (PE(s) && (CPL(s) > IOPL(s) || VM86(s))) {
805         gen_helper_check_io(cpu_env, port, tcg_constant_i32(1 << ot));
806     }
807     if (GUEST(s)) {
808         gen_update_cc_op(s);
809         gen_update_eip_cur(s);
810         if (s->prefix & (PREFIX_REPZ | PREFIX_REPNZ)) {
811             svm_flags |= SVM_IOIO_REP_MASK;
812         }
813         svm_flags |= 1 << (SVM_IOIO_SIZE_SHIFT + ot);
814         gen_helper_svm_check_io(cpu_env, port,
815                                 tcg_constant_i32(svm_flags),
816                                 cur_insn_len_i32(s));
817     }
818     return true;
819 #endif
820 }
821 
822 static void gen_movs(DisasContext *s, MemOp ot)
823 {
824     gen_string_movl_A0_ESI(s);
825     gen_op_ld_v(s, ot, s->T0, s->A0);
826     gen_string_movl_A0_EDI(s);
827     gen_op_st_v(s, ot, s->T0, s->A0);
828     gen_op_movl_T0_Dshift(s, ot);
829     gen_op_add_reg_T0(s, s->aflag, R_ESI);
830     gen_op_add_reg_T0(s, s->aflag, R_EDI);
831 }
832 
833 static void gen_op_update1_cc(DisasContext *s)
834 {
835     tcg_gen_mov_tl(cpu_cc_dst, s->T0);
836 }
837 
838 static void gen_op_update2_cc(DisasContext *s)
839 {
840     tcg_gen_mov_tl(cpu_cc_src, s->T1);
841     tcg_gen_mov_tl(cpu_cc_dst, s->T0);
842 }
843 
844 static void gen_op_update3_cc(DisasContext *s, TCGv reg)
845 {
846     tcg_gen_mov_tl(cpu_cc_src2, reg);
847     tcg_gen_mov_tl(cpu_cc_src, s->T1);
848     tcg_gen_mov_tl(cpu_cc_dst, s->T0);
849 }
850 
851 static inline void gen_op_testl_T0_T1_cc(DisasContext *s)
852 {
853     tcg_gen_and_tl(cpu_cc_dst, s->T0, s->T1);
854 }
855 
856 static void gen_op_update_neg_cc(DisasContext *s)
857 {
858     tcg_gen_mov_tl(cpu_cc_dst, s->T0);
859     tcg_gen_neg_tl(cpu_cc_src, s->T0);
860     tcg_gen_movi_tl(s->cc_srcT, 0);
861 }
862 
863 /* compute all eflags to cc_src */
864 static void gen_compute_eflags(DisasContext *s)
865 {
866     TCGv zero, dst, src1, src2;
867     int live, dead;
868 
869     if (s->cc_op == CC_OP_EFLAGS) {
870         return;
871     }
872     if (s->cc_op == CC_OP_CLR) {
873         tcg_gen_movi_tl(cpu_cc_src, CC_Z | CC_P);
874         set_cc_op(s, CC_OP_EFLAGS);
875         return;
876     }
877 
878     zero = NULL;
879     dst = cpu_cc_dst;
880     src1 = cpu_cc_src;
881     src2 = cpu_cc_src2;
882 
883     /* Take care to not read values that are not live.  */
884     live = cc_op_live[s->cc_op] & ~USES_CC_SRCT;
885     dead = live ^ (USES_CC_DST | USES_CC_SRC | USES_CC_SRC2);
886     if (dead) {
887         zero = tcg_const_tl(0);
888         if (dead & USES_CC_DST) {
889             dst = zero;
890         }
891         if (dead & USES_CC_SRC) {
892             src1 = zero;
893         }
894         if (dead & USES_CC_SRC2) {
895             src2 = zero;
896         }
897     }
898 
899     gen_update_cc_op(s);
900     gen_helper_cc_compute_all(cpu_cc_src, dst, src1, src2, cpu_cc_op);
901     set_cc_op(s, CC_OP_EFLAGS);
902 
903     if (dead) {
904         tcg_temp_free(zero);
905     }
906 }
907 
908 typedef struct CCPrepare {
909     TCGCond cond;
910     TCGv reg;
911     TCGv reg2;
912     target_ulong imm;
913     target_ulong mask;
914     bool use_reg2;
915     bool no_setcond;
916 } CCPrepare;
917 
918 /* compute eflags.C to reg */
919 static CCPrepare gen_prepare_eflags_c(DisasContext *s, TCGv reg)
920 {
921     TCGv t0, t1;
922     int size, shift;
923 
924     switch (s->cc_op) {
925     case CC_OP_SUBB ... CC_OP_SUBQ:
926         /* (DATA_TYPE)CC_SRCT < (DATA_TYPE)CC_SRC */
927         size = s->cc_op - CC_OP_SUBB;
928         t1 = gen_ext_tl(s->tmp0, cpu_cc_src, size, false);
929         /* If no temporary was used, be careful not to alias t1 and t0.  */
930         t0 = t1 == cpu_cc_src ? s->tmp0 : reg;
931         tcg_gen_mov_tl(t0, s->cc_srcT);
932         gen_extu(size, t0);
933         goto add_sub;
934 
935     case CC_OP_ADDB ... CC_OP_ADDQ:
936         /* (DATA_TYPE)CC_DST < (DATA_TYPE)CC_SRC */
937         size = s->cc_op - CC_OP_ADDB;
938         t1 = gen_ext_tl(s->tmp0, cpu_cc_src, size, false);
939         t0 = gen_ext_tl(reg, cpu_cc_dst, size, false);
940     add_sub:
941         return (CCPrepare) { .cond = TCG_COND_LTU, .reg = t0,
942                              .reg2 = t1, .mask = -1, .use_reg2 = true };
943 
944     case CC_OP_LOGICB ... CC_OP_LOGICQ:
945     case CC_OP_CLR:
946     case CC_OP_POPCNT:
947         return (CCPrepare) { .cond = TCG_COND_NEVER, .mask = -1 };
948 
949     case CC_OP_INCB ... CC_OP_INCQ:
950     case CC_OP_DECB ... CC_OP_DECQ:
951         return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src,
952                              .mask = -1, .no_setcond = true };
953 
954     case CC_OP_SHLB ... CC_OP_SHLQ:
955         /* (CC_SRC >> (DATA_BITS - 1)) & 1 */
956         size = s->cc_op - CC_OP_SHLB;
957         shift = (8 << size) - 1;
958         return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src,
959                              .mask = (target_ulong)1 << shift };
960 
961     case CC_OP_MULB ... CC_OP_MULQ:
962         return (CCPrepare) { .cond = TCG_COND_NE,
963                              .reg = cpu_cc_src, .mask = -1 };
964 
965     case CC_OP_BMILGB ... CC_OP_BMILGQ:
966         size = s->cc_op - CC_OP_BMILGB;
967         t0 = gen_ext_tl(reg, cpu_cc_src, size, false);
968         return (CCPrepare) { .cond = TCG_COND_EQ, .reg = t0, .mask = -1 };
969 
970     case CC_OP_ADCX:
971     case CC_OP_ADCOX:
972         return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_dst,
973                              .mask = -1, .no_setcond = true };
974 
975     case CC_OP_EFLAGS:
976     case CC_OP_SARB ... CC_OP_SARQ:
977         /* CC_SRC & 1 */
978         return (CCPrepare) { .cond = TCG_COND_NE,
979                              .reg = cpu_cc_src, .mask = CC_C };
980 
981     default:
982        /* The need to compute only C from CC_OP_DYNAMIC is important
983           in efficiently implementing e.g. INC at the start of a TB.  */
984        gen_update_cc_op(s);
985        gen_helper_cc_compute_c(reg, cpu_cc_dst, cpu_cc_src,
986                                cpu_cc_src2, cpu_cc_op);
987        return (CCPrepare) { .cond = TCG_COND_NE, .reg = reg,
988                             .mask = -1, .no_setcond = true };
989     }
990 }
991 
992 /* compute eflags.P to reg */
993 static CCPrepare gen_prepare_eflags_p(DisasContext *s, TCGv reg)
994 {
995     gen_compute_eflags(s);
996     return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src,
997                          .mask = CC_P };
998 }
999 
1000 /* compute eflags.S to reg */
1001 static CCPrepare gen_prepare_eflags_s(DisasContext *s, TCGv reg)
1002 {
1003     switch (s->cc_op) {
1004     case CC_OP_DYNAMIC:
1005         gen_compute_eflags(s);
1006         /* FALLTHRU */
1007     case CC_OP_EFLAGS:
1008     case CC_OP_ADCX:
1009     case CC_OP_ADOX:
1010     case CC_OP_ADCOX:
1011         return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src,
1012                              .mask = CC_S };
1013     case CC_OP_CLR:
1014     case CC_OP_POPCNT:
1015         return (CCPrepare) { .cond = TCG_COND_NEVER, .mask = -1 };
1016     default:
1017         {
1018             MemOp size = (s->cc_op - CC_OP_ADDB) & 3;
1019             TCGv t0 = gen_ext_tl(reg, cpu_cc_dst, size, true);
1020             return (CCPrepare) { .cond = TCG_COND_LT, .reg = t0, .mask = -1 };
1021         }
1022     }
1023 }
1024 
1025 /* compute eflags.O to reg */
1026 static CCPrepare gen_prepare_eflags_o(DisasContext *s, TCGv reg)
1027 {
1028     switch (s->cc_op) {
1029     case CC_OP_ADOX:
1030     case CC_OP_ADCOX:
1031         return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src2,
1032                              .mask = -1, .no_setcond = true };
1033     case CC_OP_CLR:
1034     case CC_OP_POPCNT:
1035         return (CCPrepare) { .cond = TCG_COND_NEVER, .mask = -1 };
1036     default:
1037         gen_compute_eflags(s);
1038         return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src,
1039                              .mask = CC_O };
1040     }
1041 }
1042 
1043 /* compute eflags.Z to reg */
1044 static CCPrepare gen_prepare_eflags_z(DisasContext *s, TCGv reg)
1045 {
1046     switch (s->cc_op) {
1047     case CC_OP_DYNAMIC:
1048         gen_compute_eflags(s);
1049         /* FALLTHRU */
1050     case CC_OP_EFLAGS:
1051     case CC_OP_ADCX:
1052     case CC_OP_ADOX:
1053     case CC_OP_ADCOX:
1054         return (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src,
1055                              .mask = CC_Z };
1056     case CC_OP_CLR:
1057         return (CCPrepare) { .cond = TCG_COND_ALWAYS, .mask = -1 };
1058     case CC_OP_POPCNT:
1059         return (CCPrepare) { .cond = TCG_COND_EQ, .reg = cpu_cc_src,
1060                              .mask = -1 };
1061     default:
1062         {
1063             MemOp size = (s->cc_op - CC_OP_ADDB) & 3;
1064             TCGv t0 = gen_ext_tl(reg, cpu_cc_dst, size, false);
1065             return (CCPrepare) { .cond = TCG_COND_EQ, .reg = t0, .mask = -1 };
1066         }
1067     }
1068 }
1069 
1070 /* perform a conditional store into register 'reg' according to jump opcode
1071    value 'b'. In the fast case, T0 is guaranted not to be used. */
1072 static CCPrepare gen_prepare_cc(DisasContext *s, int b, TCGv reg)
1073 {
1074     int inv, jcc_op, cond;
1075     MemOp size;
1076     CCPrepare cc;
1077     TCGv t0;
1078 
1079     inv = b & 1;
1080     jcc_op = (b >> 1) & 7;
1081 
1082     switch (s->cc_op) {
1083     case CC_OP_SUBB ... CC_OP_SUBQ:
1084         /* We optimize relational operators for the cmp/jcc case.  */
1085         size = s->cc_op - CC_OP_SUBB;
1086         switch (jcc_op) {
1087         case JCC_BE:
1088             tcg_gen_mov_tl(s->tmp4, s->cc_srcT);
1089             gen_extu(size, s->tmp4);
1090             t0 = gen_ext_tl(s->tmp0, cpu_cc_src, size, false);
1091             cc = (CCPrepare) { .cond = TCG_COND_LEU, .reg = s->tmp4,
1092                                .reg2 = t0, .mask = -1, .use_reg2 = true };
1093             break;
1094 
1095         case JCC_L:
1096             cond = TCG_COND_LT;
1097             goto fast_jcc_l;
1098         case JCC_LE:
1099             cond = TCG_COND_LE;
1100         fast_jcc_l:
1101             tcg_gen_mov_tl(s->tmp4, s->cc_srcT);
1102             gen_exts(size, s->tmp4);
1103             t0 = gen_ext_tl(s->tmp0, cpu_cc_src, size, true);
1104             cc = (CCPrepare) { .cond = cond, .reg = s->tmp4,
1105                                .reg2 = t0, .mask = -1, .use_reg2 = true };
1106             break;
1107 
1108         default:
1109             goto slow_jcc;
1110         }
1111         break;
1112 
1113     default:
1114     slow_jcc:
1115         /* This actually generates good code for JC, JZ and JS.  */
1116         switch (jcc_op) {
1117         case JCC_O:
1118             cc = gen_prepare_eflags_o(s, reg);
1119             break;
1120         case JCC_B:
1121             cc = gen_prepare_eflags_c(s, reg);
1122             break;
1123         case JCC_Z:
1124             cc = gen_prepare_eflags_z(s, reg);
1125             break;
1126         case JCC_BE:
1127             gen_compute_eflags(s);
1128             cc = (CCPrepare) { .cond = TCG_COND_NE, .reg = cpu_cc_src,
1129                                .mask = CC_Z | CC_C };
1130             break;
1131         case JCC_S:
1132             cc = gen_prepare_eflags_s(s, reg);
1133             break;
1134         case JCC_P:
1135             cc = gen_prepare_eflags_p(s, reg);
1136             break;
1137         case JCC_L:
1138             gen_compute_eflags(s);
1139             if (reg == cpu_cc_src) {
1140                 reg = s->tmp0;
1141             }
1142             tcg_gen_shri_tl(reg, cpu_cc_src, 4); /* CC_O -> CC_S */
1143             tcg_gen_xor_tl(reg, reg, cpu_cc_src);
1144             cc = (CCPrepare) { .cond = TCG_COND_NE, .reg = reg,
1145                                .mask = CC_S };
1146             break;
1147         default:
1148         case JCC_LE:
1149             gen_compute_eflags(s);
1150             if (reg == cpu_cc_src) {
1151                 reg = s->tmp0;
1152             }
1153             tcg_gen_shri_tl(reg, cpu_cc_src, 4); /* CC_O -> CC_S */
1154             tcg_gen_xor_tl(reg, reg, cpu_cc_src);
1155             cc = (CCPrepare) { .cond = TCG_COND_NE, .reg = reg,
1156                                .mask = CC_S | CC_Z };
1157             break;
1158         }
1159         break;
1160     }
1161 
1162     if (inv) {
1163         cc.cond = tcg_invert_cond(cc.cond);
1164     }
1165     return cc;
1166 }
1167 
1168 static void gen_setcc1(DisasContext *s, int b, TCGv reg)
1169 {
1170     CCPrepare cc = gen_prepare_cc(s, b, reg);
1171 
1172     if (cc.no_setcond) {
1173         if (cc.cond == TCG_COND_EQ) {
1174             tcg_gen_xori_tl(reg, cc.reg, 1);
1175         } else {
1176             tcg_gen_mov_tl(reg, cc.reg);
1177         }
1178         return;
1179     }
1180 
1181     if (cc.cond == TCG_COND_NE && !cc.use_reg2 && cc.imm == 0 &&
1182         cc.mask != 0 && (cc.mask & (cc.mask - 1)) == 0) {
1183         tcg_gen_shri_tl(reg, cc.reg, ctztl(cc.mask));
1184         tcg_gen_andi_tl(reg, reg, 1);
1185         return;
1186     }
1187     if (cc.mask != -1) {
1188         tcg_gen_andi_tl(reg, cc.reg, cc.mask);
1189         cc.reg = reg;
1190     }
1191     if (cc.use_reg2) {
1192         tcg_gen_setcond_tl(cc.cond, reg, cc.reg, cc.reg2);
1193     } else {
1194         tcg_gen_setcondi_tl(cc.cond, reg, cc.reg, cc.imm);
1195     }
1196 }
1197 
1198 static inline void gen_compute_eflags_c(DisasContext *s, TCGv reg)
1199 {
1200     gen_setcc1(s, JCC_B << 1, reg);
1201 }
1202 
1203 /* generate a conditional jump to label 'l1' according to jump opcode
1204    value 'b'. In the fast case, T0 is guaranted not to be used. */
1205 static inline void gen_jcc1_noeob(DisasContext *s, int b, TCGLabel *l1)
1206 {
1207     CCPrepare cc = gen_prepare_cc(s, b, s->T0);
1208 
1209     if (cc.mask != -1) {
1210         tcg_gen_andi_tl(s->T0, cc.reg, cc.mask);
1211         cc.reg = s->T0;
1212     }
1213     if (cc.use_reg2) {
1214         tcg_gen_brcond_tl(cc.cond, cc.reg, cc.reg2, l1);
1215     } else {
1216         tcg_gen_brcondi_tl(cc.cond, cc.reg, cc.imm, l1);
1217     }
1218 }
1219 
1220 /* Generate a conditional jump to label 'l1' according to jump opcode
1221    value 'b'. In the fast case, T0 is guaranted not to be used.
1222    A translation block must end soon.  */
1223 static inline void gen_jcc1(DisasContext *s, int b, TCGLabel *l1)
1224 {
1225     CCPrepare cc = gen_prepare_cc(s, b, s->T0);
1226 
1227     gen_update_cc_op(s);
1228     if (cc.mask != -1) {
1229         tcg_gen_andi_tl(s->T0, cc.reg, cc.mask);
1230         cc.reg = s->T0;
1231     }
1232     set_cc_op(s, CC_OP_DYNAMIC);
1233     if (cc.use_reg2) {
1234         tcg_gen_brcond_tl(cc.cond, cc.reg, cc.reg2, l1);
1235     } else {
1236         tcg_gen_brcondi_tl(cc.cond, cc.reg, cc.imm, l1);
1237     }
1238 }
1239 
1240 /* XXX: does not work with gdbstub "ice" single step - not a
1241    serious problem */
1242 static TCGLabel *gen_jz_ecx_string(DisasContext *s)
1243 {
1244     TCGLabel *l1 = gen_new_label();
1245     TCGLabel *l2 = gen_new_label();
1246     gen_op_jnz_ecx(s, l1);
1247     gen_set_label(l2);
1248     gen_jmp_rel_csize(s, 0, 1);
1249     gen_set_label(l1);
1250     return l2;
1251 }
1252 
1253 static void gen_stos(DisasContext *s, MemOp ot)
1254 {
1255     gen_op_mov_v_reg(s, MO_32, s->T0, R_EAX);
1256     gen_string_movl_A0_EDI(s);
1257     gen_op_st_v(s, ot, s->T0, s->A0);
1258     gen_op_movl_T0_Dshift(s, ot);
1259     gen_op_add_reg_T0(s, s->aflag, R_EDI);
1260 }
1261 
1262 static void gen_lods(DisasContext *s, MemOp ot)
1263 {
1264     gen_string_movl_A0_ESI(s);
1265     gen_op_ld_v(s, ot, s->T0, s->A0);
1266     gen_op_mov_reg_v(s, ot, R_EAX, s->T0);
1267     gen_op_movl_T0_Dshift(s, ot);
1268     gen_op_add_reg_T0(s, s->aflag, R_ESI);
1269 }
1270 
1271 static void gen_scas(DisasContext *s, MemOp ot)
1272 {
1273     gen_string_movl_A0_EDI(s);
1274     gen_op_ld_v(s, ot, s->T1, s->A0);
1275     gen_op(s, OP_CMPL, ot, R_EAX);
1276     gen_op_movl_T0_Dshift(s, ot);
1277     gen_op_add_reg_T0(s, s->aflag, R_EDI);
1278 }
1279 
1280 static void gen_cmps(DisasContext *s, MemOp ot)
1281 {
1282     gen_string_movl_A0_EDI(s);
1283     gen_op_ld_v(s, ot, s->T1, s->A0);
1284     gen_string_movl_A0_ESI(s);
1285     gen_op(s, OP_CMPL, ot, OR_TMP0);
1286     gen_op_movl_T0_Dshift(s, ot);
1287     gen_op_add_reg_T0(s, s->aflag, R_ESI);
1288     gen_op_add_reg_T0(s, s->aflag, R_EDI);
1289 }
1290 
1291 static void gen_bpt_io(DisasContext *s, TCGv_i32 t_port, int ot)
1292 {
1293     if (s->flags & HF_IOBPT_MASK) {
1294 #ifdef CONFIG_USER_ONLY
1295         /* user-mode cpu should not be in IOBPT mode */
1296         g_assert_not_reached();
1297 #else
1298         TCGv_i32 t_size = tcg_constant_i32(1 << ot);
1299         TCGv t_next = eip_next_tl(s);
1300         gen_helper_bpt_io(cpu_env, t_port, t_size, t_next);
1301 #endif /* CONFIG_USER_ONLY */
1302     }
1303 }
1304 
1305 static void gen_ins(DisasContext *s, MemOp ot)
1306 {
1307     gen_string_movl_A0_EDI(s);
1308     /* Note: we must do this dummy write first to be restartable in
1309        case of page fault. */
1310     tcg_gen_movi_tl(s->T0, 0);
1311     gen_op_st_v(s, ot, s->T0, s->A0);
1312     tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_EDX]);
1313     tcg_gen_andi_i32(s->tmp2_i32, s->tmp2_i32, 0xffff);
1314     gen_helper_in_func(ot, s->T0, s->tmp2_i32);
1315     gen_op_st_v(s, ot, s->T0, s->A0);
1316     gen_op_movl_T0_Dshift(s, ot);
1317     gen_op_add_reg_T0(s, s->aflag, R_EDI);
1318     gen_bpt_io(s, s->tmp2_i32, ot);
1319 }
1320 
1321 static void gen_outs(DisasContext *s, MemOp ot)
1322 {
1323     gen_string_movl_A0_ESI(s);
1324     gen_op_ld_v(s, ot, s->T0, s->A0);
1325 
1326     tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_EDX]);
1327     tcg_gen_andi_i32(s->tmp2_i32, s->tmp2_i32, 0xffff);
1328     tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T0);
1329     gen_helper_out_func(ot, s->tmp2_i32, s->tmp3_i32);
1330     gen_op_movl_T0_Dshift(s, ot);
1331     gen_op_add_reg_T0(s, s->aflag, R_ESI);
1332     gen_bpt_io(s, s->tmp2_i32, ot);
1333 }
1334 
1335 /* Generate jumps to current or next instruction */
1336 static void gen_repz(DisasContext *s, MemOp ot,
1337                      void (*fn)(DisasContext *s, MemOp ot))
1338 {
1339     TCGLabel *l2;
1340     gen_update_cc_op(s);
1341     l2 = gen_jz_ecx_string(s);
1342     fn(s, ot);
1343     gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
1344     /*
1345      * A loop would cause two single step exceptions if ECX = 1
1346      * before rep string_insn
1347      */
1348     if (s->repz_opt) {
1349         gen_op_jz_ecx(s, l2);
1350     }
1351     gen_jmp_rel_csize(s, -cur_insn_len(s), 0);
1352 }
1353 
1354 #define GEN_REPZ(op) \
1355     static inline void gen_repz_ ## op(DisasContext *s, MemOp ot) \
1356     { gen_repz(s, ot, gen_##op); }
1357 
1358 static void gen_repz2(DisasContext *s, MemOp ot, int nz,
1359                       void (*fn)(DisasContext *s, MemOp ot))
1360 {
1361     TCGLabel *l2;
1362     gen_update_cc_op(s);
1363     l2 = gen_jz_ecx_string(s);
1364     fn(s, ot);
1365     gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
1366     gen_update_cc_op(s);
1367     gen_jcc1(s, (JCC_Z << 1) | (nz ^ 1), l2);
1368     if (s->repz_opt) {
1369         gen_op_jz_ecx(s, l2);
1370     }
1371     gen_jmp_rel_csize(s, -cur_insn_len(s), 0);
1372 }
1373 
1374 #define GEN_REPZ2(op) \
1375     static inline void gen_repz_ ## op(DisasContext *s, MemOp ot, int nz) \
1376     { gen_repz2(s, ot, nz, gen_##op); }
1377 
1378 GEN_REPZ(movs)
1379 GEN_REPZ(stos)
1380 GEN_REPZ(lods)
1381 GEN_REPZ(ins)
1382 GEN_REPZ(outs)
1383 GEN_REPZ2(scas)
1384 GEN_REPZ2(cmps)
1385 
1386 static void gen_helper_fp_arith_ST0_FT0(int op)
1387 {
1388     switch (op) {
1389     case 0:
1390         gen_helper_fadd_ST0_FT0(cpu_env);
1391         break;
1392     case 1:
1393         gen_helper_fmul_ST0_FT0(cpu_env);
1394         break;
1395     case 2:
1396         gen_helper_fcom_ST0_FT0(cpu_env);
1397         break;
1398     case 3:
1399         gen_helper_fcom_ST0_FT0(cpu_env);
1400         break;
1401     case 4:
1402         gen_helper_fsub_ST0_FT0(cpu_env);
1403         break;
1404     case 5:
1405         gen_helper_fsubr_ST0_FT0(cpu_env);
1406         break;
1407     case 6:
1408         gen_helper_fdiv_ST0_FT0(cpu_env);
1409         break;
1410     case 7:
1411         gen_helper_fdivr_ST0_FT0(cpu_env);
1412         break;
1413     }
1414 }
1415 
1416 /* NOTE the exception in "r" op ordering */
1417 static void gen_helper_fp_arith_STN_ST0(int op, int opreg)
1418 {
1419     TCGv_i32 tmp = tcg_const_i32(opreg);
1420     switch (op) {
1421     case 0:
1422         gen_helper_fadd_STN_ST0(cpu_env, tmp);
1423         break;
1424     case 1:
1425         gen_helper_fmul_STN_ST0(cpu_env, tmp);
1426         break;
1427     case 4:
1428         gen_helper_fsubr_STN_ST0(cpu_env, tmp);
1429         break;
1430     case 5:
1431         gen_helper_fsub_STN_ST0(cpu_env, tmp);
1432         break;
1433     case 6:
1434         gen_helper_fdivr_STN_ST0(cpu_env, tmp);
1435         break;
1436     case 7:
1437         gen_helper_fdiv_STN_ST0(cpu_env, tmp);
1438         break;
1439     }
1440 }
1441 
1442 static void gen_exception(DisasContext *s, int trapno)
1443 {
1444     gen_update_cc_op(s);
1445     gen_update_eip_cur(s);
1446     gen_helper_raise_exception(cpu_env, tcg_const_i32(trapno));
1447     s->base.is_jmp = DISAS_NORETURN;
1448 }
1449 
1450 /* Generate #UD for the current instruction.  The assumption here is that
1451    the instruction is known, but it isn't allowed in the current cpu mode.  */
1452 static void gen_illegal_opcode(DisasContext *s)
1453 {
1454     gen_exception(s, EXCP06_ILLOP);
1455 }
1456 
1457 /* Generate #GP for the current instruction. */
1458 static void gen_exception_gpf(DisasContext *s)
1459 {
1460     gen_exception(s, EXCP0D_GPF);
1461 }
1462 
1463 /* Check for cpl == 0; if not, raise #GP and return false. */
1464 static bool check_cpl0(DisasContext *s)
1465 {
1466     if (CPL(s) == 0) {
1467         return true;
1468     }
1469     gen_exception_gpf(s);
1470     return false;
1471 }
1472 
1473 /* If vm86, check for iopl == 3; if not, raise #GP and return false. */
1474 static bool check_vm86_iopl(DisasContext *s)
1475 {
1476     if (!VM86(s) || IOPL(s) == 3) {
1477         return true;
1478     }
1479     gen_exception_gpf(s);
1480     return false;
1481 }
1482 
1483 /* Check for iopl allowing access; if not, raise #GP and return false. */
1484 static bool check_iopl(DisasContext *s)
1485 {
1486     if (VM86(s) ? IOPL(s) == 3 : CPL(s) <= IOPL(s)) {
1487         return true;
1488     }
1489     gen_exception_gpf(s);
1490     return false;
1491 }
1492 
1493 /* if d == OR_TMP0, it means memory operand (address in A0) */
1494 static void gen_op(DisasContext *s1, int op, MemOp ot, int d)
1495 {
1496     if (d != OR_TMP0) {
1497         if (s1->prefix & PREFIX_LOCK) {
1498             /* Lock prefix when destination is not memory.  */
1499             gen_illegal_opcode(s1);
1500             return;
1501         }
1502         gen_op_mov_v_reg(s1, ot, s1->T0, d);
1503     } else if (!(s1->prefix & PREFIX_LOCK)) {
1504         gen_op_ld_v(s1, ot, s1->T0, s1->A0);
1505     }
1506     switch(op) {
1507     case OP_ADCL:
1508         gen_compute_eflags_c(s1, s1->tmp4);
1509         if (s1->prefix & PREFIX_LOCK) {
1510             tcg_gen_add_tl(s1->T0, s1->tmp4, s1->T1);
1511             tcg_gen_atomic_add_fetch_tl(s1->T0, s1->A0, s1->T0,
1512                                         s1->mem_index, ot | MO_LE);
1513         } else {
1514             tcg_gen_add_tl(s1->T0, s1->T0, s1->T1);
1515             tcg_gen_add_tl(s1->T0, s1->T0, s1->tmp4);
1516             gen_op_st_rm_T0_A0(s1, ot, d);
1517         }
1518         gen_op_update3_cc(s1, s1->tmp4);
1519         set_cc_op(s1, CC_OP_ADCB + ot);
1520         break;
1521     case OP_SBBL:
1522         gen_compute_eflags_c(s1, s1->tmp4);
1523         if (s1->prefix & PREFIX_LOCK) {
1524             tcg_gen_add_tl(s1->T0, s1->T1, s1->tmp4);
1525             tcg_gen_neg_tl(s1->T0, s1->T0);
1526             tcg_gen_atomic_add_fetch_tl(s1->T0, s1->A0, s1->T0,
1527                                         s1->mem_index, ot | MO_LE);
1528         } else {
1529             tcg_gen_sub_tl(s1->T0, s1->T0, s1->T1);
1530             tcg_gen_sub_tl(s1->T0, s1->T0, s1->tmp4);
1531             gen_op_st_rm_T0_A0(s1, ot, d);
1532         }
1533         gen_op_update3_cc(s1, s1->tmp4);
1534         set_cc_op(s1, CC_OP_SBBB + ot);
1535         break;
1536     case OP_ADDL:
1537         if (s1->prefix & PREFIX_LOCK) {
1538             tcg_gen_atomic_add_fetch_tl(s1->T0, s1->A0, s1->T1,
1539                                         s1->mem_index, ot | MO_LE);
1540         } else {
1541             tcg_gen_add_tl(s1->T0, s1->T0, s1->T1);
1542             gen_op_st_rm_T0_A0(s1, ot, d);
1543         }
1544         gen_op_update2_cc(s1);
1545         set_cc_op(s1, CC_OP_ADDB + ot);
1546         break;
1547     case OP_SUBL:
1548         if (s1->prefix & PREFIX_LOCK) {
1549             tcg_gen_neg_tl(s1->T0, s1->T1);
1550             tcg_gen_atomic_fetch_add_tl(s1->cc_srcT, s1->A0, s1->T0,
1551                                         s1->mem_index, ot | MO_LE);
1552             tcg_gen_sub_tl(s1->T0, s1->cc_srcT, s1->T1);
1553         } else {
1554             tcg_gen_mov_tl(s1->cc_srcT, s1->T0);
1555             tcg_gen_sub_tl(s1->T0, s1->T0, s1->T1);
1556             gen_op_st_rm_T0_A0(s1, ot, d);
1557         }
1558         gen_op_update2_cc(s1);
1559         set_cc_op(s1, CC_OP_SUBB + ot);
1560         break;
1561     default:
1562     case OP_ANDL:
1563         if (s1->prefix & PREFIX_LOCK) {
1564             tcg_gen_atomic_and_fetch_tl(s1->T0, s1->A0, s1->T1,
1565                                         s1->mem_index, ot | MO_LE);
1566         } else {
1567             tcg_gen_and_tl(s1->T0, s1->T0, s1->T1);
1568             gen_op_st_rm_T0_A0(s1, ot, d);
1569         }
1570         gen_op_update1_cc(s1);
1571         set_cc_op(s1, CC_OP_LOGICB + ot);
1572         break;
1573     case OP_ORL:
1574         if (s1->prefix & PREFIX_LOCK) {
1575             tcg_gen_atomic_or_fetch_tl(s1->T0, s1->A0, s1->T1,
1576                                        s1->mem_index, ot | MO_LE);
1577         } else {
1578             tcg_gen_or_tl(s1->T0, s1->T0, s1->T1);
1579             gen_op_st_rm_T0_A0(s1, ot, d);
1580         }
1581         gen_op_update1_cc(s1);
1582         set_cc_op(s1, CC_OP_LOGICB + ot);
1583         break;
1584     case OP_XORL:
1585         if (s1->prefix & PREFIX_LOCK) {
1586             tcg_gen_atomic_xor_fetch_tl(s1->T0, s1->A0, s1->T1,
1587                                         s1->mem_index, ot | MO_LE);
1588         } else {
1589             tcg_gen_xor_tl(s1->T0, s1->T0, s1->T1);
1590             gen_op_st_rm_T0_A0(s1, ot, d);
1591         }
1592         gen_op_update1_cc(s1);
1593         set_cc_op(s1, CC_OP_LOGICB + ot);
1594         break;
1595     case OP_CMPL:
1596         tcg_gen_mov_tl(cpu_cc_src, s1->T1);
1597         tcg_gen_mov_tl(s1->cc_srcT, s1->T0);
1598         tcg_gen_sub_tl(cpu_cc_dst, s1->T0, s1->T1);
1599         set_cc_op(s1, CC_OP_SUBB + ot);
1600         break;
1601     }
1602 }
1603 
1604 /* if d == OR_TMP0, it means memory operand (address in A0) */
1605 static void gen_inc(DisasContext *s1, MemOp ot, int d, int c)
1606 {
1607     if (s1->prefix & PREFIX_LOCK) {
1608         if (d != OR_TMP0) {
1609             /* Lock prefix when destination is not memory */
1610             gen_illegal_opcode(s1);
1611             return;
1612         }
1613         tcg_gen_movi_tl(s1->T0, c > 0 ? 1 : -1);
1614         tcg_gen_atomic_add_fetch_tl(s1->T0, s1->A0, s1->T0,
1615                                     s1->mem_index, ot | MO_LE);
1616     } else {
1617         if (d != OR_TMP0) {
1618             gen_op_mov_v_reg(s1, ot, s1->T0, d);
1619         } else {
1620             gen_op_ld_v(s1, ot, s1->T0, s1->A0);
1621         }
1622         tcg_gen_addi_tl(s1->T0, s1->T0, (c > 0 ? 1 : -1));
1623         gen_op_st_rm_T0_A0(s1, ot, d);
1624     }
1625 
1626     gen_compute_eflags_c(s1, cpu_cc_src);
1627     tcg_gen_mov_tl(cpu_cc_dst, s1->T0);
1628     set_cc_op(s1, (c > 0 ? CC_OP_INCB : CC_OP_DECB) + ot);
1629 }
1630 
1631 static void gen_shift_flags(DisasContext *s, MemOp ot, TCGv result,
1632                             TCGv shm1, TCGv count, bool is_right)
1633 {
1634     TCGv_i32 z32, s32, oldop;
1635     TCGv z_tl;
1636 
1637     /* Store the results into the CC variables.  If we know that the
1638        variable must be dead, store unconditionally.  Otherwise we'll
1639        need to not disrupt the current contents.  */
1640     z_tl = tcg_const_tl(0);
1641     if (cc_op_live[s->cc_op] & USES_CC_DST) {
1642         tcg_gen_movcond_tl(TCG_COND_NE, cpu_cc_dst, count, z_tl,
1643                            result, cpu_cc_dst);
1644     } else {
1645         tcg_gen_mov_tl(cpu_cc_dst, result);
1646     }
1647     if (cc_op_live[s->cc_op] & USES_CC_SRC) {
1648         tcg_gen_movcond_tl(TCG_COND_NE, cpu_cc_src, count, z_tl,
1649                            shm1, cpu_cc_src);
1650     } else {
1651         tcg_gen_mov_tl(cpu_cc_src, shm1);
1652     }
1653     tcg_temp_free(z_tl);
1654 
1655     /* Get the two potential CC_OP values into temporaries.  */
1656     tcg_gen_movi_i32(s->tmp2_i32, (is_right ? CC_OP_SARB : CC_OP_SHLB) + ot);
1657     if (s->cc_op == CC_OP_DYNAMIC) {
1658         oldop = cpu_cc_op;
1659     } else {
1660         tcg_gen_movi_i32(s->tmp3_i32, s->cc_op);
1661         oldop = s->tmp3_i32;
1662     }
1663 
1664     /* Conditionally store the CC_OP value.  */
1665     z32 = tcg_const_i32(0);
1666     s32 = tcg_temp_new_i32();
1667     tcg_gen_trunc_tl_i32(s32, count);
1668     tcg_gen_movcond_i32(TCG_COND_NE, cpu_cc_op, s32, z32, s->tmp2_i32, oldop);
1669     tcg_temp_free_i32(z32);
1670     tcg_temp_free_i32(s32);
1671 
1672     /* The CC_OP value is no longer predictable.  */
1673     set_cc_op(s, CC_OP_DYNAMIC);
1674 }
1675 
1676 static void gen_shift_rm_T1(DisasContext *s, MemOp ot, int op1,
1677                             int is_right, int is_arith)
1678 {
1679     target_ulong mask = (ot == MO_64 ? 0x3f : 0x1f);
1680 
1681     /* load */
1682     if (op1 == OR_TMP0) {
1683         gen_op_ld_v(s, ot, s->T0, s->A0);
1684     } else {
1685         gen_op_mov_v_reg(s, ot, s->T0, op1);
1686     }
1687 
1688     tcg_gen_andi_tl(s->T1, s->T1, mask);
1689     tcg_gen_subi_tl(s->tmp0, s->T1, 1);
1690 
1691     if (is_right) {
1692         if (is_arith) {
1693             gen_exts(ot, s->T0);
1694             tcg_gen_sar_tl(s->tmp0, s->T0, s->tmp0);
1695             tcg_gen_sar_tl(s->T0, s->T0, s->T1);
1696         } else {
1697             gen_extu(ot, s->T0);
1698             tcg_gen_shr_tl(s->tmp0, s->T0, s->tmp0);
1699             tcg_gen_shr_tl(s->T0, s->T0, s->T1);
1700         }
1701     } else {
1702         tcg_gen_shl_tl(s->tmp0, s->T0, s->tmp0);
1703         tcg_gen_shl_tl(s->T0, s->T0, s->T1);
1704     }
1705 
1706     /* store */
1707     gen_op_st_rm_T0_A0(s, ot, op1);
1708 
1709     gen_shift_flags(s, ot, s->T0, s->tmp0, s->T1, is_right);
1710 }
1711 
1712 static void gen_shift_rm_im(DisasContext *s, MemOp ot, int op1, int op2,
1713                             int is_right, int is_arith)
1714 {
1715     int mask = (ot == MO_64 ? 0x3f : 0x1f);
1716 
1717     /* load */
1718     if (op1 == OR_TMP0)
1719         gen_op_ld_v(s, ot, s->T0, s->A0);
1720     else
1721         gen_op_mov_v_reg(s, ot, s->T0, op1);
1722 
1723     op2 &= mask;
1724     if (op2 != 0) {
1725         if (is_right) {
1726             if (is_arith) {
1727                 gen_exts(ot, s->T0);
1728                 tcg_gen_sari_tl(s->tmp4, s->T0, op2 - 1);
1729                 tcg_gen_sari_tl(s->T0, s->T0, op2);
1730             } else {
1731                 gen_extu(ot, s->T0);
1732                 tcg_gen_shri_tl(s->tmp4, s->T0, op2 - 1);
1733                 tcg_gen_shri_tl(s->T0, s->T0, op2);
1734             }
1735         } else {
1736             tcg_gen_shli_tl(s->tmp4, s->T0, op2 - 1);
1737             tcg_gen_shli_tl(s->T0, s->T0, op2);
1738         }
1739     }
1740 
1741     /* store */
1742     gen_op_st_rm_T0_A0(s, ot, op1);
1743 
1744     /* update eflags if non zero shift */
1745     if (op2 != 0) {
1746         tcg_gen_mov_tl(cpu_cc_src, s->tmp4);
1747         tcg_gen_mov_tl(cpu_cc_dst, s->T0);
1748         set_cc_op(s, (is_right ? CC_OP_SARB : CC_OP_SHLB) + ot);
1749     }
1750 }
1751 
1752 static void gen_rot_rm_T1(DisasContext *s, MemOp ot, int op1, int is_right)
1753 {
1754     target_ulong mask = (ot == MO_64 ? 0x3f : 0x1f);
1755     TCGv_i32 t0, t1;
1756 
1757     /* load */
1758     if (op1 == OR_TMP0) {
1759         gen_op_ld_v(s, ot, s->T0, s->A0);
1760     } else {
1761         gen_op_mov_v_reg(s, ot, s->T0, op1);
1762     }
1763 
1764     tcg_gen_andi_tl(s->T1, s->T1, mask);
1765 
1766     switch (ot) {
1767     case MO_8:
1768         /* Replicate the 8-bit input so that a 32-bit rotate works.  */
1769         tcg_gen_ext8u_tl(s->T0, s->T0);
1770         tcg_gen_muli_tl(s->T0, s->T0, 0x01010101);
1771         goto do_long;
1772     case MO_16:
1773         /* Replicate the 16-bit input so that a 32-bit rotate works.  */
1774         tcg_gen_deposit_tl(s->T0, s->T0, s->T0, 16, 16);
1775         goto do_long;
1776     do_long:
1777 #ifdef TARGET_X86_64
1778     case MO_32:
1779         tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
1780         tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
1781         if (is_right) {
1782             tcg_gen_rotr_i32(s->tmp2_i32, s->tmp2_i32, s->tmp3_i32);
1783         } else {
1784             tcg_gen_rotl_i32(s->tmp2_i32, s->tmp2_i32, s->tmp3_i32);
1785         }
1786         tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
1787         break;
1788 #endif
1789     default:
1790         if (is_right) {
1791             tcg_gen_rotr_tl(s->T0, s->T0, s->T1);
1792         } else {
1793             tcg_gen_rotl_tl(s->T0, s->T0, s->T1);
1794         }
1795         break;
1796     }
1797 
1798     /* store */
1799     gen_op_st_rm_T0_A0(s, ot, op1);
1800 
1801     /* We'll need the flags computed into CC_SRC.  */
1802     gen_compute_eflags(s);
1803 
1804     /* The value that was "rotated out" is now present at the other end
1805        of the word.  Compute C into CC_DST and O into CC_SRC2.  Note that
1806        since we've computed the flags into CC_SRC, these variables are
1807        currently dead.  */
1808     if (is_right) {
1809         tcg_gen_shri_tl(cpu_cc_src2, s->T0, mask - 1);
1810         tcg_gen_shri_tl(cpu_cc_dst, s->T0, mask);
1811         tcg_gen_andi_tl(cpu_cc_dst, cpu_cc_dst, 1);
1812     } else {
1813         tcg_gen_shri_tl(cpu_cc_src2, s->T0, mask);
1814         tcg_gen_andi_tl(cpu_cc_dst, s->T0, 1);
1815     }
1816     tcg_gen_andi_tl(cpu_cc_src2, cpu_cc_src2, 1);
1817     tcg_gen_xor_tl(cpu_cc_src2, cpu_cc_src2, cpu_cc_dst);
1818 
1819     /* Now conditionally store the new CC_OP value.  If the shift count
1820        is 0 we keep the CC_OP_EFLAGS setting so that only CC_SRC is live.
1821        Otherwise reuse CC_OP_ADCOX which have the C and O flags split out
1822        exactly as we computed above.  */
1823     t0 = tcg_const_i32(0);
1824     t1 = tcg_temp_new_i32();
1825     tcg_gen_trunc_tl_i32(t1, s->T1);
1826     tcg_gen_movi_i32(s->tmp2_i32, CC_OP_ADCOX);
1827     tcg_gen_movi_i32(s->tmp3_i32, CC_OP_EFLAGS);
1828     tcg_gen_movcond_i32(TCG_COND_NE, cpu_cc_op, t1, t0,
1829                         s->tmp2_i32, s->tmp3_i32);
1830     tcg_temp_free_i32(t0);
1831     tcg_temp_free_i32(t1);
1832 
1833     /* The CC_OP value is no longer predictable.  */
1834     set_cc_op(s, CC_OP_DYNAMIC);
1835 }
1836 
1837 static void gen_rot_rm_im(DisasContext *s, MemOp ot, int op1, int op2,
1838                           int is_right)
1839 {
1840     int mask = (ot == MO_64 ? 0x3f : 0x1f);
1841     int shift;
1842 
1843     /* load */
1844     if (op1 == OR_TMP0) {
1845         gen_op_ld_v(s, ot, s->T0, s->A0);
1846     } else {
1847         gen_op_mov_v_reg(s, ot, s->T0, op1);
1848     }
1849 
1850     op2 &= mask;
1851     if (op2 != 0) {
1852         switch (ot) {
1853 #ifdef TARGET_X86_64
1854         case MO_32:
1855             tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
1856             if (is_right) {
1857                 tcg_gen_rotri_i32(s->tmp2_i32, s->tmp2_i32, op2);
1858             } else {
1859                 tcg_gen_rotli_i32(s->tmp2_i32, s->tmp2_i32, op2);
1860             }
1861             tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
1862             break;
1863 #endif
1864         default:
1865             if (is_right) {
1866                 tcg_gen_rotri_tl(s->T0, s->T0, op2);
1867             } else {
1868                 tcg_gen_rotli_tl(s->T0, s->T0, op2);
1869             }
1870             break;
1871         case MO_8:
1872             mask = 7;
1873             goto do_shifts;
1874         case MO_16:
1875             mask = 15;
1876         do_shifts:
1877             shift = op2 & mask;
1878             if (is_right) {
1879                 shift = mask + 1 - shift;
1880             }
1881             gen_extu(ot, s->T0);
1882             tcg_gen_shli_tl(s->tmp0, s->T0, shift);
1883             tcg_gen_shri_tl(s->T0, s->T0, mask + 1 - shift);
1884             tcg_gen_or_tl(s->T0, s->T0, s->tmp0);
1885             break;
1886         }
1887     }
1888 
1889     /* store */
1890     gen_op_st_rm_T0_A0(s, ot, op1);
1891 
1892     if (op2 != 0) {
1893         /* Compute the flags into CC_SRC.  */
1894         gen_compute_eflags(s);
1895 
1896         /* The value that was "rotated out" is now present at the other end
1897            of the word.  Compute C into CC_DST and O into CC_SRC2.  Note that
1898            since we've computed the flags into CC_SRC, these variables are
1899            currently dead.  */
1900         if (is_right) {
1901             tcg_gen_shri_tl(cpu_cc_src2, s->T0, mask - 1);
1902             tcg_gen_shri_tl(cpu_cc_dst, s->T0, mask);
1903             tcg_gen_andi_tl(cpu_cc_dst, cpu_cc_dst, 1);
1904         } else {
1905             tcg_gen_shri_tl(cpu_cc_src2, s->T0, mask);
1906             tcg_gen_andi_tl(cpu_cc_dst, s->T0, 1);
1907         }
1908         tcg_gen_andi_tl(cpu_cc_src2, cpu_cc_src2, 1);
1909         tcg_gen_xor_tl(cpu_cc_src2, cpu_cc_src2, cpu_cc_dst);
1910         set_cc_op(s, CC_OP_ADCOX);
1911     }
1912 }
1913 
1914 /* XXX: add faster immediate = 1 case */
1915 static void gen_rotc_rm_T1(DisasContext *s, MemOp ot, int op1,
1916                            int is_right)
1917 {
1918     gen_compute_eflags(s);
1919     assert(s->cc_op == CC_OP_EFLAGS);
1920 
1921     /* load */
1922     if (op1 == OR_TMP0)
1923         gen_op_ld_v(s, ot, s->T0, s->A0);
1924     else
1925         gen_op_mov_v_reg(s, ot, s->T0, op1);
1926 
1927     if (is_right) {
1928         switch (ot) {
1929         case MO_8:
1930             gen_helper_rcrb(s->T0, cpu_env, s->T0, s->T1);
1931             break;
1932         case MO_16:
1933             gen_helper_rcrw(s->T0, cpu_env, s->T0, s->T1);
1934             break;
1935         case MO_32:
1936             gen_helper_rcrl(s->T0, cpu_env, s->T0, s->T1);
1937             break;
1938 #ifdef TARGET_X86_64
1939         case MO_64:
1940             gen_helper_rcrq(s->T0, cpu_env, s->T0, s->T1);
1941             break;
1942 #endif
1943         default:
1944             tcg_abort();
1945         }
1946     } else {
1947         switch (ot) {
1948         case MO_8:
1949             gen_helper_rclb(s->T0, cpu_env, s->T0, s->T1);
1950             break;
1951         case MO_16:
1952             gen_helper_rclw(s->T0, cpu_env, s->T0, s->T1);
1953             break;
1954         case MO_32:
1955             gen_helper_rcll(s->T0, cpu_env, s->T0, s->T1);
1956             break;
1957 #ifdef TARGET_X86_64
1958         case MO_64:
1959             gen_helper_rclq(s->T0, cpu_env, s->T0, s->T1);
1960             break;
1961 #endif
1962         default:
1963             tcg_abort();
1964         }
1965     }
1966     /* store */
1967     gen_op_st_rm_T0_A0(s, ot, op1);
1968 }
1969 
1970 /* XXX: add faster immediate case */
1971 static void gen_shiftd_rm_T1(DisasContext *s, MemOp ot, int op1,
1972                              bool is_right, TCGv count_in)
1973 {
1974     target_ulong mask = (ot == MO_64 ? 63 : 31);
1975     TCGv count;
1976 
1977     /* load */
1978     if (op1 == OR_TMP0) {
1979         gen_op_ld_v(s, ot, s->T0, s->A0);
1980     } else {
1981         gen_op_mov_v_reg(s, ot, s->T0, op1);
1982     }
1983 
1984     count = tcg_temp_new();
1985     tcg_gen_andi_tl(count, count_in, mask);
1986 
1987     switch (ot) {
1988     case MO_16:
1989         /* Note: we implement the Intel behaviour for shift count > 16.
1990            This means "shrdw C, B, A" shifts A:B:A >> C.  Build the B:A
1991            portion by constructing it as a 32-bit value.  */
1992         if (is_right) {
1993             tcg_gen_deposit_tl(s->tmp0, s->T0, s->T1, 16, 16);
1994             tcg_gen_mov_tl(s->T1, s->T0);
1995             tcg_gen_mov_tl(s->T0, s->tmp0);
1996         } else {
1997             tcg_gen_deposit_tl(s->T1, s->T0, s->T1, 16, 16);
1998         }
1999         /*
2000          * If TARGET_X86_64 defined then fall through into MO_32 case,
2001          * otherwise fall through default case.
2002          */
2003     case MO_32:
2004 #ifdef TARGET_X86_64
2005         /* Concatenate the two 32-bit values and use a 64-bit shift.  */
2006         tcg_gen_subi_tl(s->tmp0, count, 1);
2007         if (is_right) {
2008             tcg_gen_concat_tl_i64(s->T0, s->T0, s->T1);
2009             tcg_gen_shr_i64(s->tmp0, s->T0, s->tmp0);
2010             tcg_gen_shr_i64(s->T0, s->T0, count);
2011         } else {
2012             tcg_gen_concat_tl_i64(s->T0, s->T1, s->T0);
2013             tcg_gen_shl_i64(s->tmp0, s->T0, s->tmp0);
2014             tcg_gen_shl_i64(s->T0, s->T0, count);
2015             tcg_gen_shri_i64(s->tmp0, s->tmp0, 32);
2016             tcg_gen_shri_i64(s->T0, s->T0, 32);
2017         }
2018         break;
2019 #endif
2020     default:
2021         tcg_gen_subi_tl(s->tmp0, count, 1);
2022         if (is_right) {
2023             tcg_gen_shr_tl(s->tmp0, s->T0, s->tmp0);
2024 
2025             tcg_gen_subfi_tl(s->tmp4, mask + 1, count);
2026             tcg_gen_shr_tl(s->T0, s->T0, count);
2027             tcg_gen_shl_tl(s->T1, s->T1, s->tmp4);
2028         } else {
2029             tcg_gen_shl_tl(s->tmp0, s->T0, s->tmp0);
2030             if (ot == MO_16) {
2031                 /* Only needed if count > 16, for Intel behaviour.  */
2032                 tcg_gen_subfi_tl(s->tmp4, 33, count);
2033                 tcg_gen_shr_tl(s->tmp4, s->T1, s->tmp4);
2034                 tcg_gen_or_tl(s->tmp0, s->tmp0, s->tmp4);
2035             }
2036 
2037             tcg_gen_subfi_tl(s->tmp4, mask + 1, count);
2038             tcg_gen_shl_tl(s->T0, s->T0, count);
2039             tcg_gen_shr_tl(s->T1, s->T1, s->tmp4);
2040         }
2041         tcg_gen_movi_tl(s->tmp4, 0);
2042         tcg_gen_movcond_tl(TCG_COND_EQ, s->T1, count, s->tmp4,
2043                            s->tmp4, s->T1);
2044         tcg_gen_or_tl(s->T0, s->T0, s->T1);
2045         break;
2046     }
2047 
2048     /* store */
2049     gen_op_st_rm_T0_A0(s, ot, op1);
2050 
2051     gen_shift_flags(s, ot, s->T0, s->tmp0, count, is_right);
2052     tcg_temp_free(count);
2053 }
2054 
2055 static void gen_shift(DisasContext *s1, int op, MemOp ot, int d, int s)
2056 {
2057     if (s != OR_TMP1)
2058         gen_op_mov_v_reg(s1, ot, s1->T1, s);
2059     switch(op) {
2060     case OP_ROL:
2061         gen_rot_rm_T1(s1, ot, d, 0);
2062         break;
2063     case OP_ROR:
2064         gen_rot_rm_T1(s1, ot, d, 1);
2065         break;
2066     case OP_SHL:
2067     case OP_SHL1:
2068         gen_shift_rm_T1(s1, ot, d, 0, 0);
2069         break;
2070     case OP_SHR:
2071         gen_shift_rm_T1(s1, ot, d, 1, 0);
2072         break;
2073     case OP_SAR:
2074         gen_shift_rm_T1(s1, ot, d, 1, 1);
2075         break;
2076     case OP_RCL:
2077         gen_rotc_rm_T1(s1, ot, d, 0);
2078         break;
2079     case OP_RCR:
2080         gen_rotc_rm_T1(s1, ot, d, 1);
2081         break;
2082     }
2083 }
2084 
2085 static void gen_shifti(DisasContext *s1, int op, MemOp ot, int d, int c)
2086 {
2087     switch(op) {
2088     case OP_ROL:
2089         gen_rot_rm_im(s1, ot, d, c, 0);
2090         break;
2091     case OP_ROR:
2092         gen_rot_rm_im(s1, ot, d, c, 1);
2093         break;
2094     case OP_SHL:
2095     case OP_SHL1:
2096         gen_shift_rm_im(s1, ot, d, c, 0, 0);
2097         break;
2098     case OP_SHR:
2099         gen_shift_rm_im(s1, ot, d, c, 1, 0);
2100         break;
2101     case OP_SAR:
2102         gen_shift_rm_im(s1, ot, d, c, 1, 1);
2103         break;
2104     default:
2105         /* currently not optimized */
2106         tcg_gen_movi_tl(s1->T1, c);
2107         gen_shift(s1, op, ot, d, OR_TMP1);
2108         break;
2109     }
2110 }
2111 
2112 #define X86_MAX_INSN_LENGTH 15
2113 
2114 static uint64_t advance_pc(CPUX86State *env, DisasContext *s, int num_bytes)
2115 {
2116     uint64_t pc = s->pc;
2117 
2118     /* This is a subsequent insn that crosses a page boundary.  */
2119     if (s->base.num_insns > 1 &&
2120         !is_same_page(&s->base, s->pc + num_bytes - 1)) {
2121         siglongjmp(s->jmpbuf, 2);
2122     }
2123 
2124     s->pc += num_bytes;
2125     if (unlikely(cur_insn_len(s) > X86_MAX_INSN_LENGTH)) {
2126         /* If the instruction's 16th byte is on a different page than the 1st, a
2127          * page fault on the second page wins over the general protection fault
2128          * caused by the instruction being too long.
2129          * This can happen even if the operand is only one byte long!
2130          */
2131         if (((s->pc - 1) ^ (pc - 1)) & TARGET_PAGE_MASK) {
2132             volatile uint8_t unused =
2133                 cpu_ldub_code(env, (s->pc - 1) & TARGET_PAGE_MASK);
2134             (void) unused;
2135         }
2136         siglongjmp(s->jmpbuf, 1);
2137     }
2138 
2139     return pc;
2140 }
2141 
2142 static inline uint8_t x86_ldub_code(CPUX86State *env, DisasContext *s)
2143 {
2144     return translator_ldub(env, &s->base, advance_pc(env, s, 1));
2145 }
2146 
2147 static inline int16_t x86_ldsw_code(CPUX86State *env, DisasContext *s)
2148 {
2149     return translator_lduw(env, &s->base, advance_pc(env, s, 2));
2150 }
2151 
2152 static inline uint16_t x86_lduw_code(CPUX86State *env, DisasContext *s)
2153 {
2154     return translator_lduw(env, &s->base, advance_pc(env, s, 2));
2155 }
2156 
2157 static inline uint32_t x86_ldl_code(CPUX86State *env, DisasContext *s)
2158 {
2159     return translator_ldl(env, &s->base, advance_pc(env, s, 4));
2160 }
2161 
2162 #ifdef TARGET_X86_64
2163 static inline uint64_t x86_ldq_code(CPUX86State *env, DisasContext *s)
2164 {
2165     return translator_ldq(env, &s->base, advance_pc(env, s, 8));
2166 }
2167 #endif
2168 
2169 /* Decompose an address.  */
2170 
2171 typedef struct AddressParts {
2172     int def_seg;
2173     int base;
2174     int index;
2175     int scale;
2176     target_long disp;
2177 } AddressParts;
2178 
2179 static AddressParts gen_lea_modrm_0(CPUX86State *env, DisasContext *s,
2180                                     int modrm)
2181 {
2182     int def_seg, base, index, scale, mod, rm;
2183     target_long disp;
2184     bool havesib;
2185 
2186     def_seg = R_DS;
2187     index = -1;
2188     scale = 0;
2189     disp = 0;
2190 
2191     mod = (modrm >> 6) & 3;
2192     rm = modrm & 7;
2193     base = rm | REX_B(s);
2194 
2195     if (mod == 3) {
2196         /* Normally filtered out earlier, but including this path
2197            simplifies multi-byte nop, as well as bndcl, bndcu, bndcn.  */
2198         goto done;
2199     }
2200 
2201     switch (s->aflag) {
2202     case MO_64:
2203     case MO_32:
2204         havesib = 0;
2205         if (rm == 4) {
2206             int code = x86_ldub_code(env, s);
2207             scale = (code >> 6) & 3;
2208             index = ((code >> 3) & 7) | REX_X(s);
2209             if (index == 4) {
2210                 index = -1;  /* no index */
2211             }
2212             base = (code & 7) | REX_B(s);
2213             havesib = 1;
2214         }
2215 
2216         switch (mod) {
2217         case 0:
2218             if ((base & 7) == 5) {
2219                 base = -1;
2220                 disp = (int32_t)x86_ldl_code(env, s);
2221                 if (CODE64(s) && !havesib) {
2222                     base = -2;
2223                     disp += s->pc + s->rip_offset;
2224                 }
2225             }
2226             break;
2227         case 1:
2228             disp = (int8_t)x86_ldub_code(env, s);
2229             break;
2230         default:
2231         case 2:
2232             disp = (int32_t)x86_ldl_code(env, s);
2233             break;
2234         }
2235 
2236         /* For correct popl handling with esp.  */
2237         if (base == R_ESP && s->popl_esp_hack) {
2238             disp += s->popl_esp_hack;
2239         }
2240         if (base == R_EBP || base == R_ESP) {
2241             def_seg = R_SS;
2242         }
2243         break;
2244 
2245     case MO_16:
2246         if (mod == 0) {
2247             if (rm == 6) {
2248                 base = -1;
2249                 disp = x86_lduw_code(env, s);
2250                 break;
2251             }
2252         } else if (mod == 1) {
2253             disp = (int8_t)x86_ldub_code(env, s);
2254         } else {
2255             disp = (int16_t)x86_lduw_code(env, s);
2256         }
2257 
2258         switch (rm) {
2259         case 0:
2260             base = R_EBX;
2261             index = R_ESI;
2262             break;
2263         case 1:
2264             base = R_EBX;
2265             index = R_EDI;
2266             break;
2267         case 2:
2268             base = R_EBP;
2269             index = R_ESI;
2270             def_seg = R_SS;
2271             break;
2272         case 3:
2273             base = R_EBP;
2274             index = R_EDI;
2275             def_seg = R_SS;
2276             break;
2277         case 4:
2278             base = R_ESI;
2279             break;
2280         case 5:
2281             base = R_EDI;
2282             break;
2283         case 6:
2284             base = R_EBP;
2285             def_seg = R_SS;
2286             break;
2287         default:
2288         case 7:
2289             base = R_EBX;
2290             break;
2291         }
2292         break;
2293 
2294     default:
2295         tcg_abort();
2296     }
2297 
2298  done:
2299     return (AddressParts){ def_seg, base, index, scale, disp };
2300 }
2301 
2302 /* Compute the address, with a minimum number of TCG ops.  */
2303 static TCGv gen_lea_modrm_1(DisasContext *s, AddressParts a, bool is_vsib)
2304 {
2305     TCGv ea = NULL;
2306 
2307     if (a.index >= 0 && !is_vsib) {
2308         if (a.scale == 0) {
2309             ea = cpu_regs[a.index];
2310         } else {
2311             tcg_gen_shli_tl(s->A0, cpu_regs[a.index], a.scale);
2312             ea = s->A0;
2313         }
2314         if (a.base >= 0) {
2315             tcg_gen_add_tl(s->A0, ea, cpu_regs[a.base]);
2316             ea = s->A0;
2317         }
2318     } else if (a.base >= 0) {
2319         ea = cpu_regs[a.base];
2320     }
2321     if (!ea) {
2322         if (TARGET_TB_PCREL && a.base == -2) {
2323             /* With cpu_eip ~= pc_save, the expression is pc-relative. */
2324             tcg_gen_addi_tl(s->A0, cpu_eip, a.disp - s->pc_save);
2325         } else {
2326             tcg_gen_movi_tl(s->A0, a.disp);
2327         }
2328         ea = s->A0;
2329     } else if (a.disp != 0) {
2330         tcg_gen_addi_tl(s->A0, ea, a.disp);
2331         ea = s->A0;
2332     }
2333 
2334     return ea;
2335 }
2336 
2337 static void gen_lea_modrm(CPUX86State *env, DisasContext *s, int modrm)
2338 {
2339     AddressParts a = gen_lea_modrm_0(env, s, modrm);
2340     TCGv ea = gen_lea_modrm_1(s, a, false);
2341     gen_lea_v_seg(s, s->aflag, ea, a.def_seg, s->override);
2342 }
2343 
2344 static void gen_nop_modrm(CPUX86State *env, DisasContext *s, int modrm)
2345 {
2346     (void)gen_lea_modrm_0(env, s, modrm);
2347 }
2348 
2349 /* Used for BNDCL, BNDCU, BNDCN.  */
2350 static void gen_bndck(CPUX86State *env, DisasContext *s, int modrm,
2351                       TCGCond cond, TCGv_i64 bndv)
2352 {
2353     AddressParts a = gen_lea_modrm_0(env, s, modrm);
2354     TCGv ea = gen_lea_modrm_1(s, a, false);
2355 
2356     tcg_gen_extu_tl_i64(s->tmp1_i64, ea);
2357     if (!CODE64(s)) {
2358         tcg_gen_ext32u_i64(s->tmp1_i64, s->tmp1_i64);
2359     }
2360     tcg_gen_setcond_i64(cond, s->tmp1_i64, s->tmp1_i64, bndv);
2361     tcg_gen_extrl_i64_i32(s->tmp2_i32, s->tmp1_i64);
2362     gen_helper_bndck(cpu_env, s->tmp2_i32);
2363 }
2364 
2365 /* used for LEA and MOV AX, mem */
2366 static void gen_add_A0_ds_seg(DisasContext *s)
2367 {
2368     gen_lea_v_seg(s, s->aflag, s->A0, R_DS, s->override);
2369 }
2370 
2371 /* generate modrm memory load or store of 'reg'. TMP0 is used if reg ==
2372    OR_TMP0 */
2373 static void gen_ldst_modrm(CPUX86State *env, DisasContext *s, int modrm,
2374                            MemOp ot, int reg, int is_store)
2375 {
2376     int mod, rm;
2377 
2378     mod = (modrm >> 6) & 3;
2379     rm = (modrm & 7) | REX_B(s);
2380     if (mod == 3) {
2381         if (is_store) {
2382             if (reg != OR_TMP0)
2383                 gen_op_mov_v_reg(s, ot, s->T0, reg);
2384             gen_op_mov_reg_v(s, ot, rm, s->T0);
2385         } else {
2386             gen_op_mov_v_reg(s, ot, s->T0, rm);
2387             if (reg != OR_TMP0)
2388                 gen_op_mov_reg_v(s, ot, reg, s->T0);
2389         }
2390     } else {
2391         gen_lea_modrm(env, s, modrm);
2392         if (is_store) {
2393             if (reg != OR_TMP0)
2394                 gen_op_mov_v_reg(s, ot, s->T0, reg);
2395             gen_op_st_v(s, ot, s->T0, s->A0);
2396         } else {
2397             gen_op_ld_v(s, ot, s->T0, s->A0);
2398             if (reg != OR_TMP0)
2399                 gen_op_mov_reg_v(s, ot, reg, s->T0);
2400         }
2401     }
2402 }
2403 
2404 static target_ulong insn_get_addr(CPUX86State *env, DisasContext *s, MemOp ot)
2405 {
2406     target_ulong ret;
2407 
2408     switch (ot) {
2409     case MO_8:
2410         ret = x86_ldub_code(env, s);
2411         break;
2412     case MO_16:
2413         ret = x86_lduw_code(env, s);
2414         break;
2415     case MO_32:
2416         ret = x86_ldl_code(env, s);
2417         break;
2418 #ifdef TARGET_X86_64
2419     case MO_64:
2420         ret = x86_ldq_code(env, s);
2421         break;
2422 #endif
2423     default:
2424         g_assert_not_reached();
2425     }
2426     return ret;
2427 }
2428 
2429 static inline uint32_t insn_get(CPUX86State *env, DisasContext *s, MemOp ot)
2430 {
2431     uint32_t ret;
2432 
2433     switch (ot) {
2434     case MO_8:
2435         ret = x86_ldub_code(env, s);
2436         break;
2437     case MO_16:
2438         ret = x86_lduw_code(env, s);
2439         break;
2440     case MO_32:
2441 #ifdef TARGET_X86_64
2442     case MO_64:
2443 #endif
2444         ret = x86_ldl_code(env, s);
2445         break;
2446     default:
2447         tcg_abort();
2448     }
2449     return ret;
2450 }
2451 
2452 static target_long insn_get_signed(CPUX86State *env, DisasContext *s, MemOp ot)
2453 {
2454     target_long ret;
2455 
2456     switch (ot) {
2457     case MO_8:
2458         ret = (int8_t) x86_ldub_code(env, s);
2459         break;
2460     case MO_16:
2461         ret = (int16_t) x86_lduw_code(env, s);
2462         break;
2463     case MO_32:
2464         ret = (int32_t) x86_ldl_code(env, s);
2465         break;
2466 #ifdef TARGET_X86_64
2467     case MO_64:
2468         ret = x86_ldq_code(env, s);
2469         break;
2470 #endif
2471     default:
2472         g_assert_not_reached();
2473     }
2474     return ret;
2475 }
2476 
2477 static inline int insn_const_size(MemOp ot)
2478 {
2479     if (ot <= MO_32) {
2480         return 1 << ot;
2481     } else {
2482         return 4;
2483     }
2484 }
2485 
2486 static void gen_jcc(DisasContext *s, int b, int diff)
2487 {
2488     TCGLabel *l1 = gen_new_label();
2489 
2490     gen_jcc1(s, b, l1);
2491     gen_jmp_rel_csize(s, 0, 1);
2492     gen_set_label(l1);
2493     gen_jmp_rel(s, s->dflag, diff, 0);
2494 }
2495 
2496 static void gen_cmovcc1(CPUX86State *env, DisasContext *s, MemOp ot, int b,
2497                         int modrm, int reg)
2498 {
2499     CCPrepare cc;
2500 
2501     gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 0);
2502 
2503     cc = gen_prepare_cc(s, b, s->T1);
2504     if (cc.mask != -1) {
2505         TCGv t0 = tcg_temp_new();
2506         tcg_gen_andi_tl(t0, cc.reg, cc.mask);
2507         cc.reg = t0;
2508     }
2509     if (!cc.use_reg2) {
2510         cc.reg2 = tcg_const_tl(cc.imm);
2511     }
2512 
2513     tcg_gen_movcond_tl(cc.cond, s->T0, cc.reg, cc.reg2,
2514                        s->T0, cpu_regs[reg]);
2515     gen_op_mov_reg_v(s, ot, reg, s->T0);
2516 
2517     if (cc.mask != -1) {
2518         tcg_temp_free(cc.reg);
2519     }
2520     if (!cc.use_reg2) {
2521         tcg_temp_free(cc.reg2);
2522     }
2523 }
2524 
2525 static inline void gen_op_movl_T0_seg(DisasContext *s, X86Seg seg_reg)
2526 {
2527     tcg_gen_ld32u_tl(s->T0, cpu_env,
2528                      offsetof(CPUX86State,segs[seg_reg].selector));
2529 }
2530 
2531 static inline void gen_op_movl_seg_T0_vm(DisasContext *s, X86Seg seg_reg)
2532 {
2533     tcg_gen_ext16u_tl(s->T0, s->T0);
2534     tcg_gen_st32_tl(s->T0, cpu_env,
2535                     offsetof(CPUX86State,segs[seg_reg].selector));
2536     tcg_gen_shli_tl(cpu_seg_base[seg_reg], s->T0, 4);
2537 }
2538 
2539 /* move T0 to seg_reg and compute if the CPU state may change. Never
2540    call this function with seg_reg == R_CS */
2541 static void gen_movl_seg_T0(DisasContext *s, X86Seg seg_reg)
2542 {
2543     if (PE(s) && !VM86(s)) {
2544         tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
2545         gen_helper_load_seg(cpu_env, tcg_const_i32(seg_reg), s->tmp2_i32);
2546         /* abort translation because the addseg value may change or
2547            because ss32 may change. For R_SS, translation must always
2548            stop as a special handling must be done to disable hardware
2549            interrupts for the next instruction */
2550         if (seg_reg == R_SS) {
2551             s->base.is_jmp = DISAS_EOB_INHIBIT_IRQ;
2552         } else if (CODE32(s) && seg_reg < R_FS) {
2553             s->base.is_jmp = DISAS_EOB_NEXT;
2554         }
2555     } else {
2556         gen_op_movl_seg_T0_vm(s, seg_reg);
2557         if (seg_reg == R_SS) {
2558             s->base.is_jmp = DISAS_EOB_INHIBIT_IRQ;
2559         }
2560     }
2561 }
2562 
2563 static void gen_svm_check_intercept(DisasContext *s, uint32_t type)
2564 {
2565     /* no SVM activated; fast case */
2566     if (likely(!GUEST(s))) {
2567         return;
2568     }
2569     gen_helper_svm_check_intercept(cpu_env, tcg_constant_i32(type));
2570 }
2571 
2572 static inline void gen_stack_update(DisasContext *s, int addend)
2573 {
2574     gen_op_add_reg_im(s, mo_stacksize(s), R_ESP, addend);
2575 }
2576 
2577 /* Generate a push. It depends on ss32, addseg and dflag.  */
2578 static void gen_push_v(DisasContext *s, TCGv val)
2579 {
2580     MemOp d_ot = mo_pushpop(s, s->dflag);
2581     MemOp a_ot = mo_stacksize(s);
2582     int size = 1 << d_ot;
2583     TCGv new_esp = s->A0;
2584 
2585     tcg_gen_subi_tl(s->A0, cpu_regs[R_ESP], size);
2586 
2587     if (!CODE64(s)) {
2588         if (ADDSEG(s)) {
2589             new_esp = s->tmp4;
2590             tcg_gen_mov_tl(new_esp, s->A0);
2591         }
2592         gen_lea_v_seg(s, a_ot, s->A0, R_SS, -1);
2593     }
2594 
2595     gen_op_st_v(s, d_ot, val, s->A0);
2596     gen_op_mov_reg_v(s, a_ot, R_ESP, new_esp);
2597 }
2598 
2599 /* two step pop is necessary for precise exceptions */
2600 static MemOp gen_pop_T0(DisasContext *s)
2601 {
2602     MemOp d_ot = mo_pushpop(s, s->dflag);
2603 
2604     gen_lea_v_seg(s, mo_stacksize(s), cpu_regs[R_ESP], R_SS, -1);
2605     gen_op_ld_v(s, d_ot, s->T0, s->A0);
2606 
2607     return d_ot;
2608 }
2609 
2610 static inline void gen_pop_update(DisasContext *s, MemOp ot)
2611 {
2612     gen_stack_update(s, 1 << ot);
2613 }
2614 
2615 static inline void gen_stack_A0(DisasContext *s)
2616 {
2617     gen_lea_v_seg(s, SS32(s) ? MO_32 : MO_16, cpu_regs[R_ESP], R_SS, -1);
2618 }
2619 
2620 static void gen_pusha(DisasContext *s)
2621 {
2622     MemOp s_ot = SS32(s) ? MO_32 : MO_16;
2623     MemOp d_ot = s->dflag;
2624     int size = 1 << d_ot;
2625     int i;
2626 
2627     for (i = 0; i < 8; i++) {
2628         tcg_gen_addi_tl(s->A0, cpu_regs[R_ESP], (i - 8) * size);
2629         gen_lea_v_seg(s, s_ot, s->A0, R_SS, -1);
2630         gen_op_st_v(s, d_ot, cpu_regs[7 - i], s->A0);
2631     }
2632 
2633     gen_stack_update(s, -8 * size);
2634 }
2635 
2636 static void gen_popa(DisasContext *s)
2637 {
2638     MemOp s_ot = SS32(s) ? MO_32 : MO_16;
2639     MemOp d_ot = s->dflag;
2640     int size = 1 << d_ot;
2641     int i;
2642 
2643     for (i = 0; i < 8; i++) {
2644         /* ESP is not reloaded */
2645         if (7 - i == R_ESP) {
2646             continue;
2647         }
2648         tcg_gen_addi_tl(s->A0, cpu_regs[R_ESP], i * size);
2649         gen_lea_v_seg(s, s_ot, s->A0, R_SS, -1);
2650         gen_op_ld_v(s, d_ot, s->T0, s->A0);
2651         gen_op_mov_reg_v(s, d_ot, 7 - i, s->T0);
2652     }
2653 
2654     gen_stack_update(s, 8 * size);
2655 }
2656 
2657 static void gen_enter(DisasContext *s, int esp_addend, int level)
2658 {
2659     MemOp d_ot = mo_pushpop(s, s->dflag);
2660     MemOp a_ot = CODE64(s) ? MO_64 : SS32(s) ? MO_32 : MO_16;
2661     int size = 1 << d_ot;
2662 
2663     /* Push BP; compute FrameTemp into T1.  */
2664     tcg_gen_subi_tl(s->T1, cpu_regs[R_ESP], size);
2665     gen_lea_v_seg(s, a_ot, s->T1, R_SS, -1);
2666     gen_op_st_v(s, d_ot, cpu_regs[R_EBP], s->A0);
2667 
2668     level &= 31;
2669     if (level != 0) {
2670         int i;
2671 
2672         /* Copy level-1 pointers from the previous frame.  */
2673         for (i = 1; i < level; ++i) {
2674             tcg_gen_subi_tl(s->A0, cpu_regs[R_EBP], size * i);
2675             gen_lea_v_seg(s, a_ot, s->A0, R_SS, -1);
2676             gen_op_ld_v(s, d_ot, s->tmp0, s->A0);
2677 
2678             tcg_gen_subi_tl(s->A0, s->T1, size * i);
2679             gen_lea_v_seg(s, a_ot, s->A0, R_SS, -1);
2680             gen_op_st_v(s, d_ot, s->tmp0, s->A0);
2681         }
2682 
2683         /* Push the current FrameTemp as the last level.  */
2684         tcg_gen_subi_tl(s->A0, s->T1, size * level);
2685         gen_lea_v_seg(s, a_ot, s->A0, R_SS, -1);
2686         gen_op_st_v(s, d_ot, s->T1, s->A0);
2687     }
2688 
2689     /* Copy the FrameTemp value to EBP.  */
2690     gen_op_mov_reg_v(s, a_ot, R_EBP, s->T1);
2691 
2692     /* Compute the final value of ESP.  */
2693     tcg_gen_subi_tl(s->T1, s->T1, esp_addend + size * level);
2694     gen_op_mov_reg_v(s, a_ot, R_ESP, s->T1);
2695 }
2696 
2697 static void gen_leave(DisasContext *s)
2698 {
2699     MemOp d_ot = mo_pushpop(s, s->dflag);
2700     MemOp a_ot = mo_stacksize(s);
2701 
2702     gen_lea_v_seg(s, a_ot, cpu_regs[R_EBP], R_SS, -1);
2703     gen_op_ld_v(s, d_ot, s->T0, s->A0);
2704 
2705     tcg_gen_addi_tl(s->T1, cpu_regs[R_EBP], 1 << d_ot);
2706 
2707     gen_op_mov_reg_v(s, d_ot, R_EBP, s->T0);
2708     gen_op_mov_reg_v(s, a_ot, R_ESP, s->T1);
2709 }
2710 
2711 /* Similarly, except that the assumption here is that we don't decode
2712    the instruction at all -- either a missing opcode, an unimplemented
2713    feature, or just a bogus instruction stream.  */
2714 static void gen_unknown_opcode(CPUX86State *env, DisasContext *s)
2715 {
2716     gen_illegal_opcode(s);
2717 
2718     if (qemu_loglevel_mask(LOG_UNIMP)) {
2719         FILE *logfile = qemu_log_trylock();
2720         if (logfile) {
2721             target_ulong pc = s->base.pc_next, end = s->pc;
2722 
2723             fprintf(logfile, "ILLOPC: " TARGET_FMT_lx ":", pc);
2724             for (; pc < end; ++pc) {
2725                 fprintf(logfile, " %02x", cpu_ldub_code(env, pc));
2726             }
2727             fprintf(logfile, "\n");
2728             qemu_log_unlock(logfile);
2729         }
2730     }
2731 }
2732 
2733 /* an interrupt is different from an exception because of the
2734    privilege checks */
2735 static void gen_interrupt(DisasContext *s, int intno)
2736 {
2737     gen_update_cc_op(s);
2738     gen_update_eip_cur(s);
2739     gen_helper_raise_interrupt(cpu_env, tcg_constant_i32(intno),
2740                                cur_insn_len_i32(s));
2741     s->base.is_jmp = DISAS_NORETURN;
2742 }
2743 
2744 static void gen_set_hflag(DisasContext *s, uint32_t mask)
2745 {
2746     if ((s->flags & mask) == 0) {
2747         TCGv_i32 t = tcg_temp_new_i32();
2748         tcg_gen_ld_i32(t, cpu_env, offsetof(CPUX86State, hflags));
2749         tcg_gen_ori_i32(t, t, mask);
2750         tcg_gen_st_i32(t, cpu_env, offsetof(CPUX86State, hflags));
2751         tcg_temp_free_i32(t);
2752         s->flags |= mask;
2753     }
2754 }
2755 
2756 static void gen_reset_hflag(DisasContext *s, uint32_t mask)
2757 {
2758     if (s->flags & mask) {
2759         TCGv_i32 t = tcg_temp_new_i32();
2760         tcg_gen_ld_i32(t, cpu_env, offsetof(CPUX86State, hflags));
2761         tcg_gen_andi_i32(t, t, ~mask);
2762         tcg_gen_st_i32(t, cpu_env, offsetof(CPUX86State, hflags));
2763         tcg_temp_free_i32(t);
2764         s->flags &= ~mask;
2765     }
2766 }
2767 
2768 static void gen_set_eflags(DisasContext *s, target_ulong mask)
2769 {
2770     TCGv t = tcg_temp_new();
2771 
2772     tcg_gen_ld_tl(t, cpu_env, offsetof(CPUX86State, eflags));
2773     tcg_gen_ori_tl(t, t, mask);
2774     tcg_gen_st_tl(t, cpu_env, offsetof(CPUX86State, eflags));
2775     tcg_temp_free(t);
2776 }
2777 
2778 static void gen_reset_eflags(DisasContext *s, target_ulong mask)
2779 {
2780     TCGv t = tcg_temp_new();
2781 
2782     tcg_gen_ld_tl(t, cpu_env, offsetof(CPUX86State, eflags));
2783     tcg_gen_andi_tl(t, t, ~mask);
2784     tcg_gen_st_tl(t, cpu_env, offsetof(CPUX86State, eflags));
2785     tcg_temp_free(t);
2786 }
2787 
2788 /* Clear BND registers during legacy branches.  */
2789 static void gen_bnd_jmp(DisasContext *s)
2790 {
2791     /* Clear the registers only if BND prefix is missing, MPX is enabled,
2792        and if the BNDREGs are known to be in use (non-zero) already.
2793        The helper itself will check BNDPRESERVE at runtime.  */
2794     if ((s->prefix & PREFIX_REPNZ) == 0
2795         && (s->flags & HF_MPX_EN_MASK) != 0
2796         && (s->flags & HF_MPX_IU_MASK) != 0) {
2797         gen_helper_bnd_jmp(cpu_env);
2798     }
2799 }
2800 
2801 /* Generate an end of block. Trace exception is also generated if needed.
2802    If INHIBIT, set HF_INHIBIT_IRQ_MASK if it isn't already set.
2803    If RECHECK_TF, emit a rechecking helper for #DB, ignoring the state of
2804    S->TF.  This is used by the syscall/sysret insns.  */
2805 static void
2806 do_gen_eob_worker(DisasContext *s, bool inhibit, bool recheck_tf, bool jr)
2807 {
2808     gen_update_cc_op(s);
2809 
2810     /* If several instructions disable interrupts, only the first does it.  */
2811     if (inhibit && !(s->flags & HF_INHIBIT_IRQ_MASK)) {
2812         gen_set_hflag(s, HF_INHIBIT_IRQ_MASK);
2813     } else {
2814         gen_reset_hflag(s, HF_INHIBIT_IRQ_MASK);
2815     }
2816 
2817     if (s->base.tb->flags & HF_RF_MASK) {
2818         gen_reset_eflags(s, RF_MASK);
2819     }
2820     if (recheck_tf) {
2821         gen_helper_rechecking_single_step(cpu_env);
2822         tcg_gen_exit_tb(NULL, 0);
2823     } else if (s->flags & HF_TF_MASK) {
2824         gen_helper_single_step(cpu_env);
2825     } else if (jr) {
2826         tcg_gen_lookup_and_goto_ptr();
2827     } else {
2828         tcg_gen_exit_tb(NULL, 0);
2829     }
2830     s->base.is_jmp = DISAS_NORETURN;
2831 }
2832 
2833 static inline void
2834 gen_eob_worker(DisasContext *s, bool inhibit, bool recheck_tf)
2835 {
2836     do_gen_eob_worker(s, inhibit, recheck_tf, false);
2837 }
2838 
2839 /* End of block.
2840    If INHIBIT, set HF_INHIBIT_IRQ_MASK if it isn't already set.  */
2841 static void gen_eob_inhibit_irq(DisasContext *s, bool inhibit)
2842 {
2843     gen_eob_worker(s, inhibit, false);
2844 }
2845 
2846 /* End of block, resetting the inhibit irq flag.  */
2847 static void gen_eob(DisasContext *s)
2848 {
2849     gen_eob_worker(s, false, false);
2850 }
2851 
2852 /* Jump to register */
2853 static void gen_jr(DisasContext *s)
2854 {
2855     do_gen_eob_worker(s, false, false, true);
2856 }
2857 
2858 /* Jump to eip+diff, truncating the result to OT. */
2859 static void gen_jmp_rel(DisasContext *s, MemOp ot, int diff, int tb_num)
2860 {
2861     bool use_goto_tb = s->jmp_opt;
2862     target_ulong mask = -1;
2863     target_ulong new_pc = s->pc + diff;
2864     target_ulong new_eip = new_pc - s->cs_base;
2865 
2866     /* In 64-bit mode, operand size is fixed at 64 bits. */
2867     if (!CODE64(s)) {
2868         if (ot == MO_16) {
2869             mask = 0xffff;
2870             if (TARGET_TB_PCREL && CODE32(s)) {
2871                 use_goto_tb = false;
2872             }
2873         } else {
2874             mask = 0xffffffff;
2875         }
2876     }
2877     new_eip &= mask;
2878 
2879     gen_update_cc_op(s);
2880     set_cc_op(s, CC_OP_DYNAMIC);
2881 
2882     if (TARGET_TB_PCREL) {
2883         tcg_gen_addi_tl(cpu_eip, cpu_eip, new_pc - s->pc_save);
2884         /*
2885          * If we can prove the branch does not leave the page and we have
2886          * no extra masking to apply (data16 branch in code32, see above),
2887          * then we have also proven that the addition does not wrap.
2888          */
2889         if (!use_goto_tb || !is_same_page(&s->base, new_pc)) {
2890             tcg_gen_andi_tl(cpu_eip, cpu_eip, mask);
2891             use_goto_tb = false;
2892         }
2893     }
2894 
2895     if (use_goto_tb &&
2896         translator_use_goto_tb(&s->base, new_eip + s->cs_base)) {
2897         /* jump to same page: we can use a direct jump */
2898         tcg_gen_goto_tb(tb_num);
2899         if (!TARGET_TB_PCREL) {
2900             tcg_gen_movi_tl(cpu_eip, new_eip);
2901         }
2902         tcg_gen_exit_tb(s->base.tb, tb_num);
2903         s->base.is_jmp = DISAS_NORETURN;
2904     } else {
2905         if (!TARGET_TB_PCREL) {
2906             tcg_gen_movi_tl(cpu_eip, new_eip);
2907         }
2908         if (s->jmp_opt) {
2909             gen_jr(s);   /* jump to another page */
2910         } else {
2911             gen_eob(s);  /* exit to main loop */
2912         }
2913     }
2914 }
2915 
2916 /* Jump to eip+diff, truncating to the current code size. */
2917 static void gen_jmp_rel_csize(DisasContext *s, int diff, int tb_num)
2918 {
2919     /* CODE64 ignores the OT argument, so we need not consider it. */
2920     gen_jmp_rel(s, CODE32(s) ? MO_32 : MO_16, diff, tb_num);
2921 }
2922 
2923 static inline void gen_ldq_env_A0(DisasContext *s, int offset)
2924 {
2925     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
2926     tcg_gen_st_i64(s->tmp1_i64, cpu_env, offset);
2927 }
2928 
2929 static inline void gen_stq_env_A0(DisasContext *s, int offset)
2930 {
2931     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, offset);
2932     tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, s->mem_index, MO_LEUQ);
2933 }
2934 
2935 static inline void gen_ldo_env_A0(DisasContext *s, int offset, bool align)
2936 {
2937     int mem_index = s->mem_index;
2938     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, mem_index,
2939                         MO_LEUQ | (align ? MO_ALIGN_16 : 0));
2940     tcg_gen_st_i64(s->tmp1_i64, cpu_env, offset + offsetof(XMMReg, XMM_Q(0)));
2941     tcg_gen_addi_tl(s->tmp0, s->A0, 8);
2942     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2943     tcg_gen_st_i64(s->tmp1_i64, cpu_env, offset + offsetof(XMMReg, XMM_Q(1)));
2944 }
2945 
2946 static inline void gen_sto_env_A0(DisasContext *s, int offset, bool align)
2947 {
2948     int mem_index = s->mem_index;
2949     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, offset + offsetof(XMMReg, XMM_Q(0)));
2950     tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, mem_index,
2951                         MO_LEUQ | (align ? MO_ALIGN_16 : 0));
2952     tcg_gen_addi_tl(s->tmp0, s->A0, 8);
2953     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, offset + offsetof(XMMReg, XMM_Q(1)));
2954     tcg_gen_qemu_st_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2955 }
2956 
2957 static void gen_ldy_env_A0(DisasContext *s, int offset, bool align)
2958 {
2959     int mem_index = s->mem_index;
2960     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0, mem_index,
2961                         MO_LEUQ | (align ? MO_ALIGN_32 : 0));
2962     tcg_gen_st_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(0)));
2963     tcg_gen_addi_tl(s->tmp0, s->A0, 8);
2964     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2965     tcg_gen_st_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(1)));
2966 
2967     tcg_gen_addi_tl(s->tmp0, s->A0, 16);
2968     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2969     tcg_gen_st_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(2)));
2970     tcg_gen_addi_tl(s->tmp0, s->A0, 24);
2971     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2972     tcg_gen_st_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(3)));
2973 }
2974 
2975 static void gen_sty_env_A0(DisasContext *s, int offset, bool align)
2976 {
2977     int mem_index = s->mem_index;
2978     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(0)));
2979     tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0, mem_index,
2980                         MO_LEUQ | (align ? MO_ALIGN_32 : 0));
2981     tcg_gen_addi_tl(s->tmp0, s->A0, 8);
2982     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(1)));
2983     tcg_gen_qemu_st_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2984     tcg_gen_addi_tl(s->tmp0, s->A0, 16);
2985     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(2)));
2986     tcg_gen_qemu_st_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2987     tcg_gen_addi_tl(s->tmp0, s->A0, 24);
2988     tcg_gen_ld_i64(s->tmp1_i64, cpu_env, offset + offsetof(YMMReg, YMM_Q(3)));
2989     tcg_gen_qemu_st_i64(s->tmp1_i64, s->tmp0, mem_index, MO_LEUQ);
2990 }
2991 
2992 #include "decode-new.h"
2993 #include "emit.c.inc"
2994 #include "decode-new.c.inc"
2995 
2996 static void gen_cmpxchg8b(DisasContext *s, CPUX86State *env, int modrm)
2997 {
2998     TCGv_i64 cmp, val, old;
2999     TCGv Z;
3000 
3001     gen_lea_modrm(env, s, modrm);
3002 
3003     cmp = tcg_temp_new_i64();
3004     val = tcg_temp_new_i64();
3005     old = tcg_temp_new_i64();
3006 
3007     /* Construct the comparison values from the register pair. */
3008     tcg_gen_concat_tl_i64(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
3009     tcg_gen_concat_tl_i64(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
3010 
3011     /* Only require atomic with LOCK; non-parallel handled in generator. */
3012     if (s->prefix & PREFIX_LOCK) {
3013         tcg_gen_atomic_cmpxchg_i64(old, s->A0, cmp, val, s->mem_index, MO_TEUQ);
3014     } else {
3015         tcg_gen_nonatomic_cmpxchg_i64(old, s->A0, cmp, val,
3016                                       s->mem_index, MO_TEUQ);
3017     }
3018     tcg_temp_free_i64(val);
3019 
3020     /* Set tmp0 to match the required value of Z. */
3021     tcg_gen_setcond_i64(TCG_COND_EQ, cmp, old, cmp);
3022     Z = tcg_temp_new();
3023     tcg_gen_trunc_i64_tl(Z, cmp);
3024     tcg_temp_free_i64(cmp);
3025 
3026     /*
3027      * Extract the result values for the register pair.
3028      * For 32-bit, we may do this unconditionally, because on success (Z=1),
3029      * the old value matches the previous value in EDX:EAX.  For x86_64,
3030      * the store must be conditional, because we must leave the source
3031      * registers unchanged on success, and zero-extend the writeback
3032      * on failure (Z=0).
3033      */
3034     if (TARGET_LONG_BITS == 32) {
3035         tcg_gen_extr_i64_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], old);
3036     } else {
3037         TCGv zero = tcg_constant_tl(0);
3038 
3039         tcg_gen_extr_i64_tl(s->T0, s->T1, old);
3040         tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EAX], Z, zero,
3041                            s->T0, cpu_regs[R_EAX]);
3042         tcg_gen_movcond_tl(TCG_COND_EQ, cpu_regs[R_EDX], Z, zero,
3043                            s->T1, cpu_regs[R_EDX]);
3044     }
3045     tcg_temp_free_i64(old);
3046 
3047     /* Update Z. */
3048     gen_compute_eflags(s);
3049     tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, Z, ctz32(CC_Z), 1);
3050     tcg_temp_free(Z);
3051 }
3052 
3053 #ifdef TARGET_X86_64
3054 static void gen_cmpxchg16b(DisasContext *s, CPUX86State *env, int modrm)
3055 {
3056     MemOp mop = MO_TE | MO_128 | MO_ALIGN;
3057     TCGv_i64 t0, t1;
3058     TCGv_i128 cmp, val;
3059 
3060     gen_lea_modrm(env, s, modrm);
3061 
3062     cmp = tcg_temp_new_i128();
3063     val = tcg_temp_new_i128();
3064     tcg_gen_concat_i64_i128(cmp, cpu_regs[R_EAX], cpu_regs[R_EDX]);
3065     tcg_gen_concat_i64_i128(val, cpu_regs[R_EBX], cpu_regs[R_ECX]);
3066 
3067     /* Only require atomic with LOCK; non-parallel handled in generator. */
3068     if (s->prefix & PREFIX_LOCK) {
3069         tcg_gen_atomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
3070     } else {
3071         tcg_gen_nonatomic_cmpxchg_i128(val, s->A0, cmp, val, s->mem_index, mop);
3072     }
3073 
3074     tcg_gen_extr_i128_i64(s->T0, s->T1, val);
3075     tcg_temp_free_i128(cmp);
3076     tcg_temp_free_i128(val);
3077 
3078     /* Determine success after the fact. */
3079     t0 = tcg_temp_new_i64();
3080     t1 = tcg_temp_new_i64();
3081     tcg_gen_xor_i64(t0, s->T0, cpu_regs[R_EAX]);
3082     tcg_gen_xor_i64(t1, s->T1, cpu_regs[R_EDX]);
3083     tcg_gen_or_i64(t0, t0, t1);
3084     tcg_temp_free_i64(t1);
3085 
3086     /* Update Z. */
3087     gen_compute_eflags(s);
3088     tcg_gen_setcondi_i64(TCG_COND_EQ, t0, t0, 0);
3089     tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, t0, ctz32(CC_Z), 1);
3090     tcg_temp_free_i64(t0);
3091 
3092     /*
3093      * Extract the result values for the register pair.  We may do this
3094      * unconditionally, because on success (Z=1), the old value matches
3095      * the previous value in RDX:RAX.
3096      */
3097     tcg_gen_mov_i64(cpu_regs[R_EAX], s->T0);
3098     tcg_gen_mov_i64(cpu_regs[R_EDX], s->T1);
3099 }
3100 #endif
3101 
3102 /* convert one instruction. s->base.is_jmp is set if the translation must
3103    be stopped. Return the next pc value */
3104 static bool disas_insn(DisasContext *s, CPUState *cpu)
3105 {
3106     CPUX86State *env = cpu->env_ptr;
3107     int b, prefixes;
3108     int shift;
3109     MemOp ot, aflag, dflag;
3110     int modrm, reg, rm, mod, op, opreg, val;
3111     bool orig_cc_op_dirty = s->cc_op_dirty;
3112     CCOp orig_cc_op = s->cc_op;
3113     target_ulong orig_pc_save = s->pc_save;
3114 
3115     s->pc = s->base.pc_next;
3116     s->override = -1;
3117 #ifdef TARGET_X86_64
3118     s->rex_r = 0;
3119     s->rex_x = 0;
3120     s->rex_b = 0;
3121 #endif
3122     s->rip_offset = 0; /* for relative ip address */
3123     s->vex_l = 0;
3124     s->vex_v = 0;
3125     s->vex_w = false;
3126     switch (sigsetjmp(s->jmpbuf, 0)) {
3127     case 0:
3128         break;
3129     case 1:
3130         gen_exception_gpf(s);
3131         return true;
3132     case 2:
3133         /* Restore state that may affect the next instruction. */
3134         s->pc = s->base.pc_next;
3135         /*
3136          * TODO: These save/restore can be removed after the table-based
3137          * decoder is complete; we will be decoding the insn completely
3138          * before any code generation that might affect these variables.
3139          */
3140         s->cc_op_dirty = orig_cc_op_dirty;
3141         s->cc_op = orig_cc_op;
3142         s->pc_save = orig_pc_save;
3143         /* END TODO */
3144         s->base.num_insns--;
3145         tcg_remove_ops_after(s->prev_insn_end);
3146         s->base.is_jmp = DISAS_TOO_MANY;
3147         return false;
3148     default:
3149         g_assert_not_reached();
3150     }
3151 
3152     prefixes = 0;
3153 
3154  next_byte:
3155     s->prefix = prefixes;
3156     b = x86_ldub_code(env, s);
3157     /* Collect prefixes.  */
3158     switch (b) {
3159     default:
3160         break;
3161     case 0x0f:
3162         b = x86_ldub_code(env, s) + 0x100;
3163         break;
3164     case 0xf3:
3165         prefixes |= PREFIX_REPZ;
3166         prefixes &= ~PREFIX_REPNZ;
3167         goto next_byte;
3168     case 0xf2:
3169         prefixes |= PREFIX_REPNZ;
3170         prefixes &= ~PREFIX_REPZ;
3171         goto next_byte;
3172     case 0xf0:
3173         prefixes |= PREFIX_LOCK;
3174         goto next_byte;
3175     case 0x2e:
3176         s->override = R_CS;
3177         goto next_byte;
3178     case 0x36:
3179         s->override = R_SS;
3180         goto next_byte;
3181     case 0x3e:
3182         s->override = R_DS;
3183         goto next_byte;
3184     case 0x26:
3185         s->override = R_ES;
3186         goto next_byte;
3187     case 0x64:
3188         s->override = R_FS;
3189         goto next_byte;
3190     case 0x65:
3191         s->override = R_GS;
3192         goto next_byte;
3193     case 0x66:
3194         prefixes |= PREFIX_DATA;
3195         goto next_byte;
3196     case 0x67:
3197         prefixes |= PREFIX_ADR;
3198         goto next_byte;
3199 #ifdef TARGET_X86_64
3200     case 0x40 ... 0x4f:
3201         if (CODE64(s)) {
3202             /* REX prefix */
3203             prefixes |= PREFIX_REX;
3204             s->vex_w = (b >> 3) & 1;
3205             s->rex_r = (b & 0x4) << 1;
3206             s->rex_x = (b & 0x2) << 2;
3207             s->rex_b = (b & 0x1) << 3;
3208             goto next_byte;
3209         }
3210         break;
3211 #endif
3212     case 0xc5: /* 2-byte VEX */
3213     case 0xc4: /* 3-byte VEX */
3214         if (CODE32(s) && !VM86(s)) {
3215             int vex2 = x86_ldub_code(env, s);
3216             s->pc--; /* rewind the advance_pc() x86_ldub_code() did */
3217 
3218             if (!CODE64(s) && (vex2 & 0xc0) != 0xc0) {
3219                 /* 4.1.4.6: In 32-bit mode, bits [7:6] must be 11b,
3220                    otherwise the instruction is LES or LDS.  */
3221                 break;
3222             }
3223             disas_insn_new(s, cpu, b);
3224             return s->pc;
3225         }
3226         break;
3227     }
3228 
3229     /* Post-process prefixes.  */
3230     if (CODE64(s)) {
3231         /* In 64-bit mode, the default data size is 32-bit.  Select 64-bit
3232            data with rex_w, and 16-bit data with 0x66; rex_w takes precedence
3233            over 0x66 if both are present.  */
3234         dflag = (REX_W(s) ? MO_64 : prefixes & PREFIX_DATA ? MO_16 : MO_32);
3235         /* In 64-bit mode, 0x67 selects 32-bit addressing.  */
3236         aflag = (prefixes & PREFIX_ADR ? MO_32 : MO_64);
3237     } else {
3238         /* In 16/32-bit mode, 0x66 selects the opposite data size.  */
3239         if (CODE32(s) ^ ((prefixes & PREFIX_DATA) != 0)) {
3240             dflag = MO_32;
3241         } else {
3242             dflag = MO_16;
3243         }
3244         /* In 16/32-bit mode, 0x67 selects the opposite addressing.  */
3245         if (CODE32(s) ^ ((prefixes & PREFIX_ADR) != 0)) {
3246             aflag = MO_32;
3247         }  else {
3248             aflag = MO_16;
3249         }
3250     }
3251 
3252     s->prefix = prefixes;
3253     s->aflag = aflag;
3254     s->dflag = dflag;
3255 
3256     /* now check op code */
3257     switch (b) {
3258         /**************************/
3259         /* arith & logic */
3260     case 0x00 ... 0x05:
3261     case 0x08 ... 0x0d:
3262     case 0x10 ... 0x15:
3263     case 0x18 ... 0x1d:
3264     case 0x20 ... 0x25:
3265     case 0x28 ... 0x2d:
3266     case 0x30 ... 0x35:
3267     case 0x38 ... 0x3d:
3268         {
3269             int op, f, val;
3270             op = (b >> 3) & 7;
3271             f = (b >> 1) & 3;
3272 
3273             ot = mo_b_d(b, dflag);
3274 
3275             switch(f) {
3276             case 0: /* OP Ev, Gv */
3277                 modrm = x86_ldub_code(env, s);
3278                 reg = ((modrm >> 3) & 7) | REX_R(s);
3279                 mod = (modrm >> 6) & 3;
3280                 rm = (modrm & 7) | REX_B(s);
3281                 if (mod != 3) {
3282                     gen_lea_modrm(env, s, modrm);
3283                     opreg = OR_TMP0;
3284                 } else if (op == OP_XORL && rm == reg) {
3285                 xor_zero:
3286                     /* xor reg, reg optimisation */
3287                     set_cc_op(s, CC_OP_CLR);
3288                     tcg_gen_movi_tl(s->T0, 0);
3289                     gen_op_mov_reg_v(s, ot, reg, s->T0);
3290                     break;
3291                 } else {
3292                     opreg = rm;
3293                 }
3294                 gen_op_mov_v_reg(s, ot, s->T1, reg);
3295                 gen_op(s, op, ot, opreg);
3296                 break;
3297             case 1: /* OP Gv, Ev */
3298                 modrm = x86_ldub_code(env, s);
3299                 mod = (modrm >> 6) & 3;
3300                 reg = ((modrm >> 3) & 7) | REX_R(s);
3301                 rm = (modrm & 7) | REX_B(s);
3302                 if (mod != 3) {
3303                     gen_lea_modrm(env, s, modrm);
3304                     gen_op_ld_v(s, ot, s->T1, s->A0);
3305                 } else if (op == OP_XORL && rm == reg) {
3306                     goto xor_zero;
3307                 } else {
3308                     gen_op_mov_v_reg(s, ot, s->T1, rm);
3309                 }
3310                 gen_op(s, op, ot, reg);
3311                 break;
3312             case 2: /* OP A, Iv */
3313                 val = insn_get(env, s, ot);
3314                 tcg_gen_movi_tl(s->T1, val);
3315                 gen_op(s, op, ot, OR_EAX);
3316                 break;
3317             }
3318         }
3319         break;
3320 
3321     case 0x82:
3322         if (CODE64(s))
3323             goto illegal_op;
3324         /* fall through */
3325     case 0x80: /* GRP1 */
3326     case 0x81:
3327     case 0x83:
3328         {
3329             int val;
3330 
3331             ot = mo_b_d(b, dflag);
3332 
3333             modrm = x86_ldub_code(env, s);
3334             mod = (modrm >> 6) & 3;
3335             rm = (modrm & 7) | REX_B(s);
3336             op = (modrm >> 3) & 7;
3337 
3338             if (mod != 3) {
3339                 if (b == 0x83)
3340                     s->rip_offset = 1;
3341                 else
3342                     s->rip_offset = insn_const_size(ot);
3343                 gen_lea_modrm(env, s, modrm);
3344                 opreg = OR_TMP0;
3345             } else {
3346                 opreg = rm;
3347             }
3348 
3349             switch(b) {
3350             default:
3351             case 0x80:
3352             case 0x81:
3353             case 0x82:
3354                 val = insn_get(env, s, ot);
3355                 break;
3356             case 0x83:
3357                 val = (int8_t)insn_get(env, s, MO_8);
3358                 break;
3359             }
3360             tcg_gen_movi_tl(s->T1, val);
3361             gen_op(s, op, ot, opreg);
3362         }
3363         break;
3364 
3365         /**************************/
3366         /* inc, dec, and other misc arith */
3367     case 0x40 ... 0x47: /* inc Gv */
3368         ot = dflag;
3369         gen_inc(s, ot, OR_EAX + (b & 7), 1);
3370         break;
3371     case 0x48 ... 0x4f: /* dec Gv */
3372         ot = dflag;
3373         gen_inc(s, ot, OR_EAX + (b & 7), -1);
3374         break;
3375     case 0xf6: /* GRP3 */
3376     case 0xf7:
3377         ot = mo_b_d(b, dflag);
3378 
3379         modrm = x86_ldub_code(env, s);
3380         mod = (modrm >> 6) & 3;
3381         rm = (modrm & 7) | REX_B(s);
3382         op = (modrm >> 3) & 7;
3383         if (mod != 3) {
3384             if (op == 0) {
3385                 s->rip_offset = insn_const_size(ot);
3386             }
3387             gen_lea_modrm(env, s, modrm);
3388             /* For those below that handle locked memory, don't load here.  */
3389             if (!(s->prefix & PREFIX_LOCK)
3390                 || op != 2) {
3391                 gen_op_ld_v(s, ot, s->T0, s->A0);
3392             }
3393         } else {
3394             gen_op_mov_v_reg(s, ot, s->T0, rm);
3395         }
3396 
3397         switch(op) {
3398         case 0: /* test */
3399             val = insn_get(env, s, ot);
3400             tcg_gen_movi_tl(s->T1, val);
3401             gen_op_testl_T0_T1_cc(s);
3402             set_cc_op(s, CC_OP_LOGICB + ot);
3403             break;
3404         case 2: /* not */
3405             if (s->prefix & PREFIX_LOCK) {
3406                 if (mod == 3) {
3407                     goto illegal_op;
3408                 }
3409                 tcg_gen_movi_tl(s->T0, ~0);
3410                 tcg_gen_atomic_xor_fetch_tl(s->T0, s->A0, s->T0,
3411                                             s->mem_index, ot | MO_LE);
3412             } else {
3413                 tcg_gen_not_tl(s->T0, s->T0);
3414                 if (mod != 3) {
3415                     gen_op_st_v(s, ot, s->T0, s->A0);
3416                 } else {
3417                     gen_op_mov_reg_v(s, ot, rm, s->T0);
3418                 }
3419             }
3420             break;
3421         case 3: /* neg */
3422             if (s->prefix & PREFIX_LOCK) {
3423                 TCGLabel *label1;
3424                 TCGv a0, t0, t1, t2;
3425 
3426                 if (mod == 3) {
3427                     goto illegal_op;
3428                 }
3429                 a0 = tcg_temp_local_new();
3430                 t0 = tcg_temp_local_new();
3431                 label1 = gen_new_label();
3432 
3433                 tcg_gen_mov_tl(a0, s->A0);
3434                 tcg_gen_mov_tl(t0, s->T0);
3435 
3436                 gen_set_label(label1);
3437                 t1 = tcg_temp_new();
3438                 t2 = tcg_temp_new();
3439                 tcg_gen_mov_tl(t2, t0);
3440                 tcg_gen_neg_tl(t1, t0);
3441                 tcg_gen_atomic_cmpxchg_tl(t0, a0, t0, t1,
3442                                           s->mem_index, ot | MO_LE);
3443                 tcg_temp_free(t1);
3444                 tcg_gen_brcond_tl(TCG_COND_NE, t0, t2, label1);
3445 
3446                 tcg_temp_free(t2);
3447                 tcg_temp_free(a0);
3448                 tcg_gen_neg_tl(s->T0, t0);
3449                 tcg_temp_free(t0);
3450             } else {
3451                 tcg_gen_neg_tl(s->T0, s->T0);
3452                 if (mod != 3) {
3453                     gen_op_st_v(s, ot, s->T0, s->A0);
3454                 } else {
3455                     gen_op_mov_reg_v(s, ot, rm, s->T0);
3456                 }
3457             }
3458             gen_op_update_neg_cc(s);
3459             set_cc_op(s, CC_OP_SUBB + ot);
3460             break;
3461         case 4: /* mul */
3462             switch(ot) {
3463             case MO_8:
3464                 gen_op_mov_v_reg(s, MO_8, s->T1, R_EAX);
3465                 tcg_gen_ext8u_tl(s->T0, s->T0);
3466                 tcg_gen_ext8u_tl(s->T1, s->T1);
3467                 /* XXX: use 32 bit mul which could be faster */
3468                 tcg_gen_mul_tl(s->T0, s->T0, s->T1);
3469                 gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
3470                 tcg_gen_mov_tl(cpu_cc_dst, s->T0);
3471                 tcg_gen_andi_tl(cpu_cc_src, s->T0, 0xff00);
3472                 set_cc_op(s, CC_OP_MULB);
3473                 break;
3474             case MO_16:
3475                 gen_op_mov_v_reg(s, MO_16, s->T1, R_EAX);
3476                 tcg_gen_ext16u_tl(s->T0, s->T0);
3477                 tcg_gen_ext16u_tl(s->T1, s->T1);
3478                 /* XXX: use 32 bit mul which could be faster */
3479                 tcg_gen_mul_tl(s->T0, s->T0, s->T1);
3480                 gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
3481                 tcg_gen_mov_tl(cpu_cc_dst, s->T0);
3482                 tcg_gen_shri_tl(s->T0, s->T0, 16);
3483                 gen_op_mov_reg_v(s, MO_16, R_EDX, s->T0);
3484                 tcg_gen_mov_tl(cpu_cc_src, s->T0);
3485                 set_cc_op(s, CC_OP_MULW);
3486                 break;
3487             default:
3488             case MO_32:
3489                 tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3490                 tcg_gen_trunc_tl_i32(s->tmp3_i32, cpu_regs[R_EAX]);
3491                 tcg_gen_mulu2_i32(s->tmp2_i32, s->tmp3_i32,
3492                                   s->tmp2_i32, s->tmp3_i32);
3493                 tcg_gen_extu_i32_tl(cpu_regs[R_EAX], s->tmp2_i32);
3494                 tcg_gen_extu_i32_tl(cpu_regs[R_EDX], s->tmp3_i32);
3495                 tcg_gen_mov_tl(cpu_cc_dst, cpu_regs[R_EAX]);
3496                 tcg_gen_mov_tl(cpu_cc_src, cpu_regs[R_EDX]);
3497                 set_cc_op(s, CC_OP_MULL);
3498                 break;
3499 #ifdef TARGET_X86_64
3500             case MO_64:
3501                 tcg_gen_mulu2_i64(cpu_regs[R_EAX], cpu_regs[R_EDX],
3502                                   s->T0, cpu_regs[R_EAX]);
3503                 tcg_gen_mov_tl(cpu_cc_dst, cpu_regs[R_EAX]);
3504                 tcg_gen_mov_tl(cpu_cc_src, cpu_regs[R_EDX]);
3505                 set_cc_op(s, CC_OP_MULQ);
3506                 break;
3507 #endif
3508             }
3509             break;
3510         case 5: /* imul */
3511             switch(ot) {
3512             case MO_8:
3513                 gen_op_mov_v_reg(s, MO_8, s->T1, R_EAX);
3514                 tcg_gen_ext8s_tl(s->T0, s->T0);
3515                 tcg_gen_ext8s_tl(s->T1, s->T1);
3516                 /* XXX: use 32 bit mul which could be faster */
3517                 tcg_gen_mul_tl(s->T0, s->T0, s->T1);
3518                 gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
3519                 tcg_gen_mov_tl(cpu_cc_dst, s->T0);
3520                 tcg_gen_ext8s_tl(s->tmp0, s->T0);
3521                 tcg_gen_sub_tl(cpu_cc_src, s->T0, s->tmp0);
3522                 set_cc_op(s, CC_OP_MULB);
3523                 break;
3524             case MO_16:
3525                 gen_op_mov_v_reg(s, MO_16, s->T1, R_EAX);
3526                 tcg_gen_ext16s_tl(s->T0, s->T0);
3527                 tcg_gen_ext16s_tl(s->T1, s->T1);
3528                 /* XXX: use 32 bit mul which could be faster */
3529                 tcg_gen_mul_tl(s->T0, s->T0, s->T1);
3530                 gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
3531                 tcg_gen_mov_tl(cpu_cc_dst, s->T0);
3532                 tcg_gen_ext16s_tl(s->tmp0, s->T0);
3533                 tcg_gen_sub_tl(cpu_cc_src, s->T0, s->tmp0);
3534                 tcg_gen_shri_tl(s->T0, s->T0, 16);
3535                 gen_op_mov_reg_v(s, MO_16, R_EDX, s->T0);
3536                 set_cc_op(s, CC_OP_MULW);
3537                 break;
3538             default:
3539             case MO_32:
3540                 tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3541                 tcg_gen_trunc_tl_i32(s->tmp3_i32, cpu_regs[R_EAX]);
3542                 tcg_gen_muls2_i32(s->tmp2_i32, s->tmp3_i32,
3543                                   s->tmp2_i32, s->tmp3_i32);
3544                 tcg_gen_extu_i32_tl(cpu_regs[R_EAX], s->tmp2_i32);
3545                 tcg_gen_extu_i32_tl(cpu_regs[R_EDX], s->tmp3_i32);
3546                 tcg_gen_sari_i32(s->tmp2_i32, s->tmp2_i32, 31);
3547                 tcg_gen_mov_tl(cpu_cc_dst, cpu_regs[R_EAX]);
3548                 tcg_gen_sub_i32(s->tmp2_i32, s->tmp2_i32, s->tmp3_i32);
3549                 tcg_gen_extu_i32_tl(cpu_cc_src, s->tmp2_i32);
3550                 set_cc_op(s, CC_OP_MULL);
3551                 break;
3552 #ifdef TARGET_X86_64
3553             case MO_64:
3554                 tcg_gen_muls2_i64(cpu_regs[R_EAX], cpu_regs[R_EDX],
3555                                   s->T0, cpu_regs[R_EAX]);
3556                 tcg_gen_mov_tl(cpu_cc_dst, cpu_regs[R_EAX]);
3557                 tcg_gen_sari_tl(cpu_cc_src, cpu_regs[R_EAX], 63);
3558                 tcg_gen_sub_tl(cpu_cc_src, cpu_cc_src, cpu_regs[R_EDX]);
3559                 set_cc_op(s, CC_OP_MULQ);
3560                 break;
3561 #endif
3562             }
3563             break;
3564         case 6: /* div */
3565             switch(ot) {
3566             case MO_8:
3567                 gen_helper_divb_AL(cpu_env, s->T0);
3568                 break;
3569             case MO_16:
3570                 gen_helper_divw_AX(cpu_env, s->T0);
3571                 break;
3572             default:
3573             case MO_32:
3574                 gen_helper_divl_EAX(cpu_env, s->T0);
3575                 break;
3576 #ifdef TARGET_X86_64
3577             case MO_64:
3578                 gen_helper_divq_EAX(cpu_env, s->T0);
3579                 break;
3580 #endif
3581             }
3582             break;
3583         case 7: /* idiv */
3584             switch(ot) {
3585             case MO_8:
3586                 gen_helper_idivb_AL(cpu_env, s->T0);
3587                 break;
3588             case MO_16:
3589                 gen_helper_idivw_AX(cpu_env, s->T0);
3590                 break;
3591             default:
3592             case MO_32:
3593                 gen_helper_idivl_EAX(cpu_env, s->T0);
3594                 break;
3595 #ifdef TARGET_X86_64
3596             case MO_64:
3597                 gen_helper_idivq_EAX(cpu_env, s->T0);
3598                 break;
3599 #endif
3600             }
3601             break;
3602         default:
3603             goto unknown_op;
3604         }
3605         break;
3606 
3607     case 0xfe: /* GRP4 */
3608     case 0xff: /* GRP5 */
3609         ot = mo_b_d(b, dflag);
3610 
3611         modrm = x86_ldub_code(env, s);
3612         mod = (modrm >> 6) & 3;
3613         rm = (modrm & 7) | REX_B(s);
3614         op = (modrm >> 3) & 7;
3615         if (op >= 2 && b == 0xfe) {
3616             goto unknown_op;
3617         }
3618         if (CODE64(s)) {
3619             if (op == 2 || op == 4) {
3620                 /* operand size for jumps is 64 bit */
3621                 ot = MO_64;
3622             } else if (op == 3 || op == 5) {
3623                 ot = dflag != MO_16 ? MO_32 + REX_W(s) : MO_16;
3624             } else if (op == 6) {
3625                 /* default push size is 64 bit */
3626                 ot = mo_pushpop(s, dflag);
3627             }
3628         }
3629         if (mod != 3) {
3630             gen_lea_modrm(env, s, modrm);
3631             if (op >= 2 && op != 3 && op != 5)
3632                 gen_op_ld_v(s, ot, s->T0, s->A0);
3633         } else {
3634             gen_op_mov_v_reg(s, ot, s->T0, rm);
3635         }
3636 
3637         switch(op) {
3638         case 0: /* inc Ev */
3639             if (mod != 3)
3640                 opreg = OR_TMP0;
3641             else
3642                 opreg = rm;
3643             gen_inc(s, ot, opreg, 1);
3644             break;
3645         case 1: /* dec Ev */
3646             if (mod != 3)
3647                 opreg = OR_TMP0;
3648             else
3649                 opreg = rm;
3650             gen_inc(s, ot, opreg, -1);
3651             break;
3652         case 2: /* call Ev */
3653             /* XXX: optimize if memory (no 'and' is necessary) */
3654             if (dflag == MO_16) {
3655                 tcg_gen_ext16u_tl(s->T0, s->T0);
3656             }
3657             gen_push_v(s, eip_next_tl(s));
3658             gen_op_jmp_v(s, s->T0);
3659             gen_bnd_jmp(s);
3660             s->base.is_jmp = DISAS_JUMP;
3661             break;
3662         case 3: /* lcall Ev */
3663             if (mod == 3) {
3664                 goto illegal_op;
3665             }
3666             gen_op_ld_v(s, ot, s->T1, s->A0);
3667             gen_add_A0_im(s, 1 << ot);
3668             gen_op_ld_v(s, MO_16, s->T0, s->A0);
3669         do_lcall:
3670             if (PE(s) && !VM86(s)) {
3671                 tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3672                 gen_helper_lcall_protected(cpu_env, s->tmp2_i32, s->T1,
3673                                            tcg_constant_i32(dflag - 1),
3674                                            eip_next_tl(s));
3675             } else {
3676                 tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3677                 tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
3678                 gen_helper_lcall_real(cpu_env, s->tmp2_i32, s->tmp3_i32,
3679                                       tcg_constant_i32(dflag - 1),
3680                                       eip_next_i32(s));
3681             }
3682             s->base.is_jmp = DISAS_JUMP;
3683             break;
3684         case 4: /* jmp Ev */
3685             if (dflag == MO_16) {
3686                 tcg_gen_ext16u_tl(s->T0, s->T0);
3687             }
3688             gen_op_jmp_v(s, s->T0);
3689             gen_bnd_jmp(s);
3690             s->base.is_jmp = DISAS_JUMP;
3691             break;
3692         case 5: /* ljmp Ev */
3693             if (mod == 3) {
3694                 goto illegal_op;
3695             }
3696             gen_op_ld_v(s, ot, s->T1, s->A0);
3697             gen_add_A0_im(s, 1 << ot);
3698             gen_op_ld_v(s, MO_16, s->T0, s->A0);
3699         do_ljmp:
3700             if (PE(s) && !VM86(s)) {
3701                 tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3702                 gen_helper_ljmp_protected(cpu_env, s->tmp2_i32, s->T1,
3703                                           eip_next_tl(s));
3704             } else {
3705                 gen_op_movl_seg_T0_vm(s, R_CS);
3706                 gen_op_jmp_v(s, s->T1);
3707             }
3708             s->base.is_jmp = DISAS_JUMP;
3709             break;
3710         case 6: /* push Ev */
3711             gen_push_v(s, s->T0);
3712             break;
3713         default:
3714             goto unknown_op;
3715         }
3716         break;
3717 
3718     case 0x84: /* test Ev, Gv */
3719     case 0x85:
3720         ot = mo_b_d(b, dflag);
3721 
3722         modrm = x86_ldub_code(env, s);
3723         reg = ((modrm >> 3) & 7) | REX_R(s);
3724 
3725         gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 0);
3726         gen_op_mov_v_reg(s, ot, s->T1, reg);
3727         gen_op_testl_T0_T1_cc(s);
3728         set_cc_op(s, CC_OP_LOGICB + ot);
3729         break;
3730 
3731     case 0xa8: /* test eAX, Iv */
3732     case 0xa9:
3733         ot = mo_b_d(b, dflag);
3734         val = insn_get(env, s, ot);
3735 
3736         gen_op_mov_v_reg(s, ot, s->T0, OR_EAX);
3737         tcg_gen_movi_tl(s->T1, val);
3738         gen_op_testl_T0_T1_cc(s);
3739         set_cc_op(s, CC_OP_LOGICB + ot);
3740         break;
3741 
3742     case 0x98: /* CWDE/CBW */
3743         switch (dflag) {
3744 #ifdef TARGET_X86_64
3745         case MO_64:
3746             gen_op_mov_v_reg(s, MO_32, s->T0, R_EAX);
3747             tcg_gen_ext32s_tl(s->T0, s->T0);
3748             gen_op_mov_reg_v(s, MO_64, R_EAX, s->T0);
3749             break;
3750 #endif
3751         case MO_32:
3752             gen_op_mov_v_reg(s, MO_16, s->T0, R_EAX);
3753             tcg_gen_ext16s_tl(s->T0, s->T0);
3754             gen_op_mov_reg_v(s, MO_32, R_EAX, s->T0);
3755             break;
3756         case MO_16:
3757             gen_op_mov_v_reg(s, MO_8, s->T0, R_EAX);
3758             tcg_gen_ext8s_tl(s->T0, s->T0);
3759             gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
3760             break;
3761         default:
3762             tcg_abort();
3763         }
3764         break;
3765     case 0x99: /* CDQ/CWD */
3766         switch (dflag) {
3767 #ifdef TARGET_X86_64
3768         case MO_64:
3769             gen_op_mov_v_reg(s, MO_64, s->T0, R_EAX);
3770             tcg_gen_sari_tl(s->T0, s->T0, 63);
3771             gen_op_mov_reg_v(s, MO_64, R_EDX, s->T0);
3772             break;
3773 #endif
3774         case MO_32:
3775             gen_op_mov_v_reg(s, MO_32, s->T0, R_EAX);
3776             tcg_gen_ext32s_tl(s->T0, s->T0);
3777             tcg_gen_sari_tl(s->T0, s->T0, 31);
3778             gen_op_mov_reg_v(s, MO_32, R_EDX, s->T0);
3779             break;
3780         case MO_16:
3781             gen_op_mov_v_reg(s, MO_16, s->T0, R_EAX);
3782             tcg_gen_ext16s_tl(s->T0, s->T0);
3783             tcg_gen_sari_tl(s->T0, s->T0, 15);
3784             gen_op_mov_reg_v(s, MO_16, R_EDX, s->T0);
3785             break;
3786         default:
3787             tcg_abort();
3788         }
3789         break;
3790     case 0x1af: /* imul Gv, Ev */
3791     case 0x69: /* imul Gv, Ev, I */
3792     case 0x6b:
3793         ot = dflag;
3794         modrm = x86_ldub_code(env, s);
3795         reg = ((modrm >> 3) & 7) | REX_R(s);
3796         if (b == 0x69)
3797             s->rip_offset = insn_const_size(ot);
3798         else if (b == 0x6b)
3799             s->rip_offset = 1;
3800         gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 0);
3801         if (b == 0x69) {
3802             val = insn_get(env, s, ot);
3803             tcg_gen_movi_tl(s->T1, val);
3804         } else if (b == 0x6b) {
3805             val = (int8_t)insn_get(env, s, MO_8);
3806             tcg_gen_movi_tl(s->T1, val);
3807         } else {
3808             gen_op_mov_v_reg(s, ot, s->T1, reg);
3809         }
3810         switch (ot) {
3811 #ifdef TARGET_X86_64
3812         case MO_64:
3813             tcg_gen_muls2_i64(cpu_regs[reg], s->T1, s->T0, s->T1);
3814             tcg_gen_mov_tl(cpu_cc_dst, cpu_regs[reg]);
3815             tcg_gen_sari_tl(cpu_cc_src, cpu_cc_dst, 63);
3816             tcg_gen_sub_tl(cpu_cc_src, cpu_cc_src, s->T1);
3817             break;
3818 #endif
3819         case MO_32:
3820             tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
3821             tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
3822             tcg_gen_muls2_i32(s->tmp2_i32, s->tmp3_i32,
3823                               s->tmp2_i32, s->tmp3_i32);
3824             tcg_gen_extu_i32_tl(cpu_regs[reg], s->tmp2_i32);
3825             tcg_gen_sari_i32(s->tmp2_i32, s->tmp2_i32, 31);
3826             tcg_gen_mov_tl(cpu_cc_dst, cpu_regs[reg]);
3827             tcg_gen_sub_i32(s->tmp2_i32, s->tmp2_i32, s->tmp3_i32);
3828             tcg_gen_extu_i32_tl(cpu_cc_src, s->tmp2_i32);
3829             break;
3830         default:
3831             tcg_gen_ext16s_tl(s->T0, s->T0);
3832             tcg_gen_ext16s_tl(s->T1, s->T1);
3833             /* XXX: use 32 bit mul which could be faster */
3834             tcg_gen_mul_tl(s->T0, s->T0, s->T1);
3835             tcg_gen_mov_tl(cpu_cc_dst, s->T0);
3836             tcg_gen_ext16s_tl(s->tmp0, s->T0);
3837             tcg_gen_sub_tl(cpu_cc_src, s->T0, s->tmp0);
3838             gen_op_mov_reg_v(s, ot, reg, s->T0);
3839             break;
3840         }
3841         set_cc_op(s, CC_OP_MULB + ot);
3842         break;
3843     case 0x1c0:
3844     case 0x1c1: /* xadd Ev, Gv */
3845         ot = mo_b_d(b, dflag);
3846         modrm = x86_ldub_code(env, s);
3847         reg = ((modrm >> 3) & 7) | REX_R(s);
3848         mod = (modrm >> 6) & 3;
3849         gen_op_mov_v_reg(s, ot, s->T0, reg);
3850         if (mod == 3) {
3851             rm = (modrm & 7) | REX_B(s);
3852             gen_op_mov_v_reg(s, ot, s->T1, rm);
3853             tcg_gen_add_tl(s->T0, s->T0, s->T1);
3854             gen_op_mov_reg_v(s, ot, reg, s->T1);
3855             gen_op_mov_reg_v(s, ot, rm, s->T0);
3856         } else {
3857             gen_lea_modrm(env, s, modrm);
3858             if (s->prefix & PREFIX_LOCK) {
3859                 tcg_gen_atomic_fetch_add_tl(s->T1, s->A0, s->T0,
3860                                             s->mem_index, ot | MO_LE);
3861                 tcg_gen_add_tl(s->T0, s->T0, s->T1);
3862             } else {
3863                 gen_op_ld_v(s, ot, s->T1, s->A0);
3864                 tcg_gen_add_tl(s->T0, s->T0, s->T1);
3865                 gen_op_st_v(s, ot, s->T0, s->A0);
3866             }
3867             gen_op_mov_reg_v(s, ot, reg, s->T1);
3868         }
3869         gen_op_update2_cc(s);
3870         set_cc_op(s, CC_OP_ADDB + ot);
3871         break;
3872     case 0x1b0:
3873     case 0x1b1: /* cmpxchg Ev, Gv */
3874         {
3875             TCGv oldv, newv, cmpv, dest;
3876 
3877             ot = mo_b_d(b, dflag);
3878             modrm = x86_ldub_code(env, s);
3879             reg = ((modrm >> 3) & 7) | REX_R(s);
3880             mod = (modrm >> 6) & 3;
3881             oldv = tcg_temp_new();
3882             newv = tcg_temp_new();
3883             cmpv = tcg_temp_new();
3884             gen_op_mov_v_reg(s, ot, newv, reg);
3885             tcg_gen_mov_tl(cmpv, cpu_regs[R_EAX]);
3886             gen_extu(ot, cmpv);
3887             if (s->prefix & PREFIX_LOCK) {
3888                 if (mod == 3) {
3889                     goto illegal_op;
3890                 }
3891                 gen_lea_modrm(env, s, modrm);
3892                 tcg_gen_atomic_cmpxchg_tl(oldv, s->A0, cmpv, newv,
3893                                           s->mem_index, ot | MO_LE);
3894             } else {
3895                 if (mod == 3) {
3896                     rm = (modrm & 7) | REX_B(s);
3897                     gen_op_mov_v_reg(s, ot, oldv, rm);
3898                     gen_extu(ot, oldv);
3899 
3900                     /*
3901                      * Unlike the memory case, where "the destination operand receives
3902                      * a write cycle without regard to the result of the comparison",
3903                      * rm must not be touched altogether if the write fails, including
3904                      * not zero-extending it on 64-bit processors.  So, precompute
3905                      * the result of a successful writeback and perform the movcond
3906                      * directly on cpu_regs.  Also need to write accumulator first, in
3907                      * case rm is part of RAX too.
3908                      */
3909                     dest = gen_op_deposit_reg_v(s, ot, rm, newv, newv);
3910                     tcg_gen_movcond_tl(TCG_COND_EQ, dest, oldv, cmpv, newv, dest);
3911                 } else {
3912                     gen_lea_modrm(env, s, modrm);
3913                     gen_op_ld_v(s, ot, oldv, s->A0);
3914 
3915                     /*
3916                      * Perform an unconditional store cycle like physical cpu;
3917                      * must be before changing accumulator to ensure
3918                      * idempotency if the store faults and the instruction
3919                      * is restarted
3920                      */
3921                     tcg_gen_movcond_tl(TCG_COND_EQ, newv, oldv, cmpv, newv, oldv);
3922                     gen_op_st_v(s, ot, newv, s->A0);
3923                 }
3924             }
3925 	    /*
3926 	     * Write EAX only if the cmpxchg fails; reuse newv as the destination,
3927 	     * since it's dead here.
3928 	     */
3929             dest = gen_op_deposit_reg_v(s, ot, R_EAX, newv, oldv);
3930             tcg_gen_movcond_tl(TCG_COND_EQ, dest, oldv, cmpv, dest, newv);
3931             tcg_gen_mov_tl(cpu_cc_src, oldv);
3932             tcg_gen_mov_tl(s->cc_srcT, cmpv);
3933             tcg_gen_sub_tl(cpu_cc_dst, cmpv, oldv);
3934             set_cc_op(s, CC_OP_SUBB + ot);
3935             tcg_temp_free(oldv);
3936             tcg_temp_free(newv);
3937             tcg_temp_free(cmpv);
3938         }
3939         break;
3940     case 0x1c7: /* cmpxchg8b */
3941         modrm = x86_ldub_code(env, s);
3942         mod = (modrm >> 6) & 3;
3943         switch ((modrm >> 3) & 7) {
3944         case 1: /* CMPXCHG8, CMPXCHG16 */
3945             if (mod == 3) {
3946                 goto illegal_op;
3947             }
3948 #ifdef TARGET_X86_64
3949             if (dflag == MO_64) {
3950                 if (!(s->cpuid_ext_features & CPUID_EXT_CX16)) {
3951                     goto illegal_op;
3952                 }
3953                 gen_cmpxchg16b(s, env, modrm);
3954                 break;
3955             }
3956 #endif
3957             if (!(s->cpuid_features & CPUID_CX8)) {
3958                 goto illegal_op;
3959             }
3960             gen_cmpxchg8b(s, env, modrm);
3961             break;
3962 
3963         case 7: /* RDSEED */
3964         case 6: /* RDRAND */
3965             if (mod != 3 ||
3966                 (s->prefix & (PREFIX_LOCK | PREFIX_REPZ | PREFIX_REPNZ)) ||
3967                 !(s->cpuid_ext_features & CPUID_EXT_RDRAND)) {
3968                 goto illegal_op;
3969             }
3970             if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
3971                 gen_io_start();
3972                 s->base.is_jmp = DISAS_TOO_MANY;
3973             }
3974             gen_helper_rdrand(s->T0, cpu_env);
3975             rm = (modrm & 7) | REX_B(s);
3976             gen_op_mov_reg_v(s, dflag, rm, s->T0);
3977             set_cc_op(s, CC_OP_EFLAGS);
3978             break;
3979 
3980         default:
3981             goto illegal_op;
3982         }
3983         break;
3984 
3985         /**************************/
3986         /* push/pop */
3987     case 0x50 ... 0x57: /* push */
3988         gen_op_mov_v_reg(s, MO_32, s->T0, (b & 7) | REX_B(s));
3989         gen_push_v(s, s->T0);
3990         break;
3991     case 0x58 ... 0x5f: /* pop */
3992         ot = gen_pop_T0(s);
3993         /* NOTE: order is important for pop %sp */
3994         gen_pop_update(s, ot);
3995         gen_op_mov_reg_v(s, ot, (b & 7) | REX_B(s), s->T0);
3996         break;
3997     case 0x60: /* pusha */
3998         if (CODE64(s))
3999             goto illegal_op;
4000         gen_pusha(s);
4001         break;
4002     case 0x61: /* popa */
4003         if (CODE64(s))
4004             goto illegal_op;
4005         gen_popa(s);
4006         break;
4007     case 0x68: /* push Iv */
4008     case 0x6a:
4009         ot = mo_pushpop(s, dflag);
4010         if (b == 0x68)
4011             val = insn_get(env, s, ot);
4012         else
4013             val = (int8_t)insn_get(env, s, MO_8);
4014         tcg_gen_movi_tl(s->T0, val);
4015         gen_push_v(s, s->T0);
4016         break;
4017     case 0x8f: /* pop Ev */
4018         modrm = x86_ldub_code(env, s);
4019         mod = (modrm >> 6) & 3;
4020         ot = gen_pop_T0(s);
4021         if (mod == 3) {
4022             /* NOTE: order is important for pop %sp */
4023             gen_pop_update(s, ot);
4024             rm = (modrm & 7) | REX_B(s);
4025             gen_op_mov_reg_v(s, ot, rm, s->T0);
4026         } else {
4027             /* NOTE: order is important too for MMU exceptions */
4028             s->popl_esp_hack = 1 << ot;
4029             gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 1);
4030             s->popl_esp_hack = 0;
4031             gen_pop_update(s, ot);
4032         }
4033         break;
4034     case 0xc8: /* enter */
4035         {
4036             int level;
4037             val = x86_lduw_code(env, s);
4038             level = x86_ldub_code(env, s);
4039             gen_enter(s, val, level);
4040         }
4041         break;
4042     case 0xc9: /* leave */
4043         gen_leave(s);
4044         break;
4045     case 0x06: /* push es */
4046     case 0x0e: /* push cs */
4047     case 0x16: /* push ss */
4048     case 0x1e: /* push ds */
4049         if (CODE64(s))
4050             goto illegal_op;
4051         gen_op_movl_T0_seg(s, b >> 3);
4052         gen_push_v(s, s->T0);
4053         break;
4054     case 0x1a0: /* push fs */
4055     case 0x1a8: /* push gs */
4056         gen_op_movl_T0_seg(s, (b >> 3) & 7);
4057         gen_push_v(s, s->T0);
4058         break;
4059     case 0x07: /* pop es */
4060     case 0x17: /* pop ss */
4061     case 0x1f: /* pop ds */
4062         if (CODE64(s))
4063             goto illegal_op;
4064         reg = b >> 3;
4065         ot = gen_pop_T0(s);
4066         gen_movl_seg_T0(s, reg);
4067         gen_pop_update(s, ot);
4068         break;
4069     case 0x1a1: /* pop fs */
4070     case 0x1a9: /* pop gs */
4071         ot = gen_pop_T0(s);
4072         gen_movl_seg_T0(s, (b >> 3) & 7);
4073         gen_pop_update(s, ot);
4074         break;
4075 
4076         /**************************/
4077         /* mov */
4078     case 0x88:
4079     case 0x89: /* mov Gv, Ev */
4080         ot = mo_b_d(b, dflag);
4081         modrm = x86_ldub_code(env, s);
4082         reg = ((modrm >> 3) & 7) | REX_R(s);
4083 
4084         /* generate a generic store */
4085         gen_ldst_modrm(env, s, modrm, ot, reg, 1);
4086         break;
4087     case 0xc6:
4088     case 0xc7: /* mov Ev, Iv */
4089         ot = mo_b_d(b, dflag);
4090         modrm = x86_ldub_code(env, s);
4091         mod = (modrm >> 6) & 3;
4092         if (mod != 3) {
4093             s->rip_offset = insn_const_size(ot);
4094             gen_lea_modrm(env, s, modrm);
4095         }
4096         val = insn_get(env, s, ot);
4097         tcg_gen_movi_tl(s->T0, val);
4098         if (mod != 3) {
4099             gen_op_st_v(s, ot, s->T0, s->A0);
4100         } else {
4101             gen_op_mov_reg_v(s, ot, (modrm & 7) | REX_B(s), s->T0);
4102         }
4103         break;
4104     case 0x8a:
4105     case 0x8b: /* mov Ev, Gv */
4106         ot = mo_b_d(b, dflag);
4107         modrm = x86_ldub_code(env, s);
4108         reg = ((modrm >> 3) & 7) | REX_R(s);
4109 
4110         gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 0);
4111         gen_op_mov_reg_v(s, ot, reg, s->T0);
4112         break;
4113     case 0x8e: /* mov seg, Gv */
4114         modrm = x86_ldub_code(env, s);
4115         reg = (modrm >> 3) & 7;
4116         if (reg >= 6 || reg == R_CS)
4117             goto illegal_op;
4118         gen_ldst_modrm(env, s, modrm, MO_16, OR_TMP0, 0);
4119         gen_movl_seg_T0(s, reg);
4120         break;
4121     case 0x8c: /* mov Gv, seg */
4122         modrm = x86_ldub_code(env, s);
4123         reg = (modrm >> 3) & 7;
4124         mod = (modrm >> 6) & 3;
4125         if (reg >= 6)
4126             goto illegal_op;
4127         gen_op_movl_T0_seg(s, reg);
4128         ot = mod == 3 ? dflag : MO_16;
4129         gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 1);
4130         break;
4131 
4132     case 0x1b6: /* movzbS Gv, Eb */
4133     case 0x1b7: /* movzwS Gv, Eb */
4134     case 0x1be: /* movsbS Gv, Eb */
4135     case 0x1bf: /* movswS Gv, Eb */
4136         {
4137             MemOp d_ot;
4138             MemOp s_ot;
4139 
4140             /* d_ot is the size of destination */
4141             d_ot = dflag;
4142             /* ot is the size of source */
4143             ot = (b & 1) + MO_8;
4144             /* s_ot is the sign+size of source */
4145             s_ot = b & 8 ? MO_SIGN | ot : ot;
4146 
4147             modrm = x86_ldub_code(env, s);
4148             reg = ((modrm >> 3) & 7) | REX_R(s);
4149             mod = (modrm >> 6) & 3;
4150             rm = (modrm & 7) | REX_B(s);
4151 
4152             if (mod == 3) {
4153                 if (s_ot == MO_SB && byte_reg_is_xH(s, rm)) {
4154                     tcg_gen_sextract_tl(s->T0, cpu_regs[rm - 4], 8, 8);
4155                 } else {
4156                     gen_op_mov_v_reg(s, ot, s->T0, rm);
4157                     switch (s_ot) {
4158                     case MO_UB:
4159                         tcg_gen_ext8u_tl(s->T0, s->T0);
4160                         break;
4161                     case MO_SB:
4162                         tcg_gen_ext8s_tl(s->T0, s->T0);
4163                         break;
4164                     case MO_UW:
4165                         tcg_gen_ext16u_tl(s->T0, s->T0);
4166                         break;
4167                     default:
4168                     case MO_SW:
4169                         tcg_gen_ext16s_tl(s->T0, s->T0);
4170                         break;
4171                     }
4172                 }
4173                 gen_op_mov_reg_v(s, d_ot, reg, s->T0);
4174             } else {
4175                 gen_lea_modrm(env, s, modrm);
4176                 gen_op_ld_v(s, s_ot, s->T0, s->A0);
4177                 gen_op_mov_reg_v(s, d_ot, reg, s->T0);
4178             }
4179         }
4180         break;
4181 
4182     case 0x8d: /* lea */
4183         modrm = x86_ldub_code(env, s);
4184         mod = (modrm >> 6) & 3;
4185         if (mod == 3)
4186             goto illegal_op;
4187         reg = ((modrm >> 3) & 7) | REX_R(s);
4188         {
4189             AddressParts a = gen_lea_modrm_0(env, s, modrm);
4190             TCGv ea = gen_lea_modrm_1(s, a, false);
4191             gen_lea_v_seg(s, s->aflag, ea, -1, -1);
4192             gen_op_mov_reg_v(s, dflag, reg, s->A0);
4193         }
4194         break;
4195 
4196     case 0xa0: /* mov EAX, Ov */
4197     case 0xa1:
4198     case 0xa2: /* mov Ov, EAX */
4199     case 0xa3:
4200         {
4201             target_ulong offset_addr;
4202 
4203             ot = mo_b_d(b, dflag);
4204             offset_addr = insn_get_addr(env, s, s->aflag);
4205             tcg_gen_movi_tl(s->A0, offset_addr);
4206             gen_add_A0_ds_seg(s);
4207             if ((b & 2) == 0) {
4208                 gen_op_ld_v(s, ot, s->T0, s->A0);
4209                 gen_op_mov_reg_v(s, ot, R_EAX, s->T0);
4210             } else {
4211                 gen_op_mov_v_reg(s, ot, s->T0, R_EAX);
4212                 gen_op_st_v(s, ot, s->T0, s->A0);
4213             }
4214         }
4215         break;
4216     case 0xd7: /* xlat */
4217         tcg_gen_mov_tl(s->A0, cpu_regs[R_EBX]);
4218         tcg_gen_ext8u_tl(s->T0, cpu_regs[R_EAX]);
4219         tcg_gen_add_tl(s->A0, s->A0, s->T0);
4220         gen_extu(s->aflag, s->A0);
4221         gen_add_A0_ds_seg(s);
4222         gen_op_ld_v(s, MO_8, s->T0, s->A0);
4223         gen_op_mov_reg_v(s, MO_8, R_EAX, s->T0);
4224         break;
4225     case 0xb0 ... 0xb7: /* mov R, Ib */
4226         val = insn_get(env, s, MO_8);
4227         tcg_gen_movi_tl(s->T0, val);
4228         gen_op_mov_reg_v(s, MO_8, (b & 7) | REX_B(s), s->T0);
4229         break;
4230     case 0xb8 ... 0xbf: /* mov R, Iv */
4231 #ifdef TARGET_X86_64
4232         if (dflag == MO_64) {
4233             uint64_t tmp;
4234             /* 64 bit case */
4235             tmp = x86_ldq_code(env, s);
4236             reg = (b & 7) | REX_B(s);
4237             tcg_gen_movi_tl(s->T0, tmp);
4238             gen_op_mov_reg_v(s, MO_64, reg, s->T0);
4239         } else
4240 #endif
4241         {
4242             ot = dflag;
4243             val = insn_get(env, s, ot);
4244             reg = (b & 7) | REX_B(s);
4245             tcg_gen_movi_tl(s->T0, val);
4246             gen_op_mov_reg_v(s, ot, reg, s->T0);
4247         }
4248         break;
4249 
4250     case 0x91 ... 0x97: /* xchg R, EAX */
4251     do_xchg_reg_eax:
4252         ot = dflag;
4253         reg = (b & 7) | REX_B(s);
4254         rm = R_EAX;
4255         goto do_xchg_reg;
4256     case 0x86:
4257     case 0x87: /* xchg Ev, Gv */
4258         ot = mo_b_d(b, dflag);
4259         modrm = x86_ldub_code(env, s);
4260         reg = ((modrm >> 3) & 7) | REX_R(s);
4261         mod = (modrm >> 6) & 3;
4262         if (mod == 3) {
4263             rm = (modrm & 7) | REX_B(s);
4264         do_xchg_reg:
4265             gen_op_mov_v_reg(s, ot, s->T0, reg);
4266             gen_op_mov_v_reg(s, ot, s->T1, rm);
4267             gen_op_mov_reg_v(s, ot, rm, s->T0);
4268             gen_op_mov_reg_v(s, ot, reg, s->T1);
4269         } else {
4270             gen_lea_modrm(env, s, modrm);
4271             gen_op_mov_v_reg(s, ot, s->T0, reg);
4272             /* for xchg, lock is implicit */
4273             tcg_gen_atomic_xchg_tl(s->T1, s->A0, s->T0,
4274                                    s->mem_index, ot | MO_LE);
4275             gen_op_mov_reg_v(s, ot, reg, s->T1);
4276         }
4277         break;
4278     case 0xc4: /* les Gv */
4279         /* In CODE64 this is VEX3; see above.  */
4280         op = R_ES;
4281         goto do_lxx;
4282     case 0xc5: /* lds Gv */
4283         /* In CODE64 this is VEX2; see above.  */
4284         op = R_DS;
4285         goto do_lxx;
4286     case 0x1b2: /* lss Gv */
4287         op = R_SS;
4288         goto do_lxx;
4289     case 0x1b4: /* lfs Gv */
4290         op = R_FS;
4291         goto do_lxx;
4292     case 0x1b5: /* lgs Gv */
4293         op = R_GS;
4294     do_lxx:
4295         ot = dflag != MO_16 ? MO_32 : MO_16;
4296         modrm = x86_ldub_code(env, s);
4297         reg = ((modrm >> 3) & 7) | REX_R(s);
4298         mod = (modrm >> 6) & 3;
4299         if (mod == 3)
4300             goto illegal_op;
4301         gen_lea_modrm(env, s, modrm);
4302         gen_op_ld_v(s, ot, s->T1, s->A0);
4303         gen_add_A0_im(s, 1 << ot);
4304         /* load the segment first to handle exceptions properly */
4305         gen_op_ld_v(s, MO_16, s->T0, s->A0);
4306         gen_movl_seg_T0(s, op);
4307         /* then put the data */
4308         gen_op_mov_reg_v(s, ot, reg, s->T1);
4309         break;
4310 
4311         /************************/
4312         /* shifts */
4313     case 0xc0:
4314     case 0xc1:
4315         /* shift Ev,Ib */
4316         shift = 2;
4317     grp2:
4318         {
4319             ot = mo_b_d(b, dflag);
4320             modrm = x86_ldub_code(env, s);
4321             mod = (modrm >> 6) & 3;
4322             op = (modrm >> 3) & 7;
4323 
4324             if (mod != 3) {
4325                 if (shift == 2) {
4326                     s->rip_offset = 1;
4327                 }
4328                 gen_lea_modrm(env, s, modrm);
4329                 opreg = OR_TMP0;
4330             } else {
4331                 opreg = (modrm & 7) | REX_B(s);
4332             }
4333 
4334             /* simpler op */
4335             if (shift == 0) {
4336                 gen_shift(s, op, ot, opreg, OR_ECX);
4337             } else {
4338                 if (shift == 2) {
4339                     shift = x86_ldub_code(env, s);
4340                 }
4341                 gen_shifti(s, op, ot, opreg, shift);
4342             }
4343         }
4344         break;
4345     case 0xd0:
4346     case 0xd1:
4347         /* shift Ev,1 */
4348         shift = 1;
4349         goto grp2;
4350     case 0xd2:
4351     case 0xd3:
4352         /* shift Ev,cl */
4353         shift = 0;
4354         goto grp2;
4355 
4356     case 0x1a4: /* shld imm */
4357         op = 0;
4358         shift = 1;
4359         goto do_shiftd;
4360     case 0x1a5: /* shld cl */
4361         op = 0;
4362         shift = 0;
4363         goto do_shiftd;
4364     case 0x1ac: /* shrd imm */
4365         op = 1;
4366         shift = 1;
4367         goto do_shiftd;
4368     case 0x1ad: /* shrd cl */
4369         op = 1;
4370         shift = 0;
4371     do_shiftd:
4372         ot = dflag;
4373         modrm = x86_ldub_code(env, s);
4374         mod = (modrm >> 6) & 3;
4375         rm = (modrm & 7) | REX_B(s);
4376         reg = ((modrm >> 3) & 7) | REX_R(s);
4377         if (mod != 3) {
4378             gen_lea_modrm(env, s, modrm);
4379             opreg = OR_TMP0;
4380         } else {
4381             opreg = rm;
4382         }
4383         gen_op_mov_v_reg(s, ot, s->T1, reg);
4384 
4385         if (shift) {
4386             TCGv imm = tcg_const_tl(x86_ldub_code(env, s));
4387             gen_shiftd_rm_T1(s, ot, opreg, op, imm);
4388             tcg_temp_free(imm);
4389         } else {
4390             gen_shiftd_rm_T1(s, ot, opreg, op, cpu_regs[R_ECX]);
4391         }
4392         break;
4393 
4394         /************************/
4395         /* floats */
4396     case 0xd8 ... 0xdf:
4397         {
4398             bool update_fip = true;
4399 
4400             if (s->flags & (HF_EM_MASK | HF_TS_MASK)) {
4401                 /* if CR0.EM or CR0.TS are set, generate an FPU exception */
4402                 /* XXX: what to do if illegal op ? */
4403                 gen_exception(s, EXCP07_PREX);
4404                 break;
4405             }
4406             modrm = x86_ldub_code(env, s);
4407             mod = (modrm >> 6) & 3;
4408             rm = modrm & 7;
4409             op = ((b & 7) << 3) | ((modrm >> 3) & 7);
4410             if (mod != 3) {
4411                 /* memory op */
4412                 AddressParts a = gen_lea_modrm_0(env, s, modrm);
4413                 TCGv ea = gen_lea_modrm_1(s, a, false);
4414                 TCGv last_addr = tcg_temp_new();
4415                 bool update_fdp = true;
4416 
4417                 tcg_gen_mov_tl(last_addr, ea);
4418                 gen_lea_v_seg(s, s->aflag, ea, a.def_seg, s->override);
4419 
4420                 switch (op) {
4421                 case 0x00 ... 0x07: /* fxxxs */
4422                 case 0x10 ... 0x17: /* fixxxl */
4423                 case 0x20 ... 0x27: /* fxxxl */
4424                 case 0x30 ... 0x37: /* fixxx */
4425                     {
4426                         int op1;
4427                         op1 = op & 7;
4428 
4429                         switch (op >> 4) {
4430                         case 0:
4431                             tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
4432                                                 s->mem_index, MO_LEUL);
4433                             gen_helper_flds_FT0(cpu_env, s->tmp2_i32);
4434                             break;
4435                         case 1:
4436                             tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
4437                                                 s->mem_index, MO_LEUL);
4438                             gen_helper_fildl_FT0(cpu_env, s->tmp2_i32);
4439                             break;
4440                         case 2:
4441                             tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0,
4442                                                 s->mem_index, MO_LEUQ);
4443                             gen_helper_fldl_FT0(cpu_env, s->tmp1_i64);
4444                             break;
4445                         case 3:
4446                         default:
4447                             tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
4448                                                 s->mem_index, MO_LESW);
4449                             gen_helper_fildl_FT0(cpu_env, s->tmp2_i32);
4450                             break;
4451                         }
4452 
4453                         gen_helper_fp_arith_ST0_FT0(op1);
4454                         if (op1 == 3) {
4455                             /* fcomp needs pop */
4456                             gen_helper_fpop(cpu_env);
4457                         }
4458                     }
4459                     break;
4460                 case 0x08: /* flds */
4461                 case 0x0a: /* fsts */
4462                 case 0x0b: /* fstps */
4463                 case 0x18 ... 0x1b: /* fildl, fisttpl, fistl, fistpl */
4464                 case 0x28 ... 0x2b: /* fldl, fisttpll, fstl, fstpl */
4465                 case 0x38 ... 0x3b: /* filds, fisttps, fists, fistps */
4466                     switch (op & 7) {
4467                     case 0:
4468                         switch (op >> 4) {
4469                         case 0:
4470                             tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
4471                                                 s->mem_index, MO_LEUL);
4472                             gen_helper_flds_ST0(cpu_env, s->tmp2_i32);
4473                             break;
4474                         case 1:
4475                             tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
4476                                                 s->mem_index, MO_LEUL);
4477                             gen_helper_fildl_ST0(cpu_env, s->tmp2_i32);
4478                             break;
4479                         case 2:
4480                             tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0,
4481                                                 s->mem_index, MO_LEUQ);
4482                             gen_helper_fldl_ST0(cpu_env, s->tmp1_i64);
4483                             break;
4484                         case 3:
4485                         default:
4486                             tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
4487                                                 s->mem_index, MO_LESW);
4488                             gen_helper_fildl_ST0(cpu_env, s->tmp2_i32);
4489                             break;
4490                         }
4491                         break;
4492                     case 1:
4493                         /* XXX: the corresponding CPUID bit must be tested ! */
4494                         switch (op >> 4) {
4495                         case 1:
4496                             gen_helper_fisttl_ST0(s->tmp2_i32, cpu_env);
4497                             tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
4498                                                 s->mem_index, MO_LEUL);
4499                             break;
4500                         case 2:
4501                             gen_helper_fisttll_ST0(s->tmp1_i64, cpu_env);
4502                             tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0,
4503                                                 s->mem_index, MO_LEUQ);
4504                             break;
4505                         case 3:
4506                         default:
4507                             gen_helper_fistt_ST0(s->tmp2_i32, cpu_env);
4508                             tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
4509                                                 s->mem_index, MO_LEUW);
4510                             break;
4511                         }
4512                         gen_helper_fpop(cpu_env);
4513                         break;
4514                     default:
4515                         switch (op >> 4) {
4516                         case 0:
4517                             gen_helper_fsts_ST0(s->tmp2_i32, cpu_env);
4518                             tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
4519                                                 s->mem_index, MO_LEUL);
4520                             break;
4521                         case 1:
4522                             gen_helper_fistl_ST0(s->tmp2_i32, cpu_env);
4523                             tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
4524                                                 s->mem_index, MO_LEUL);
4525                             break;
4526                         case 2:
4527                             gen_helper_fstl_ST0(s->tmp1_i64, cpu_env);
4528                             tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0,
4529                                                 s->mem_index, MO_LEUQ);
4530                             break;
4531                         case 3:
4532                         default:
4533                             gen_helper_fist_ST0(s->tmp2_i32, cpu_env);
4534                             tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
4535                                                 s->mem_index, MO_LEUW);
4536                             break;
4537                         }
4538                         if ((op & 7) == 3) {
4539                             gen_helper_fpop(cpu_env);
4540                         }
4541                         break;
4542                     }
4543                     break;
4544                 case 0x0c: /* fldenv mem */
4545                     gen_helper_fldenv(cpu_env, s->A0,
4546                                       tcg_const_i32(dflag - 1));
4547                     update_fip = update_fdp = false;
4548                     break;
4549                 case 0x0d: /* fldcw mem */
4550                     tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0,
4551                                         s->mem_index, MO_LEUW);
4552                     gen_helper_fldcw(cpu_env, s->tmp2_i32);
4553                     update_fip = update_fdp = false;
4554                     break;
4555                 case 0x0e: /* fnstenv mem */
4556                     gen_helper_fstenv(cpu_env, s->A0,
4557                                       tcg_const_i32(dflag - 1));
4558                     update_fip = update_fdp = false;
4559                     break;
4560                 case 0x0f: /* fnstcw mem */
4561                     gen_helper_fnstcw(s->tmp2_i32, cpu_env);
4562                     tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
4563                                         s->mem_index, MO_LEUW);
4564                     update_fip = update_fdp = false;
4565                     break;
4566                 case 0x1d: /* fldt mem */
4567                     gen_helper_fldt_ST0(cpu_env, s->A0);
4568                     break;
4569                 case 0x1f: /* fstpt mem */
4570                     gen_helper_fstt_ST0(cpu_env, s->A0);
4571                     gen_helper_fpop(cpu_env);
4572                     break;
4573                 case 0x2c: /* frstor mem */
4574                     gen_helper_frstor(cpu_env, s->A0,
4575                                       tcg_const_i32(dflag - 1));
4576                     update_fip = update_fdp = false;
4577                     break;
4578                 case 0x2e: /* fnsave mem */
4579                     gen_helper_fsave(cpu_env, s->A0,
4580                                      tcg_const_i32(dflag - 1));
4581                     update_fip = update_fdp = false;
4582                     break;
4583                 case 0x2f: /* fnstsw mem */
4584                     gen_helper_fnstsw(s->tmp2_i32, cpu_env);
4585                     tcg_gen_qemu_st_i32(s->tmp2_i32, s->A0,
4586                                         s->mem_index, MO_LEUW);
4587                     update_fip = update_fdp = false;
4588                     break;
4589                 case 0x3c: /* fbld */
4590                     gen_helper_fbld_ST0(cpu_env, s->A0);
4591                     break;
4592                 case 0x3e: /* fbstp */
4593                     gen_helper_fbst_ST0(cpu_env, s->A0);
4594                     gen_helper_fpop(cpu_env);
4595                     break;
4596                 case 0x3d: /* fildll */
4597                     tcg_gen_qemu_ld_i64(s->tmp1_i64, s->A0,
4598                                         s->mem_index, MO_LEUQ);
4599                     gen_helper_fildll_ST0(cpu_env, s->tmp1_i64);
4600                     break;
4601                 case 0x3f: /* fistpll */
4602                     gen_helper_fistll_ST0(s->tmp1_i64, cpu_env);
4603                     tcg_gen_qemu_st_i64(s->tmp1_i64, s->A0,
4604                                         s->mem_index, MO_LEUQ);
4605                     gen_helper_fpop(cpu_env);
4606                     break;
4607                 default:
4608                     goto unknown_op;
4609                 }
4610 
4611                 if (update_fdp) {
4612                     int last_seg = s->override >= 0 ? s->override : a.def_seg;
4613 
4614                     tcg_gen_ld_i32(s->tmp2_i32, cpu_env,
4615                                    offsetof(CPUX86State,
4616                                             segs[last_seg].selector));
4617                     tcg_gen_st16_i32(s->tmp2_i32, cpu_env,
4618                                      offsetof(CPUX86State, fpds));
4619                     tcg_gen_st_tl(last_addr, cpu_env,
4620                                   offsetof(CPUX86State, fpdp));
4621                 }
4622                 tcg_temp_free(last_addr);
4623             } else {
4624                 /* register float ops */
4625                 opreg = rm;
4626 
4627                 switch (op) {
4628                 case 0x08: /* fld sti */
4629                     gen_helper_fpush(cpu_env);
4630                     gen_helper_fmov_ST0_STN(cpu_env,
4631                                             tcg_const_i32((opreg + 1) & 7));
4632                     break;
4633                 case 0x09: /* fxchg sti */
4634                 case 0x29: /* fxchg4 sti, undocumented op */
4635                 case 0x39: /* fxchg7 sti, undocumented op */
4636                     gen_helper_fxchg_ST0_STN(cpu_env, tcg_const_i32(opreg));
4637                     break;
4638                 case 0x0a: /* grp d9/2 */
4639                     switch (rm) {
4640                     case 0: /* fnop */
4641                         /* check exceptions (FreeBSD FPU probe) */
4642                         gen_helper_fwait(cpu_env);
4643                         update_fip = false;
4644                         break;
4645                     default:
4646                         goto unknown_op;
4647                     }
4648                     break;
4649                 case 0x0c: /* grp d9/4 */
4650                     switch (rm) {
4651                     case 0: /* fchs */
4652                         gen_helper_fchs_ST0(cpu_env);
4653                         break;
4654                     case 1: /* fabs */
4655                         gen_helper_fabs_ST0(cpu_env);
4656                         break;
4657                     case 4: /* ftst */
4658                         gen_helper_fldz_FT0(cpu_env);
4659                         gen_helper_fcom_ST0_FT0(cpu_env);
4660                         break;
4661                     case 5: /* fxam */
4662                         gen_helper_fxam_ST0(cpu_env);
4663                         break;
4664                     default:
4665                         goto unknown_op;
4666                     }
4667                     break;
4668                 case 0x0d: /* grp d9/5 */
4669                     {
4670                         switch (rm) {
4671                         case 0:
4672                             gen_helper_fpush(cpu_env);
4673                             gen_helper_fld1_ST0(cpu_env);
4674                             break;
4675                         case 1:
4676                             gen_helper_fpush(cpu_env);
4677                             gen_helper_fldl2t_ST0(cpu_env);
4678                             break;
4679                         case 2:
4680                             gen_helper_fpush(cpu_env);
4681                             gen_helper_fldl2e_ST0(cpu_env);
4682                             break;
4683                         case 3:
4684                             gen_helper_fpush(cpu_env);
4685                             gen_helper_fldpi_ST0(cpu_env);
4686                             break;
4687                         case 4:
4688                             gen_helper_fpush(cpu_env);
4689                             gen_helper_fldlg2_ST0(cpu_env);
4690                             break;
4691                         case 5:
4692                             gen_helper_fpush(cpu_env);
4693                             gen_helper_fldln2_ST0(cpu_env);
4694                             break;
4695                         case 6:
4696                             gen_helper_fpush(cpu_env);
4697                             gen_helper_fldz_ST0(cpu_env);
4698                             break;
4699                         default:
4700                             goto unknown_op;
4701                         }
4702                     }
4703                     break;
4704                 case 0x0e: /* grp d9/6 */
4705                     switch (rm) {
4706                     case 0: /* f2xm1 */
4707                         gen_helper_f2xm1(cpu_env);
4708                         break;
4709                     case 1: /* fyl2x */
4710                         gen_helper_fyl2x(cpu_env);
4711                         break;
4712                     case 2: /* fptan */
4713                         gen_helper_fptan(cpu_env);
4714                         break;
4715                     case 3: /* fpatan */
4716                         gen_helper_fpatan(cpu_env);
4717                         break;
4718                     case 4: /* fxtract */
4719                         gen_helper_fxtract(cpu_env);
4720                         break;
4721                     case 5: /* fprem1 */
4722                         gen_helper_fprem1(cpu_env);
4723                         break;
4724                     case 6: /* fdecstp */
4725                         gen_helper_fdecstp(cpu_env);
4726                         break;
4727                     default:
4728                     case 7: /* fincstp */
4729                         gen_helper_fincstp(cpu_env);
4730                         break;
4731                     }
4732                     break;
4733                 case 0x0f: /* grp d9/7 */
4734                     switch (rm) {
4735                     case 0: /* fprem */
4736                         gen_helper_fprem(cpu_env);
4737                         break;
4738                     case 1: /* fyl2xp1 */
4739                         gen_helper_fyl2xp1(cpu_env);
4740                         break;
4741                     case 2: /* fsqrt */
4742                         gen_helper_fsqrt(cpu_env);
4743                         break;
4744                     case 3: /* fsincos */
4745                         gen_helper_fsincos(cpu_env);
4746                         break;
4747                     case 5: /* fscale */
4748                         gen_helper_fscale(cpu_env);
4749                         break;
4750                     case 4: /* frndint */
4751                         gen_helper_frndint(cpu_env);
4752                         break;
4753                     case 6: /* fsin */
4754                         gen_helper_fsin(cpu_env);
4755                         break;
4756                     default:
4757                     case 7: /* fcos */
4758                         gen_helper_fcos(cpu_env);
4759                         break;
4760                     }
4761                     break;
4762                 case 0x00: case 0x01: case 0x04 ... 0x07: /* fxxx st, sti */
4763                 case 0x20: case 0x21: case 0x24 ... 0x27: /* fxxx sti, st */
4764                 case 0x30: case 0x31: case 0x34 ... 0x37: /* fxxxp sti, st */
4765                     {
4766                         int op1;
4767 
4768                         op1 = op & 7;
4769                         if (op >= 0x20) {
4770                             gen_helper_fp_arith_STN_ST0(op1, opreg);
4771                             if (op >= 0x30) {
4772                                 gen_helper_fpop(cpu_env);
4773                             }
4774                         } else {
4775                             gen_helper_fmov_FT0_STN(cpu_env,
4776                                                     tcg_const_i32(opreg));
4777                             gen_helper_fp_arith_ST0_FT0(op1);
4778                         }
4779                     }
4780                     break;
4781                 case 0x02: /* fcom */
4782                 case 0x22: /* fcom2, undocumented op */
4783                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4784                     gen_helper_fcom_ST0_FT0(cpu_env);
4785                     break;
4786                 case 0x03: /* fcomp */
4787                 case 0x23: /* fcomp3, undocumented op */
4788                 case 0x32: /* fcomp5, undocumented op */
4789                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4790                     gen_helper_fcom_ST0_FT0(cpu_env);
4791                     gen_helper_fpop(cpu_env);
4792                     break;
4793                 case 0x15: /* da/5 */
4794                     switch (rm) {
4795                     case 1: /* fucompp */
4796                         gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(1));
4797                         gen_helper_fucom_ST0_FT0(cpu_env);
4798                         gen_helper_fpop(cpu_env);
4799                         gen_helper_fpop(cpu_env);
4800                         break;
4801                     default:
4802                         goto unknown_op;
4803                     }
4804                     break;
4805                 case 0x1c:
4806                     switch (rm) {
4807                     case 0: /* feni (287 only, just do nop here) */
4808                         break;
4809                     case 1: /* fdisi (287 only, just do nop here) */
4810                         break;
4811                     case 2: /* fclex */
4812                         gen_helper_fclex(cpu_env);
4813                         update_fip = false;
4814                         break;
4815                     case 3: /* fninit */
4816                         gen_helper_fninit(cpu_env);
4817                         update_fip = false;
4818                         break;
4819                     case 4: /* fsetpm (287 only, just do nop here) */
4820                         break;
4821                     default:
4822                         goto unknown_op;
4823                     }
4824                     break;
4825                 case 0x1d: /* fucomi */
4826                     if (!(s->cpuid_features & CPUID_CMOV)) {
4827                         goto illegal_op;
4828                     }
4829                     gen_update_cc_op(s);
4830                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4831                     gen_helper_fucomi_ST0_FT0(cpu_env);
4832                     set_cc_op(s, CC_OP_EFLAGS);
4833                     break;
4834                 case 0x1e: /* fcomi */
4835                     if (!(s->cpuid_features & CPUID_CMOV)) {
4836                         goto illegal_op;
4837                     }
4838                     gen_update_cc_op(s);
4839                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4840                     gen_helper_fcomi_ST0_FT0(cpu_env);
4841                     set_cc_op(s, CC_OP_EFLAGS);
4842                     break;
4843                 case 0x28: /* ffree sti */
4844                     gen_helper_ffree_STN(cpu_env, tcg_const_i32(opreg));
4845                     break;
4846                 case 0x2a: /* fst sti */
4847                     gen_helper_fmov_STN_ST0(cpu_env, tcg_const_i32(opreg));
4848                     break;
4849                 case 0x2b: /* fstp sti */
4850                 case 0x0b: /* fstp1 sti, undocumented op */
4851                 case 0x3a: /* fstp8 sti, undocumented op */
4852                 case 0x3b: /* fstp9 sti, undocumented op */
4853                     gen_helper_fmov_STN_ST0(cpu_env, tcg_const_i32(opreg));
4854                     gen_helper_fpop(cpu_env);
4855                     break;
4856                 case 0x2c: /* fucom st(i) */
4857                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4858                     gen_helper_fucom_ST0_FT0(cpu_env);
4859                     break;
4860                 case 0x2d: /* fucomp st(i) */
4861                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4862                     gen_helper_fucom_ST0_FT0(cpu_env);
4863                     gen_helper_fpop(cpu_env);
4864                     break;
4865                 case 0x33: /* de/3 */
4866                     switch (rm) {
4867                     case 1: /* fcompp */
4868                         gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(1));
4869                         gen_helper_fcom_ST0_FT0(cpu_env);
4870                         gen_helper_fpop(cpu_env);
4871                         gen_helper_fpop(cpu_env);
4872                         break;
4873                     default:
4874                         goto unknown_op;
4875                     }
4876                     break;
4877                 case 0x38: /* ffreep sti, undocumented op */
4878                     gen_helper_ffree_STN(cpu_env, tcg_const_i32(opreg));
4879                     gen_helper_fpop(cpu_env);
4880                     break;
4881                 case 0x3c: /* df/4 */
4882                     switch (rm) {
4883                     case 0:
4884                         gen_helper_fnstsw(s->tmp2_i32, cpu_env);
4885                         tcg_gen_extu_i32_tl(s->T0, s->tmp2_i32);
4886                         gen_op_mov_reg_v(s, MO_16, R_EAX, s->T0);
4887                         break;
4888                     default:
4889                         goto unknown_op;
4890                     }
4891                     break;
4892                 case 0x3d: /* fucomip */
4893                     if (!(s->cpuid_features & CPUID_CMOV)) {
4894                         goto illegal_op;
4895                     }
4896                     gen_update_cc_op(s);
4897                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4898                     gen_helper_fucomi_ST0_FT0(cpu_env);
4899                     gen_helper_fpop(cpu_env);
4900                     set_cc_op(s, CC_OP_EFLAGS);
4901                     break;
4902                 case 0x3e: /* fcomip */
4903                     if (!(s->cpuid_features & CPUID_CMOV)) {
4904                         goto illegal_op;
4905                     }
4906                     gen_update_cc_op(s);
4907                     gen_helper_fmov_FT0_STN(cpu_env, tcg_const_i32(opreg));
4908                     gen_helper_fcomi_ST0_FT0(cpu_env);
4909                     gen_helper_fpop(cpu_env);
4910                     set_cc_op(s, CC_OP_EFLAGS);
4911                     break;
4912                 case 0x10 ... 0x13: /* fcmovxx */
4913                 case 0x18 ... 0x1b:
4914                     {
4915                         int op1;
4916                         TCGLabel *l1;
4917                         static const uint8_t fcmov_cc[8] = {
4918                             (JCC_B << 1),
4919                             (JCC_Z << 1),
4920                             (JCC_BE << 1),
4921                             (JCC_P << 1),
4922                         };
4923 
4924                         if (!(s->cpuid_features & CPUID_CMOV)) {
4925                             goto illegal_op;
4926                         }
4927                         op1 = fcmov_cc[op & 3] | (((op >> 3) & 1) ^ 1);
4928                         l1 = gen_new_label();
4929                         gen_jcc1_noeob(s, op1, l1);
4930                         gen_helper_fmov_ST0_STN(cpu_env, tcg_const_i32(opreg));
4931                         gen_set_label(l1);
4932                     }
4933                     break;
4934                 default:
4935                     goto unknown_op;
4936                 }
4937             }
4938 
4939             if (update_fip) {
4940                 tcg_gen_ld_i32(s->tmp2_i32, cpu_env,
4941                                offsetof(CPUX86State, segs[R_CS].selector));
4942                 tcg_gen_st16_i32(s->tmp2_i32, cpu_env,
4943                                  offsetof(CPUX86State, fpcs));
4944                 tcg_gen_st_tl(eip_cur_tl(s),
4945                               cpu_env, offsetof(CPUX86State, fpip));
4946             }
4947         }
4948         break;
4949         /************************/
4950         /* string ops */
4951 
4952     case 0xa4: /* movsS */
4953     case 0xa5:
4954         ot = mo_b_d(b, dflag);
4955         if (prefixes & (PREFIX_REPZ | PREFIX_REPNZ)) {
4956             gen_repz_movs(s, ot);
4957         } else {
4958             gen_movs(s, ot);
4959         }
4960         break;
4961 
4962     case 0xaa: /* stosS */
4963     case 0xab:
4964         ot = mo_b_d(b, dflag);
4965         if (prefixes & (PREFIX_REPZ | PREFIX_REPNZ)) {
4966             gen_repz_stos(s, ot);
4967         } else {
4968             gen_stos(s, ot);
4969         }
4970         break;
4971     case 0xac: /* lodsS */
4972     case 0xad:
4973         ot = mo_b_d(b, dflag);
4974         if (prefixes & (PREFIX_REPZ | PREFIX_REPNZ)) {
4975             gen_repz_lods(s, ot);
4976         } else {
4977             gen_lods(s, ot);
4978         }
4979         break;
4980     case 0xae: /* scasS */
4981     case 0xaf:
4982         ot = mo_b_d(b, dflag);
4983         if (prefixes & PREFIX_REPNZ) {
4984             gen_repz_scas(s, ot, 1);
4985         } else if (prefixes & PREFIX_REPZ) {
4986             gen_repz_scas(s, ot, 0);
4987         } else {
4988             gen_scas(s, ot);
4989         }
4990         break;
4991 
4992     case 0xa6: /* cmpsS */
4993     case 0xa7:
4994         ot = mo_b_d(b, dflag);
4995         if (prefixes & PREFIX_REPNZ) {
4996             gen_repz_cmps(s, ot, 1);
4997         } else if (prefixes & PREFIX_REPZ) {
4998             gen_repz_cmps(s, ot, 0);
4999         } else {
5000             gen_cmps(s, ot);
5001         }
5002         break;
5003     case 0x6c: /* insS */
5004     case 0x6d:
5005         ot = mo_b_d32(b, dflag);
5006         tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_EDX]);
5007         tcg_gen_ext16u_i32(s->tmp2_i32, s->tmp2_i32);
5008         if (!gen_check_io(s, ot, s->tmp2_i32,
5009                           SVM_IOIO_TYPE_MASK | SVM_IOIO_STR_MASK)) {
5010             break;
5011         }
5012         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
5013             gen_io_start();
5014             s->base.is_jmp = DISAS_TOO_MANY;
5015         }
5016         if (prefixes & (PREFIX_REPZ | PREFIX_REPNZ)) {
5017             gen_repz_ins(s, ot);
5018         } else {
5019             gen_ins(s, ot);
5020         }
5021         break;
5022     case 0x6e: /* outsS */
5023     case 0x6f:
5024         ot = mo_b_d32(b, dflag);
5025         tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_EDX]);
5026         tcg_gen_ext16u_i32(s->tmp2_i32, s->tmp2_i32);
5027         if (!gen_check_io(s, ot, s->tmp2_i32, SVM_IOIO_STR_MASK)) {
5028             break;
5029         }
5030         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
5031             gen_io_start();
5032             s->base.is_jmp = DISAS_TOO_MANY;
5033         }
5034         if (prefixes & (PREFIX_REPZ | PREFIX_REPNZ)) {
5035             gen_repz_outs(s, ot);
5036         } else {
5037             gen_outs(s, ot);
5038         }
5039         break;
5040 
5041         /************************/
5042         /* port I/O */
5043 
5044     case 0xe4:
5045     case 0xe5:
5046         ot = mo_b_d32(b, dflag);
5047         val = x86_ldub_code(env, s);
5048         tcg_gen_movi_i32(s->tmp2_i32, val);
5049         if (!gen_check_io(s, ot, s->tmp2_i32, SVM_IOIO_TYPE_MASK)) {
5050             break;
5051         }
5052         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
5053             gen_io_start();
5054             s->base.is_jmp = DISAS_TOO_MANY;
5055         }
5056         gen_helper_in_func(ot, s->T1, s->tmp2_i32);
5057         gen_op_mov_reg_v(s, ot, R_EAX, s->T1);
5058         gen_bpt_io(s, s->tmp2_i32, ot);
5059         break;
5060     case 0xe6:
5061     case 0xe7:
5062         ot = mo_b_d32(b, dflag);
5063         val = x86_ldub_code(env, s);
5064         tcg_gen_movi_i32(s->tmp2_i32, val);
5065         if (!gen_check_io(s, ot, s->tmp2_i32, 0)) {
5066             break;
5067         }
5068         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
5069             gen_io_start();
5070             s->base.is_jmp = DISAS_TOO_MANY;
5071         }
5072         gen_op_mov_v_reg(s, ot, s->T1, R_EAX);
5073         tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
5074         gen_helper_out_func(ot, s->tmp2_i32, s->tmp3_i32);
5075         gen_bpt_io(s, s->tmp2_i32, ot);
5076         break;
5077     case 0xec:
5078     case 0xed:
5079         ot = mo_b_d32(b, dflag);
5080         tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_EDX]);
5081         tcg_gen_ext16u_i32(s->tmp2_i32, s->tmp2_i32);
5082         if (!gen_check_io(s, ot, s->tmp2_i32, SVM_IOIO_TYPE_MASK)) {
5083             break;
5084         }
5085         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
5086             gen_io_start();
5087             s->base.is_jmp = DISAS_TOO_MANY;
5088         }
5089         gen_helper_in_func(ot, s->T1, s->tmp2_i32);
5090         gen_op_mov_reg_v(s, ot, R_EAX, s->T1);
5091         gen_bpt_io(s, s->tmp2_i32, ot);
5092         break;
5093     case 0xee:
5094     case 0xef:
5095         ot = mo_b_d32(b, dflag);
5096         tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_EDX]);
5097         tcg_gen_ext16u_i32(s->tmp2_i32, s->tmp2_i32);
5098         if (!gen_check_io(s, ot, s->tmp2_i32, 0)) {
5099             break;
5100         }
5101         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
5102             gen_io_start();
5103             s->base.is_jmp = DISAS_TOO_MANY;
5104         }
5105         gen_op_mov_v_reg(s, ot, s->T1, R_EAX);
5106         tcg_gen_trunc_tl_i32(s->tmp3_i32, s->T1);
5107         gen_helper_out_func(ot, s->tmp2_i32, s->tmp3_i32);
5108         gen_bpt_io(s, s->tmp2_i32, ot);
5109         break;
5110 
5111         /************************/
5112         /* control */
5113     case 0xc2: /* ret im */
5114         val = x86_ldsw_code(env, s);
5115         ot = gen_pop_T0(s);
5116         gen_stack_update(s, val + (1 << ot));
5117         /* Note that gen_pop_T0 uses a zero-extending load.  */
5118         gen_op_jmp_v(s, s->T0);
5119         gen_bnd_jmp(s);
5120         s->base.is_jmp = DISAS_JUMP;
5121         break;
5122     case 0xc3: /* ret */
5123         ot = gen_pop_T0(s);
5124         gen_pop_update(s, ot);
5125         /* Note that gen_pop_T0 uses a zero-extending load.  */
5126         gen_op_jmp_v(s, s->T0);
5127         gen_bnd_jmp(s);
5128         s->base.is_jmp = DISAS_JUMP;
5129         break;
5130     case 0xca: /* lret im */
5131         val = x86_ldsw_code(env, s);
5132     do_lret:
5133         if (PE(s) && !VM86(s)) {
5134             gen_update_cc_op(s);
5135             gen_update_eip_cur(s);
5136             gen_helper_lret_protected(cpu_env, tcg_const_i32(dflag - 1),
5137                                       tcg_const_i32(val));
5138         } else {
5139             gen_stack_A0(s);
5140             /* pop offset */
5141             gen_op_ld_v(s, dflag, s->T0, s->A0);
5142             /* NOTE: keeping EIP updated is not a problem in case of
5143                exception */
5144             gen_op_jmp_v(s, s->T0);
5145             /* pop selector */
5146             gen_add_A0_im(s, 1 << dflag);
5147             gen_op_ld_v(s, dflag, s->T0, s->A0);
5148             gen_op_movl_seg_T0_vm(s, R_CS);
5149             /* add stack offset */
5150             gen_stack_update(s, val + (2 << dflag));
5151         }
5152         s->base.is_jmp = DISAS_EOB_ONLY;
5153         break;
5154     case 0xcb: /* lret */
5155         val = 0;
5156         goto do_lret;
5157     case 0xcf: /* iret */
5158         gen_svm_check_intercept(s, SVM_EXIT_IRET);
5159         if (!PE(s) || VM86(s)) {
5160             /* real mode or vm86 mode */
5161             if (!check_vm86_iopl(s)) {
5162                 break;
5163             }
5164             gen_helper_iret_real(cpu_env, tcg_const_i32(dflag - 1));
5165         } else {
5166             gen_helper_iret_protected(cpu_env, tcg_constant_i32(dflag - 1),
5167                                       eip_next_i32(s));
5168         }
5169         set_cc_op(s, CC_OP_EFLAGS);
5170         s->base.is_jmp = DISAS_EOB_ONLY;
5171         break;
5172     case 0xe8: /* call im */
5173         {
5174             int diff = (dflag != MO_16
5175                         ? (int32_t)insn_get(env, s, MO_32)
5176                         : (int16_t)insn_get(env, s, MO_16));
5177             gen_push_v(s, eip_next_tl(s));
5178             gen_bnd_jmp(s);
5179             gen_jmp_rel(s, dflag, diff, 0);
5180         }
5181         break;
5182     case 0x9a: /* lcall im */
5183         {
5184             unsigned int selector, offset;
5185 
5186             if (CODE64(s))
5187                 goto illegal_op;
5188             ot = dflag;
5189             offset = insn_get(env, s, ot);
5190             selector = insn_get(env, s, MO_16);
5191 
5192             tcg_gen_movi_tl(s->T0, selector);
5193             tcg_gen_movi_tl(s->T1, offset);
5194         }
5195         goto do_lcall;
5196     case 0xe9: /* jmp im */
5197         {
5198             int diff = (dflag != MO_16
5199                         ? (int32_t)insn_get(env, s, MO_32)
5200                         : (int16_t)insn_get(env, s, MO_16));
5201             gen_bnd_jmp(s);
5202             gen_jmp_rel(s, dflag, diff, 0);
5203         }
5204         break;
5205     case 0xea: /* ljmp im */
5206         {
5207             unsigned int selector, offset;
5208 
5209             if (CODE64(s))
5210                 goto illegal_op;
5211             ot = dflag;
5212             offset = insn_get(env, s, ot);
5213             selector = insn_get(env, s, MO_16);
5214 
5215             tcg_gen_movi_tl(s->T0, selector);
5216             tcg_gen_movi_tl(s->T1, offset);
5217         }
5218         goto do_ljmp;
5219     case 0xeb: /* jmp Jb */
5220         {
5221             int diff = (int8_t)insn_get(env, s, MO_8);
5222             gen_jmp_rel(s, dflag, diff, 0);
5223         }
5224         break;
5225     case 0x70 ... 0x7f: /* jcc Jb */
5226         {
5227             int diff = (int8_t)insn_get(env, s, MO_8);
5228             gen_bnd_jmp(s);
5229             gen_jcc(s, b, diff);
5230         }
5231         break;
5232     case 0x180 ... 0x18f: /* jcc Jv */
5233         {
5234             int diff = (dflag != MO_16
5235                         ? (int32_t)insn_get(env, s, MO_32)
5236                         : (int16_t)insn_get(env, s, MO_16));
5237             gen_bnd_jmp(s);
5238             gen_jcc(s, b, diff);
5239         }
5240         break;
5241 
5242     case 0x190 ... 0x19f: /* setcc Gv */
5243         modrm = x86_ldub_code(env, s);
5244         gen_setcc1(s, b, s->T0);
5245         gen_ldst_modrm(env, s, modrm, MO_8, OR_TMP0, 1);
5246         break;
5247     case 0x140 ... 0x14f: /* cmov Gv, Ev */
5248         if (!(s->cpuid_features & CPUID_CMOV)) {
5249             goto illegal_op;
5250         }
5251         ot = dflag;
5252         modrm = x86_ldub_code(env, s);
5253         reg = ((modrm >> 3) & 7) | REX_R(s);
5254         gen_cmovcc1(env, s, ot, b, modrm, reg);
5255         break;
5256 
5257         /************************/
5258         /* flags */
5259     case 0x9c: /* pushf */
5260         gen_svm_check_intercept(s, SVM_EXIT_PUSHF);
5261         if (check_vm86_iopl(s)) {
5262             gen_update_cc_op(s);
5263             gen_helper_read_eflags(s->T0, cpu_env);
5264             gen_push_v(s, s->T0);
5265         }
5266         break;
5267     case 0x9d: /* popf */
5268         gen_svm_check_intercept(s, SVM_EXIT_POPF);
5269         if (check_vm86_iopl(s)) {
5270             ot = gen_pop_T0(s);
5271             if (CPL(s) == 0) {
5272                 if (dflag != MO_16) {
5273                     gen_helper_write_eflags(cpu_env, s->T0,
5274                                             tcg_const_i32((TF_MASK | AC_MASK |
5275                                                            ID_MASK | NT_MASK |
5276                                                            IF_MASK |
5277                                                            IOPL_MASK)));
5278                 } else {
5279                     gen_helper_write_eflags(cpu_env, s->T0,
5280                                             tcg_const_i32((TF_MASK | AC_MASK |
5281                                                            ID_MASK | NT_MASK |
5282                                                            IF_MASK | IOPL_MASK)
5283                                                           & 0xffff));
5284                 }
5285             } else {
5286                 if (CPL(s) <= IOPL(s)) {
5287                     if (dflag != MO_16) {
5288                         gen_helper_write_eflags(cpu_env, s->T0,
5289                                                 tcg_const_i32((TF_MASK |
5290                                                                AC_MASK |
5291                                                                ID_MASK |
5292                                                                NT_MASK |
5293                                                                IF_MASK)));
5294                     } else {
5295                         gen_helper_write_eflags(cpu_env, s->T0,
5296                                                 tcg_const_i32((TF_MASK |
5297                                                                AC_MASK |
5298                                                                ID_MASK |
5299                                                                NT_MASK |
5300                                                                IF_MASK)
5301                                                               & 0xffff));
5302                     }
5303                 } else {
5304                     if (dflag != MO_16) {
5305                         gen_helper_write_eflags(cpu_env, s->T0,
5306                                            tcg_const_i32((TF_MASK | AC_MASK |
5307                                                           ID_MASK | NT_MASK)));
5308                     } else {
5309                         gen_helper_write_eflags(cpu_env, s->T0,
5310                                            tcg_const_i32((TF_MASK | AC_MASK |
5311                                                           ID_MASK | NT_MASK)
5312                                                          & 0xffff));
5313                     }
5314                 }
5315             }
5316             gen_pop_update(s, ot);
5317             set_cc_op(s, CC_OP_EFLAGS);
5318             /* abort translation because TF/AC flag may change */
5319             s->base.is_jmp = DISAS_EOB_NEXT;
5320         }
5321         break;
5322     case 0x9e: /* sahf */
5323         if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM))
5324             goto illegal_op;
5325         tcg_gen_shri_tl(s->T0, cpu_regs[R_EAX], 8);
5326         gen_compute_eflags(s);
5327         tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, CC_O);
5328         tcg_gen_andi_tl(s->T0, s->T0, CC_S | CC_Z | CC_A | CC_P | CC_C);
5329         tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, s->T0);
5330         break;
5331     case 0x9f: /* lahf */
5332         if (CODE64(s) && !(s->cpuid_ext3_features & CPUID_EXT3_LAHF_LM))
5333             goto illegal_op;
5334         gen_compute_eflags(s);
5335         /* Note: gen_compute_eflags() only gives the condition codes */
5336         tcg_gen_ori_tl(s->T0, cpu_cc_src, 0x02);
5337         tcg_gen_deposit_tl(cpu_regs[R_EAX], cpu_regs[R_EAX], s->T0, 8, 8);
5338         break;
5339     case 0xf5: /* cmc */
5340         gen_compute_eflags(s);
5341         tcg_gen_xori_tl(cpu_cc_src, cpu_cc_src, CC_C);
5342         break;
5343     case 0xf8: /* clc */
5344         gen_compute_eflags(s);
5345         tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, ~CC_C);
5346         break;
5347     case 0xf9: /* stc */
5348         gen_compute_eflags(s);
5349         tcg_gen_ori_tl(cpu_cc_src, cpu_cc_src, CC_C);
5350         break;
5351     case 0xfc: /* cld */
5352         tcg_gen_movi_i32(s->tmp2_i32, 1);
5353         tcg_gen_st_i32(s->tmp2_i32, cpu_env, offsetof(CPUX86State, df));
5354         break;
5355     case 0xfd: /* std */
5356         tcg_gen_movi_i32(s->tmp2_i32, -1);
5357         tcg_gen_st_i32(s->tmp2_i32, cpu_env, offsetof(CPUX86State, df));
5358         break;
5359 
5360         /************************/
5361         /* bit operations */
5362     case 0x1ba: /* bt/bts/btr/btc Gv, im */
5363         ot = dflag;
5364         modrm = x86_ldub_code(env, s);
5365         op = (modrm >> 3) & 7;
5366         mod = (modrm >> 6) & 3;
5367         rm = (modrm & 7) | REX_B(s);
5368         if (mod != 3) {
5369             s->rip_offset = 1;
5370             gen_lea_modrm(env, s, modrm);
5371             if (!(s->prefix & PREFIX_LOCK)) {
5372                 gen_op_ld_v(s, ot, s->T0, s->A0);
5373             }
5374         } else {
5375             gen_op_mov_v_reg(s, ot, s->T0, rm);
5376         }
5377         /* load shift */
5378         val = x86_ldub_code(env, s);
5379         tcg_gen_movi_tl(s->T1, val);
5380         if (op < 4)
5381             goto unknown_op;
5382         op -= 4;
5383         goto bt_op;
5384     case 0x1a3: /* bt Gv, Ev */
5385         op = 0;
5386         goto do_btx;
5387     case 0x1ab: /* bts */
5388         op = 1;
5389         goto do_btx;
5390     case 0x1b3: /* btr */
5391         op = 2;
5392         goto do_btx;
5393     case 0x1bb: /* btc */
5394         op = 3;
5395     do_btx:
5396         ot = dflag;
5397         modrm = x86_ldub_code(env, s);
5398         reg = ((modrm >> 3) & 7) | REX_R(s);
5399         mod = (modrm >> 6) & 3;
5400         rm = (modrm & 7) | REX_B(s);
5401         gen_op_mov_v_reg(s, MO_32, s->T1, reg);
5402         if (mod != 3) {
5403             AddressParts a = gen_lea_modrm_0(env, s, modrm);
5404             /* specific case: we need to add a displacement */
5405             gen_exts(ot, s->T1);
5406             tcg_gen_sari_tl(s->tmp0, s->T1, 3 + ot);
5407             tcg_gen_shli_tl(s->tmp0, s->tmp0, ot);
5408             tcg_gen_add_tl(s->A0, gen_lea_modrm_1(s, a, false), s->tmp0);
5409             gen_lea_v_seg(s, s->aflag, s->A0, a.def_seg, s->override);
5410             if (!(s->prefix & PREFIX_LOCK)) {
5411                 gen_op_ld_v(s, ot, s->T0, s->A0);
5412             }
5413         } else {
5414             gen_op_mov_v_reg(s, ot, s->T0, rm);
5415         }
5416     bt_op:
5417         tcg_gen_andi_tl(s->T1, s->T1, (1 << (3 + ot)) - 1);
5418         tcg_gen_movi_tl(s->tmp0, 1);
5419         tcg_gen_shl_tl(s->tmp0, s->tmp0, s->T1);
5420         if (s->prefix & PREFIX_LOCK) {
5421             switch (op) {
5422             case 0: /* bt */
5423                 /* Needs no atomic ops; we surpressed the normal
5424                    memory load for LOCK above so do it now.  */
5425                 gen_op_ld_v(s, ot, s->T0, s->A0);
5426                 break;
5427             case 1: /* bts */
5428                 tcg_gen_atomic_fetch_or_tl(s->T0, s->A0, s->tmp0,
5429                                            s->mem_index, ot | MO_LE);
5430                 break;
5431             case 2: /* btr */
5432                 tcg_gen_not_tl(s->tmp0, s->tmp0);
5433                 tcg_gen_atomic_fetch_and_tl(s->T0, s->A0, s->tmp0,
5434                                             s->mem_index, ot | MO_LE);
5435                 break;
5436             default:
5437             case 3: /* btc */
5438                 tcg_gen_atomic_fetch_xor_tl(s->T0, s->A0, s->tmp0,
5439                                             s->mem_index, ot | MO_LE);
5440                 break;
5441             }
5442             tcg_gen_shr_tl(s->tmp4, s->T0, s->T1);
5443         } else {
5444             tcg_gen_shr_tl(s->tmp4, s->T0, s->T1);
5445             switch (op) {
5446             case 0: /* bt */
5447                 /* Data already loaded; nothing to do.  */
5448                 break;
5449             case 1: /* bts */
5450                 tcg_gen_or_tl(s->T0, s->T0, s->tmp0);
5451                 break;
5452             case 2: /* btr */
5453                 tcg_gen_andc_tl(s->T0, s->T0, s->tmp0);
5454                 break;
5455             default:
5456             case 3: /* btc */
5457                 tcg_gen_xor_tl(s->T0, s->T0, s->tmp0);
5458                 break;
5459             }
5460             if (op != 0) {
5461                 if (mod != 3) {
5462                     gen_op_st_v(s, ot, s->T0, s->A0);
5463                 } else {
5464                     gen_op_mov_reg_v(s, ot, rm, s->T0);
5465                 }
5466             }
5467         }
5468 
5469         /* Delay all CC updates until after the store above.  Note that
5470            C is the result of the test, Z is unchanged, and the others
5471            are all undefined.  */
5472         switch (s->cc_op) {
5473         case CC_OP_MULB ... CC_OP_MULQ:
5474         case CC_OP_ADDB ... CC_OP_ADDQ:
5475         case CC_OP_ADCB ... CC_OP_ADCQ:
5476         case CC_OP_SUBB ... CC_OP_SUBQ:
5477         case CC_OP_SBBB ... CC_OP_SBBQ:
5478         case CC_OP_LOGICB ... CC_OP_LOGICQ:
5479         case CC_OP_INCB ... CC_OP_INCQ:
5480         case CC_OP_DECB ... CC_OP_DECQ:
5481         case CC_OP_SHLB ... CC_OP_SHLQ:
5482         case CC_OP_SARB ... CC_OP_SARQ:
5483         case CC_OP_BMILGB ... CC_OP_BMILGQ:
5484             /* Z was going to be computed from the non-zero status of CC_DST.
5485                We can get that same Z value (and the new C value) by leaving
5486                CC_DST alone, setting CC_SRC, and using a CC_OP_SAR of the
5487                same width.  */
5488             tcg_gen_mov_tl(cpu_cc_src, s->tmp4);
5489             set_cc_op(s, ((s->cc_op - CC_OP_MULB) & 3) + CC_OP_SARB);
5490             break;
5491         default:
5492             /* Otherwise, generate EFLAGS and replace the C bit.  */
5493             gen_compute_eflags(s);
5494             tcg_gen_deposit_tl(cpu_cc_src, cpu_cc_src, s->tmp4,
5495                                ctz32(CC_C), 1);
5496             break;
5497         }
5498         break;
5499     case 0x1bc: /* bsf / tzcnt */
5500     case 0x1bd: /* bsr / lzcnt */
5501         ot = dflag;
5502         modrm = x86_ldub_code(env, s);
5503         reg = ((modrm >> 3) & 7) | REX_R(s);
5504         gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 0);
5505         gen_extu(ot, s->T0);
5506 
5507         /* Note that lzcnt and tzcnt are in different extensions.  */
5508         if ((prefixes & PREFIX_REPZ)
5509             && (b & 1
5510                 ? s->cpuid_ext3_features & CPUID_EXT3_ABM
5511                 : s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_BMI1)) {
5512             int size = 8 << ot;
5513             /* For lzcnt/tzcnt, C bit is defined related to the input. */
5514             tcg_gen_mov_tl(cpu_cc_src, s->T0);
5515             if (b & 1) {
5516                 /* For lzcnt, reduce the target_ulong result by the
5517                    number of zeros that we expect to find at the top.  */
5518                 tcg_gen_clzi_tl(s->T0, s->T0, TARGET_LONG_BITS);
5519                 tcg_gen_subi_tl(s->T0, s->T0, TARGET_LONG_BITS - size);
5520             } else {
5521                 /* For tzcnt, a zero input must return the operand size.  */
5522                 tcg_gen_ctzi_tl(s->T0, s->T0, size);
5523             }
5524             /* For lzcnt/tzcnt, Z bit is defined related to the result.  */
5525             gen_op_update1_cc(s);
5526             set_cc_op(s, CC_OP_BMILGB + ot);
5527         } else {
5528             /* For bsr/bsf, only the Z bit is defined and it is related
5529                to the input and not the result.  */
5530             tcg_gen_mov_tl(cpu_cc_dst, s->T0);
5531             set_cc_op(s, CC_OP_LOGICB + ot);
5532 
5533             /* ??? The manual says that the output is undefined when the
5534                input is zero, but real hardware leaves it unchanged, and
5535                real programs appear to depend on that.  Accomplish this
5536                by passing the output as the value to return upon zero.  */
5537             if (b & 1) {
5538                 /* For bsr, return the bit index of the first 1 bit,
5539                    not the count of leading zeros.  */
5540                 tcg_gen_xori_tl(s->T1, cpu_regs[reg], TARGET_LONG_BITS - 1);
5541                 tcg_gen_clz_tl(s->T0, s->T0, s->T1);
5542                 tcg_gen_xori_tl(s->T0, s->T0, TARGET_LONG_BITS - 1);
5543             } else {
5544                 tcg_gen_ctz_tl(s->T0, s->T0, cpu_regs[reg]);
5545             }
5546         }
5547         gen_op_mov_reg_v(s, ot, reg, s->T0);
5548         break;
5549         /************************/
5550         /* bcd */
5551     case 0x27: /* daa */
5552         if (CODE64(s))
5553             goto illegal_op;
5554         gen_update_cc_op(s);
5555         gen_helper_daa(cpu_env);
5556         set_cc_op(s, CC_OP_EFLAGS);
5557         break;
5558     case 0x2f: /* das */
5559         if (CODE64(s))
5560             goto illegal_op;
5561         gen_update_cc_op(s);
5562         gen_helper_das(cpu_env);
5563         set_cc_op(s, CC_OP_EFLAGS);
5564         break;
5565     case 0x37: /* aaa */
5566         if (CODE64(s))
5567             goto illegal_op;
5568         gen_update_cc_op(s);
5569         gen_helper_aaa(cpu_env);
5570         set_cc_op(s, CC_OP_EFLAGS);
5571         break;
5572     case 0x3f: /* aas */
5573         if (CODE64(s))
5574             goto illegal_op;
5575         gen_update_cc_op(s);
5576         gen_helper_aas(cpu_env);
5577         set_cc_op(s, CC_OP_EFLAGS);
5578         break;
5579     case 0xd4: /* aam */
5580         if (CODE64(s))
5581             goto illegal_op;
5582         val = x86_ldub_code(env, s);
5583         if (val == 0) {
5584             gen_exception(s, EXCP00_DIVZ);
5585         } else {
5586             gen_helper_aam(cpu_env, tcg_const_i32(val));
5587             set_cc_op(s, CC_OP_LOGICB);
5588         }
5589         break;
5590     case 0xd5: /* aad */
5591         if (CODE64(s))
5592             goto illegal_op;
5593         val = x86_ldub_code(env, s);
5594         gen_helper_aad(cpu_env, tcg_const_i32(val));
5595         set_cc_op(s, CC_OP_LOGICB);
5596         break;
5597         /************************/
5598         /* misc */
5599     case 0x90: /* nop */
5600         /* XXX: correct lock test for all insn */
5601         if (prefixes & PREFIX_LOCK) {
5602             goto illegal_op;
5603         }
5604         /* If REX_B is set, then this is xchg eax, r8d, not a nop.  */
5605         if (REX_B(s)) {
5606             goto do_xchg_reg_eax;
5607         }
5608         if (prefixes & PREFIX_REPZ) {
5609             gen_update_cc_op(s);
5610             gen_update_eip_cur(s);
5611             gen_helper_pause(cpu_env, cur_insn_len_i32(s));
5612             s->base.is_jmp = DISAS_NORETURN;
5613         }
5614         break;
5615     case 0x9b: /* fwait */
5616         if ((s->flags & (HF_MP_MASK | HF_TS_MASK)) ==
5617             (HF_MP_MASK | HF_TS_MASK)) {
5618             gen_exception(s, EXCP07_PREX);
5619         } else {
5620             gen_helper_fwait(cpu_env);
5621         }
5622         break;
5623     case 0xcc: /* int3 */
5624         gen_interrupt(s, EXCP03_INT3);
5625         break;
5626     case 0xcd: /* int N */
5627         val = x86_ldub_code(env, s);
5628         if (check_vm86_iopl(s)) {
5629             gen_interrupt(s, val);
5630         }
5631         break;
5632     case 0xce: /* into */
5633         if (CODE64(s))
5634             goto illegal_op;
5635         gen_update_cc_op(s);
5636         gen_update_eip_cur(s);
5637         gen_helper_into(cpu_env, cur_insn_len_i32(s));
5638         break;
5639 #ifdef WANT_ICEBP
5640     case 0xf1: /* icebp (undocumented, exits to external debugger) */
5641         gen_svm_check_intercept(s, SVM_EXIT_ICEBP);
5642         gen_debug(s);
5643         break;
5644 #endif
5645     case 0xfa: /* cli */
5646         if (check_iopl(s)) {
5647             gen_reset_eflags(s, IF_MASK);
5648         }
5649         break;
5650     case 0xfb: /* sti */
5651         if (check_iopl(s)) {
5652             gen_set_eflags(s, IF_MASK);
5653             /* interruptions are enabled only the first insn after sti */
5654             gen_update_eip_next(s);
5655             gen_eob_inhibit_irq(s, true);
5656         }
5657         break;
5658     case 0x62: /* bound */
5659         if (CODE64(s))
5660             goto illegal_op;
5661         ot = dflag;
5662         modrm = x86_ldub_code(env, s);
5663         reg = (modrm >> 3) & 7;
5664         mod = (modrm >> 6) & 3;
5665         if (mod == 3)
5666             goto illegal_op;
5667         gen_op_mov_v_reg(s, ot, s->T0, reg);
5668         gen_lea_modrm(env, s, modrm);
5669         tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
5670         if (ot == MO_16) {
5671             gen_helper_boundw(cpu_env, s->A0, s->tmp2_i32);
5672         } else {
5673             gen_helper_boundl(cpu_env, s->A0, s->tmp2_i32);
5674         }
5675         break;
5676     case 0x1c8 ... 0x1cf: /* bswap reg */
5677         reg = (b & 7) | REX_B(s);
5678 #ifdef TARGET_X86_64
5679         if (dflag == MO_64) {
5680             tcg_gen_bswap64_i64(cpu_regs[reg], cpu_regs[reg]);
5681             break;
5682         }
5683 #endif
5684         tcg_gen_bswap32_tl(cpu_regs[reg], cpu_regs[reg], TCG_BSWAP_OZ);
5685         break;
5686     case 0xd6: /* salc */
5687         if (CODE64(s))
5688             goto illegal_op;
5689         gen_compute_eflags_c(s, s->T0);
5690         tcg_gen_neg_tl(s->T0, s->T0);
5691         gen_op_mov_reg_v(s, MO_8, R_EAX, s->T0);
5692         break;
5693     case 0xe0: /* loopnz */
5694     case 0xe1: /* loopz */
5695     case 0xe2: /* loop */
5696     case 0xe3: /* jecxz */
5697         {
5698             TCGLabel *l1, *l2;
5699             int diff = (int8_t)insn_get(env, s, MO_8);
5700 
5701             l1 = gen_new_label();
5702             l2 = gen_new_label();
5703             gen_update_cc_op(s);
5704             b &= 3;
5705             switch(b) {
5706             case 0: /* loopnz */
5707             case 1: /* loopz */
5708                 gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
5709                 gen_op_jz_ecx(s, l2);
5710                 gen_jcc1(s, (JCC_Z << 1) | (b ^ 1), l1);
5711                 break;
5712             case 2: /* loop */
5713                 gen_op_add_reg_im(s, s->aflag, R_ECX, -1);
5714                 gen_op_jnz_ecx(s, l1);
5715                 break;
5716             default:
5717             case 3: /* jcxz */
5718                 gen_op_jz_ecx(s, l1);
5719                 break;
5720             }
5721 
5722             gen_set_label(l2);
5723             gen_jmp_rel_csize(s, 0, 1);
5724 
5725             gen_set_label(l1);
5726             gen_jmp_rel(s, dflag, diff, 0);
5727         }
5728         break;
5729     case 0x130: /* wrmsr */
5730     case 0x132: /* rdmsr */
5731         if (check_cpl0(s)) {
5732             gen_update_cc_op(s);
5733             gen_update_eip_cur(s);
5734             if (b & 2) {
5735                 gen_helper_rdmsr(cpu_env);
5736             } else {
5737                 gen_helper_wrmsr(cpu_env);
5738                 s->base.is_jmp = DISAS_EOB_NEXT;
5739             }
5740         }
5741         break;
5742     case 0x131: /* rdtsc */
5743         gen_update_cc_op(s);
5744         gen_update_eip_cur(s);
5745         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
5746             gen_io_start();
5747             s->base.is_jmp = DISAS_TOO_MANY;
5748         }
5749         gen_helper_rdtsc(cpu_env);
5750         break;
5751     case 0x133: /* rdpmc */
5752         gen_update_cc_op(s);
5753         gen_update_eip_cur(s);
5754         gen_helper_rdpmc(cpu_env);
5755         s->base.is_jmp = DISAS_NORETURN;
5756         break;
5757     case 0x134: /* sysenter */
5758         /* For Intel SYSENTER is valid on 64-bit */
5759         if (CODE64(s) && env->cpuid_vendor1 != CPUID_VENDOR_INTEL_1)
5760             goto illegal_op;
5761         if (!PE(s)) {
5762             gen_exception_gpf(s);
5763         } else {
5764             gen_helper_sysenter(cpu_env);
5765             s->base.is_jmp = DISAS_EOB_ONLY;
5766         }
5767         break;
5768     case 0x135: /* sysexit */
5769         /* For Intel SYSEXIT is valid on 64-bit */
5770         if (CODE64(s) && env->cpuid_vendor1 != CPUID_VENDOR_INTEL_1)
5771             goto illegal_op;
5772         if (!PE(s)) {
5773             gen_exception_gpf(s);
5774         } else {
5775             gen_helper_sysexit(cpu_env, tcg_const_i32(dflag - 1));
5776             s->base.is_jmp = DISAS_EOB_ONLY;
5777         }
5778         break;
5779 #ifdef TARGET_X86_64
5780     case 0x105: /* syscall */
5781         /* XXX: is it usable in real mode ? */
5782         gen_update_cc_op(s);
5783         gen_update_eip_cur(s);
5784         gen_helper_syscall(cpu_env, cur_insn_len_i32(s));
5785         /* TF handling for the syscall insn is different. The TF bit is  checked
5786            after the syscall insn completes. This allows #DB to not be
5787            generated after one has entered CPL0 if TF is set in FMASK.  */
5788         gen_eob_worker(s, false, true);
5789         break;
5790     case 0x107: /* sysret */
5791         if (!PE(s)) {
5792             gen_exception_gpf(s);
5793         } else {
5794             gen_helper_sysret(cpu_env, tcg_const_i32(dflag - 1));
5795             /* condition codes are modified only in long mode */
5796             if (LMA(s)) {
5797                 set_cc_op(s, CC_OP_EFLAGS);
5798             }
5799             /* TF handling for the sysret insn is different. The TF bit is
5800                checked after the sysret insn completes. This allows #DB to be
5801                generated "as if" the syscall insn in userspace has just
5802                completed.  */
5803             gen_eob_worker(s, false, true);
5804         }
5805         break;
5806 #endif
5807     case 0x1a2: /* cpuid */
5808         gen_update_cc_op(s);
5809         gen_update_eip_cur(s);
5810         gen_helper_cpuid(cpu_env);
5811         break;
5812     case 0xf4: /* hlt */
5813         if (check_cpl0(s)) {
5814             gen_update_cc_op(s);
5815             gen_update_eip_cur(s);
5816             gen_helper_hlt(cpu_env, cur_insn_len_i32(s));
5817             s->base.is_jmp = DISAS_NORETURN;
5818         }
5819         break;
5820     case 0x100:
5821         modrm = x86_ldub_code(env, s);
5822         mod = (modrm >> 6) & 3;
5823         op = (modrm >> 3) & 7;
5824         switch(op) {
5825         case 0: /* sldt */
5826             if (!PE(s) || VM86(s))
5827                 goto illegal_op;
5828             if (s->flags & HF_UMIP_MASK && !check_cpl0(s)) {
5829                 break;
5830             }
5831             gen_svm_check_intercept(s, SVM_EXIT_LDTR_READ);
5832             tcg_gen_ld32u_tl(s->T0, cpu_env,
5833                              offsetof(CPUX86State, ldt.selector));
5834             ot = mod == 3 ? dflag : MO_16;
5835             gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 1);
5836             break;
5837         case 2: /* lldt */
5838             if (!PE(s) || VM86(s))
5839                 goto illegal_op;
5840             if (check_cpl0(s)) {
5841                 gen_svm_check_intercept(s, SVM_EXIT_LDTR_WRITE);
5842                 gen_ldst_modrm(env, s, modrm, MO_16, OR_TMP0, 0);
5843                 tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
5844                 gen_helper_lldt(cpu_env, s->tmp2_i32);
5845             }
5846             break;
5847         case 1: /* str */
5848             if (!PE(s) || VM86(s))
5849                 goto illegal_op;
5850             if (s->flags & HF_UMIP_MASK && !check_cpl0(s)) {
5851                 break;
5852             }
5853             gen_svm_check_intercept(s, SVM_EXIT_TR_READ);
5854             tcg_gen_ld32u_tl(s->T0, cpu_env,
5855                              offsetof(CPUX86State, tr.selector));
5856             ot = mod == 3 ? dflag : MO_16;
5857             gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 1);
5858             break;
5859         case 3: /* ltr */
5860             if (!PE(s) || VM86(s))
5861                 goto illegal_op;
5862             if (check_cpl0(s)) {
5863                 gen_svm_check_intercept(s, SVM_EXIT_TR_WRITE);
5864                 gen_ldst_modrm(env, s, modrm, MO_16, OR_TMP0, 0);
5865                 tcg_gen_trunc_tl_i32(s->tmp2_i32, s->T0);
5866                 gen_helper_ltr(cpu_env, s->tmp2_i32);
5867             }
5868             break;
5869         case 4: /* verr */
5870         case 5: /* verw */
5871             if (!PE(s) || VM86(s))
5872                 goto illegal_op;
5873             gen_ldst_modrm(env, s, modrm, MO_16, OR_TMP0, 0);
5874             gen_update_cc_op(s);
5875             if (op == 4) {
5876                 gen_helper_verr(cpu_env, s->T0);
5877             } else {
5878                 gen_helper_verw(cpu_env, s->T0);
5879             }
5880             set_cc_op(s, CC_OP_EFLAGS);
5881             break;
5882         default:
5883             goto unknown_op;
5884         }
5885         break;
5886 
5887     case 0x101:
5888         modrm = x86_ldub_code(env, s);
5889         switch (modrm) {
5890         CASE_MODRM_MEM_OP(0): /* sgdt */
5891             if (s->flags & HF_UMIP_MASK && !check_cpl0(s)) {
5892                 break;
5893             }
5894             gen_svm_check_intercept(s, SVM_EXIT_GDTR_READ);
5895             gen_lea_modrm(env, s, modrm);
5896             tcg_gen_ld32u_tl(s->T0,
5897                              cpu_env, offsetof(CPUX86State, gdt.limit));
5898             gen_op_st_v(s, MO_16, s->T0, s->A0);
5899             gen_add_A0_im(s, 2);
5900             tcg_gen_ld_tl(s->T0, cpu_env, offsetof(CPUX86State, gdt.base));
5901             if (dflag == MO_16) {
5902                 tcg_gen_andi_tl(s->T0, s->T0, 0xffffff);
5903             }
5904             gen_op_st_v(s, CODE64(s) + MO_32, s->T0, s->A0);
5905             break;
5906 
5907         case 0xc8: /* monitor */
5908             if (!(s->cpuid_ext_features & CPUID_EXT_MONITOR) || CPL(s) != 0) {
5909                 goto illegal_op;
5910             }
5911             gen_update_cc_op(s);
5912             gen_update_eip_cur(s);
5913             tcg_gen_mov_tl(s->A0, cpu_regs[R_EAX]);
5914             gen_extu(s->aflag, s->A0);
5915             gen_add_A0_ds_seg(s);
5916             gen_helper_monitor(cpu_env, s->A0);
5917             break;
5918 
5919         case 0xc9: /* mwait */
5920             if (!(s->cpuid_ext_features & CPUID_EXT_MONITOR) || CPL(s) != 0) {
5921                 goto illegal_op;
5922             }
5923             gen_update_cc_op(s);
5924             gen_update_eip_cur(s);
5925             gen_helper_mwait(cpu_env, cur_insn_len_i32(s));
5926             s->base.is_jmp = DISAS_NORETURN;
5927             break;
5928 
5929         case 0xca: /* clac */
5930             if (!(s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_SMAP)
5931                 || CPL(s) != 0) {
5932                 goto illegal_op;
5933             }
5934             gen_reset_eflags(s, AC_MASK);
5935             s->base.is_jmp = DISAS_EOB_NEXT;
5936             break;
5937 
5938         case 0xcb: /* stac */
5939             if (!(s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_SMAP)
5940                 || CPL(s) != 0) {
5941                 goto illegal_op;
5942             }
5943             gen_set_eflags(s, AC_MASK);
5944             s->base.is_jmp = DISAS_EOB_NEXT;
5945             break;
5946 
5947         CASE_MODRM_MEM_OP(1): /* sidt */
5948             if (s->flags & HF_UMIP_MASK && !check_cpl0(s)) {
5949                 break;
5950             }
5951             gen_svm_check_intercept(s, SVM_EXIT_IDTR_READ);
5952             gen_lea_modrm(env, s, modrm);
5953             tcg_gen_ld32u_tl(s->T0, cpu_env, offsetof(CPUX86State, idt.limit));
5954             gen_op_st_v(s, MO_16, s->T0, s->A0);
5955             gen_add_A0_im(s, 2);
5956             tcg_gen_ld_tl(s->T0, cpu_env, offsetof(CPUX86State, idt.base));
5957             if (dflag == MO_16) {
5958                 tcg_gen_andi_tl(s->T0, s->T0, 0xffffff);
5959             }
5960             gen_op_st_v(s, CODE64(s) + MO_32, s->T0, s->A0);
5961             break;
5962 
5963         case 0xd0: /* xgetbv */
5964             if ((s->cpuid_ext_features & CPUID_EXT_XSAVE) == 0
5965                 || (s->prefix & (PREFIX_LOCK | PREFIX_DATA
5966                                  | PREFIX_REPZ | PREFIX_REPNZ))) {
5967                 goto illegal_op;
5968             }
5969             tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_ECX]);
5970             gen_helper_xgetbv(s->tmp1_i64, cpu_env, s->tmp2_i32);
5971             tcg_gen_extr_i64_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], s->tmp1_i64);
5972             break;
5973 
5974         case 0xd1: /* xsetbv */
5975             if ((s->cpuid_ext_features & CPUID_EXT_XSAVE) == 0
5976                 || (s->prefix & (PREFIX_LOCK | PREFIX_DATA
5977                                  | PREFIX_REPZ | PREFIX_REPNZ))) {
5978                 goto illegal_op;
5979             }
5980             if (!check_cpl0(s)) {
5981                 break;
5982             }
5983             tcg_gen_concat_tl_i64(s->tmp1_i64, cpu_regs[R_EAX],
5984                                   cpu_regs[R_EDX]);
5985             tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_ECX]);
5986             gen_helper_xsetbv(cpu_env, s->tmp2_i32, s->tmp1_i64);
5987             /* End TB because translation flags may change.  */
5988             s->base.is_jmp = DISAS_EOB_NEXT;
5989             break;
5990 
5991         case 0xd8: /* VMRUN */
5992             if (!SVME(s) || !PE(s)) {
5993                 goto illegal_op;
5994             }
5995             if (!check_cpl0(s)) {
5996                 break;
5997             }
5998             gen_update_cc_op(s);
5999             gen_update_eip_cur(s);
6000             gen_helper_vmrun(cpu_env, tcg_const_i32(s->aflag - 1),
6001                              cur_insn_len_i32(s));
6002             tcg_gen_exit_tb(NULL, 0);
6003             s->base.is_jmp = DISAS_NORETURN;
6004             break;
6005 
6006         case 0xd9: /* VMMCALL */
6007             if (!SVME(s)) {
6008                 goto illegal_op;
6009             }
6010             gen_update_cc_op(s);
6011             gen_update_eip_cur(s);
6012             gen_helper_vmmcall(cpu_env);
6013             break;
6014 
6015         case 0xda: /* VMLOAD */
6016             if (!SVME(s) || !PE(s)) {
6017                 goto illegal_op;
6018             }
6019             if (!check_cpl0(s)) {
6020                 break;
6021             }
6022             gen_update_cc_op(s);
6023             gen_update_eip_cur(s);
6024             gen_helper_vmload(cpu_env, tcg_const_i32(s->aflag - 1));
6025             break;
6026 
6027         case 0xdb: /* VMSAVE */
6028             if (!SVME(s) || !PE(s)) {
6029                 goto illegal_op;
6030             }
6031             if (!check_cpl0(s)) {
6032                 break;
6033             }
6034             gen_update_cc_op(s);
6035             gen_update_eip_cur(s);
6036             gen_helper_vmsave(cpu_env, tcg_const_i32(s->aflag - 1));
6037             break;
6038 
6039         case 0xdc: /* STGI */
6040             if ((!SVME(s) && !(s->cpuid_ext3_features & CPUID_EXT3_SKINIT))
6041                 || !PE(s)) {
6042                 goto illegal_op;
6043             }
6044             if (!check_cpl0(s)) {
6045                 break;
6046             }
6047             gen_update_cc_op(s);
6048             gen_helper_stgi(cpu_env);
6049             s->base.is_jmp = DISAS_EOB_NEXT;
6050             break;
6051 
6052         case 0xdd: /* CLGI */
6053             if (!SVME(s) || !PE(s)) {
6054                 goto illegal_op;
6055             }
6056             if (!check_cpl0(s)) {
6057                 break;
6058             }
6059             gen_update_cc_op(s);
6060             gen_update_eip_cur(s);
6061             gen_helper_clgi(cpu_env);
6062             break;
6063 
6064         case 0xde: /* SKINIT */
6065             if ((!SVME(s) && !(s->cpuid_ext3_features & CPUID_EXT3_SKINIT))
6066                 || !PE(s)) {
6067                 goto illegal_op;
6068             }
6069             gen_svm_check_intercept(s, SVM_EXIT_SKINIT);
6070             /* If not intercepted, not implemented -- raise #UD. */
6071             goto illegal_op;
6072 
6073         case 0xdf: /* INVLPGA */
6074             if (!SVME(s) || !PE(s)) {
6075                 goto illegal_op;
6076             }
6077             if (!check_cpl0(s)) {
6078                 break;
6079             }
6080             gen_svm_check_intercept(s, SVM_EXIT_INVLPGA);
6081             if (s->aflag == MO_64) {
6082                 tcg_gen_mov_tl(s->A0, cpu_regs[R_EAX]);
6083             } else {
6084                 tcg_gen_ext32u_tl(s->A0, cpu_regs[R_EAX]);
6085             }
6086             gen_helper_flush_page(cpu_env, s->A0);
6087             s->base.is_jmp = DISAS_EOB_NEXT;
6088             break;
6089 
6090         CASE_MODRM_MEM_OP(2): /* lgdt */
6091             if (!check_cpl0(s)) {
6092                 break;
6093             }
6094             gen_svm_check_intercept(s, SVM_EXIT_GDTR_WRITE);
6095             gen_lea_modrm(env, s, modrm);
6096             gen_op_ld_v(s, MO_16, s->T1, s->A0);
6097             gen_add_A0_im(s, 2);
6098             gen_op_ld_v(s, CODE64(s) + MO_32, s->T0, s->A0);
6099             if (dflag == MO_16) {
6100                 tcg_gen_andi_tl(s->T0, s->T0, 0xffffff);
6101             }
6102             tcg_gen_st_tl(s->T0, cpu_env, offsetof(CPUX86State, gdt.base));
6103             tcg_gen_st32_tl(s->T1, cpu_env, offsetof(CPUX86State, gdt.limit));
6104             break;
6105 
6106         CASE_MODRM_MEM_OP(3): /* lidt */
6107             if (!check_cpl0(s)) {
6108                 break;
6109             }
6110             gen_svm_check_intercept(s, SVM_EXIT_IDTR_WRITE);
6111             gen_lea_modrm(env, s, modrm);
6112             gen_op_ld_v(s, MO_16, s->T1, s->A0);
6113             gen_add_A0_im(s, 2);
6114             gen_op_ld_v(s, CODE64(s) + MO_32, s->T0, s->A0);
6115             if (dflag == MO_16) {
6116                 tcg_gen_andi_tl(s->T0, s->T0, 0xffffff);
6117             }
6118             tcg_gen_st_tl(s->T0, cpu_env, offsetof(CPUX86State, idt.base));
6119             tcg_gen_st32_tl(s->T1, cpu_env, offsetof(CPUX86State, idt.limit));
6120             break;
6121 
6122         CASE_MODRM_OP(4): /* smsw */
6123             if (s->flags & HF_UMIP_MASK && !check_cpl0(s)) {
6124                 break;
6125             }
6126             gen_svm_check_intercept(s, SVM_EXIT_READ_CR0);
6127             tcg_gen_ld_tl(s->T0, cpu_env, offsetof(CPUX86State, cr[0]));
6128             /*
6129              * In 32-bit mode, the higher 16 bits of the destination
6130              * register are undefined.  In practice CR0[31:0] is stored
6131              * just like in 64-bit mode.
6132              */
6133             mod = (modrm >> 6) & 3;
6134             ot = (mod != 3 ? MO_16 : s->dflag);
6135             gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 1);
6136             break;
6137         case 0xee: /* rdpkru */
6138             if (prefixes & PREFIX_LOCK) {
6139                 goto illegal_op;
6140             }
6141             tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_ECX]);
6142             gen_helper_rdpkru(s->tmp1_i64, cpu_env, s->tmp2_i32);
6143             tcg_gen_extr_i64_tl(cpu_regs[R_EAX], cpu_regs[R_EDX], s->tmp1_i64);
6144             break;
6145         case 0xef: /* wrpkru */
6146             if (prefixes & PREFIX_LOCK) {
6147                 goto illegal_op;
6148             }
6149             tcg_gen_concat_tl_i64(s->tmp1_i64, cpu_regs[R_EAX],
6150                                   cpu_regs[R_EDX]);
6151             tcg_gen_trunc_tl_i32(s->tmp2_i32, cpu_regs[R_ECX]);
6152             gen_helper_wrpkru(cpu_env, s->tmp2_i32, s->tmp1_i64);
6153             break;
6154 
6155         CASE_MODRM_OP(6): /* lmsw */
6156             if (!check_cpl0(s)) {
6157                 break;
6158             }
6159             gen_svm_check_intercept(s, SVM_EXIT_WRITE_CR0);
6160             gen_ldst_modrm(env, s, modrm, MO_16, OR_TMP0, 0);
6161             /*
6162              * Only the 4 lower bits of CR0 are modified.
6163              * PE cannot be set to zero if already set to one.
6164              */
6165             tcg_gen_ld_tl(s->T1, cpu_env, offsetof(CPUX86State, cr[0]));
6166             tcg_gen_andi_tl(s->T0, s->T0, 0xf);
6167             tcg_gen_andi_tl(s->T1, s->T1, ~0xe);
6168             tcg_gen_or_tl(s->T0, s->T0, s->T1);
6169             gen_helper_write_crN(cpu_env, tcg_constant_i32(0), s->T0);
6170             s->base.is_jmp = DISAS_EOB_NEXT;
6171             break;
6172 
6173         CASE_MODRM_MEM_OP(7): /* invlpg */
6174             if (!check_cpl0(s)) {
6175                 break;
6176             }
6177             gen_svm_check_intercept(s, SVM_EXIT_INVLPG);
6178             gen_lea_modrm(env, s, modrm);
6179             gen_helper_flush_page(cpu_env, s->A0);
6180             s->base.is_jmp = DISAS_EOB_NEXT;
6181             break;
6182 
6183         case 0xf8: /* swapgs */
6184 #ifdef TARGET_X86_64
6185             if (CODE64(s)) {
6186                 if (check_cpl0(s)) {
6187                     tcg_gen_mov_tl(s->T0, cpu_seg_base[R_GS]);
6188                     tcg_gen_ld_tl(cpu_seg_base[R_GS], cpu_env,
6189                                   offsetof(CPUX86State, kernelgsbase));
6190                     tcg_gen_st_tl(s->T0, cpu_env,
6191                                   offsetof(CPUX86State, kernelgsbase));
6192                 }
6193                 break;
6194             }
6195 #endif
6196             goto illegal_op;
6197 
6198         case 0xf9: /* rdtscp */
6199             if (!(s->cpuid_ext2_features & CPUID_EXT2_RDTSCP)) {
6200                 goto illegal_op;
6201             }
6202             gen_update_cc_op(s);
6203             gen_update_eip_cur(s);
6204             if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
6205                 gen_io_start();
6206                 s->base.is_jmp = DISAS_TOO_MANY;
6207             }
6208             gen_helper_rdtscp(cpu_env);
6209             break;
6210 
6211         default:
6212             goto unknown_op;
6213         }
6214         break;
6215 
6216     case 0x108: /* invd */
6217     case 0x109: /* wbinvd */
6218         if (check_cpl0(s)) {
6219             gen_svm_check_intercept(s, (b & 2) ? SVM_EXIT_INVD : SVM_EXIT_WBINVD);
6220             /* nothing to do */
6221         }
6222         break;
6223     case 0x63: /* arpl or movslS (x86_64) */
6224 #ifdef TARGET_X86_64
6225         if (CODE64(s)) {
6226             int d_ot;
6227             /* d_ot is the size of destination */
6228             d_ot = dflag;
6229 
6230             modrm = x86_ldub_code(env, s);
6231             reg = ((modrm >> 3) & 7) | REX_R(s);
6232             mod = (modrm >> 6) & 3;
6233             rm = (modrm & 7) | REX_B(s);
6234 
6235             if (mod == 3) {
6236                 gen_op_mov_v_reg(s, MO_32, s->T0, rm);
6237                 /* sign extend */
6238                 if (d_ot == MO_64) {
6239                     tcg_gen_ext32s_tl(s->T0, s->T0);
6240                 }
6241                 gen_op_mov_reg_v(s, d_ot, reg, s->T0);
6242             } else {
6243                 gen_lea_modrm(env, s, modrm);
6244                 gen_op_ld_v(s, MO_32 | MO_SIGN, s->T0, s->A0);
6245                 gen_op_mov_reg_v(s, d_ot, reg, s->T0);
6246             }
6247         } else
6248 #endif
6249         {
6250             TCGLabel *label1;
6251             TCGv t0, t1, t2, a0;
6252 
6253             if (!PE(s) || VM86(s))
6254                 goto illegal_op;
6255             t0 = tcg_temp_local_new();
6256             t1 = tcg_temp_local_new();
6257             t2 = tcg_temp_local_new();
6258             ot = MO_16;
6259             modrm = x86_ldub_code(env, s);
6260             reg = (modrm >> 3) & 7;
6261             mod = (modrm >> 6) & 3;
6262             rm = modrm & 7;
6263             if (mod != 3) {
6264                 gen_lea_modrm(env, s, modrm);
6265                 gen_op_ld_v(s, ot, t0, s->A0);
6266                 a0 = tcg_temp_local_new();
6267                 tcg_gen_mov_tl(a0, s->A0);
6268             } else {
6269                 gen_op_mov_v_reg(s, ot, t0, rm);
6270                 a0 = NULL;
6271             }
6272             gen_op_mov_v_reg(s, ot, t1, reg);
6273             tcg_gen_andi_tl(s->tmp0, t0, 3);
6274             tcg_gen_andi_tl(t1, t1, 3);
6275             tcg_gen_movi_tl(t2, 0);
6276             label1 = gen_new_label();
6277             tcg_gen_brcond_tl(TCG_COND_GE, s->tmp0, t1, label1);
6278             tcg_gen_andi_tl(t0, t0, ~3);
6279             tcg_gen_or_tl(t0, t0, t1);
6280             tcg_gen_movi_tl(t2, CC_Z);
6281             gen_set_label(label1);
6282             if (mod != 3) {
6283                 gen_op_st_v(s, ot, t0, a0);
6284                 tcg_temp_free(a0);
6285            } else {
6286                 gen_op_mov_reg_v(s, ot, rm, t0);
6287             }
6288             gen_compute_eflags(s);
6289             tcg_gen_andi_tl(cpu_cc_src, cpu_cc_src, ~CC_Z);
6290             tcg_gen_or_tl(cpu_cc_src, cpu_cc_src, t2);
6291             tcg_temp_free(t0);
6292             tcg_temp_free(t1);
6293             tcg_temp_free(t2);
6294         }
6295         break;
6296     case 0x102: /* lar */
6297     case 0x103: /* lsl */
6298         {
6299             TCGLabel *label1;
6300             TCGv t0;
6301             if (!PE(s) || VM86(s))
6302                 goto illegal_op;
6303             ot = dflag != MO_16 ? MO_32 : MO_16;
6304             modrm = x86_ldub_code(env, s);
6305             reg = ((modrm >> 3) & 7) | REX_R(s);
6306             gen_ldst_modrm(env, s, modrm, MO_16, OR_TMP0, 0);
6307             t0 = tcg_temp_local_new();
6308             gen_update_cc_op(s);
6309             if (b == 0x102) {
6310                 gen_helper_lar(t0, cpu_env, s->T0);
6311             } else {
6312                 gen_helper_lsl(t0, cpu_env, s->T0);
6313             }
6314             tcg_gen_andi_tl(s->tmp0, cpu_cc_src, CC_Z);
6315             label1 = gen_new_label();
6316             tcg_gen_brcondi_tl(TCG_COND_EQ, s->tmp0, 0, label1);
6317             gen_op_mov_reg_v(s, ot, reg, t0);
6318             gen_set_label(label1);
6319             set_cc_op(s, CC_OP_EFLAGS);
6320             tcg_temp_free(t0);
6321         }
6322         break;
6323     case 0x118:
6324         modrm = x86_ldub_code(env, s);
6325         mod = (modrm >> 6) & 3;
6326         op = (modrm >> 3) & 7;
6327         switch(op) {
6328         case 0: /* prefetchnta */
6329         case 1: /* prefetchnt0 */
6330         case 2: /* prefetchnt0 */
6331         case 3: /* prefetchnt0 */
6332             if (mod == 3)
6333                 goto illegal_op;
6334             gen_nop_modrm(env, s, modrm);
6335             /* nothing more to do */
6336             break;
6337         default: /* nop (multi byte) */
6338             gen_nop_modrm(env, s, modrm);
6339             break;
6340         }
6341         break;
6342     case 0x11a:
6343         modrm = x86_ldub_code(env, s);
6344         if (s->flags & HF_MPX_EN_MASK) {
6345             mod = (modrm >> 6) & 3;
6346             reg = ((modrm >> 3) & 7) | REX_R(s);
6347             if (prefixes & PREFIX_REPZ) {
6348                 /* bndcl */
6349                 if (reg >= 4
6350                     || (prefixes & PREFIX_LOCK)
6351                     || s->aflag == MO_16) {
6352                     goto illegal_op;
6353                 }
6354                 gen_bndck(env, s, modrm, TCG_COND_LTU, cpu_bndl[reg]);
6355             } else if (prefixes & PREFIX_REPNZ) {
6356                 /* bndcu */
6357                 if (reg >= 4
6358                     || (prefixes & PREFIX_LOCK)
6359                     || s->aflag == MO_16) {
6360                     goto illegal_op;
6361                 }
6362                 TCGv_i64 notu = tcg_temp_new_i64();
6363                 tcg_gen_not_i64(notu, cpu_bndu[reg]);
6364                 gen_bndck(env, s, modrm, TCG_COND_GTU, notu);
6365                 tcg_temp_free_i64(notu);
6366             } else if (prefixes & PREFIX_DATA) {
6367                 /* bndmov -- from reg/mem */
6368                 if (reg >= 4 || s->aflag == MO_16) {
6369                     goto illegal_op;
6370                 }
6371                 if (mod == 3) {
6372                     int reg2 = (modrm & 7) | REX_B(s);
6373                     if (reg2 >= 4 || (prefixes & PREFIX_LOCK)) {
6374                         goto illegal_op;
6375                     }
6376                     if (s->flags & HF_MPX_IU_MASK) {
6377                         tcg_gen_mov_i64(cpu_bndl[reg], cpu_bndl[reg2]);
6378                         tcg_gen_mov_i64(cpu_bndu[reg], cpu_bndu[reg2]);
6379                     }
6380                 } else {
6381                     gen_lea_modrm(env, s, modrm);
6382                     if (CODE64(s)) {
6383                         tcg_gen_qemu_ld_i64(cpu_bndl[reg], s->A0,
6384                                             s->mem_index, MO_LEUQ);
6385                         tcg_gen_addi_tl(s->A0, s->A0, 8);
6386                         tcg_gen_qemu_ld_i64(cpu_bndu[reg], s->A0,
6387                                             s->mem_index, MO_LEUQ);
6388                     } else {
6389                         tcg_gen_qemu_ld_i64(cpu_bndl[reg], s->A0,
6390                                             s->mem_index, MO_LEUL);
6391                         tcg_gen_addi_tl(s->A0, s->A0, 4);
6392                         tcg_gen_qemu_ld_i64(cpu_bndu[reg], s->A0,
6393                                             s->mem_index, MO_LEUL);
6394                     }
6395                     /* bnd registers are now in-use */
6396                     gen_set_hflag(s, HF_MPX_IU_MASK);
6397                 }
6398             } else if (mod != 3) {
6399                 /* bndldx */
6400                 AddressParts a = gen_lea_modrm_0(env, s, modrm);
6401                 if (reg >= 4
6402                     || (prefixes & PREFIX_LOCK)
6403                     || s->aflag == MO_16
6404                     || a.base < -1) {
6405                     goto illegal_op;
6406                 }
6407                 if (a.base >= 0) {
6408                     tcg_gen_addi_tl(s->A0, cpu_regs[a.base], a.disp);
6409                 } else {
6410                     tcg_gen_movi_tl(s->A0, 0);
6411                 }
6412                 gen_lea_v_seg(s, s->aflag, s->A0, a.def_seg, s->override);
6413                 if (a.index >= 0) {
6414                     tcg_gen_mov_tl(s->T0, cpu_regs[a.index]);
6415                 } else {
6416                     tcg_gen_movi_tl(s->T0, 0);
6417                 }
6418                 if (CODE64(s)) {
6419                     gen_helper_bndldx64(cpu_bndl[reg], cpu_env, s->A0, s->T0);
6420                     tcg_gen_ld_i64(cpu_bndu[reg], cpu_env,
6421                                    offsetof(CPUX86State, mmx_t0.MMX_Q(0)));
6422                 } else {
6423                     gen_helper_bndldx32(cpu_bndu[reg], cpu_env, s->A0, s->T0);
6424                     tcg_gen_ext32u_i64(cpu_bndl[reg], cpu_bndu[reg]);
6425                     tcg_gen_shri_i64(cpu_bndu[reg], cpu_bndu[reg], 32);
6426                 }
6427                 gen_set_hflag(s, HF_MPX_IU_MASK);
6428             }
6429         }
6430         gen_nop_modrm(env, s, modrm);
6431         break;
6432     case 0x11b:
6433         modrm = x86_ldub_code(env, s);
6434         if (s->flags & HF_MPX_EN_MASK) {
6435             mod = (modrm >> 6) & 3;
6436             reg = ((modrm >> 3) & 7) | REX_R(s);
6437             if (mod != 3 && (prefixes & PREFIX_REPZ)) {
6438                 /* bndmk */
6439                 if (reg >= 4
6440                     || (prefixes & PREFIX_LOCK)
6441                     || s->aflag == MO_16) {
6442                     goto illegal_op;
6443                 }
6444                 AddressParts a = gen_lea_modrm_0(env, s, modrm);
6445                 if (a.base >= 0) {
6446                     tcg_gen_extu_tl_i64(cpu_bndl[reg], cpu_regs[a.base]);
6447                     if (!CODE64(s)) {
6448                         tcg_gen_ext32u_i64(cpu_bndl[reg], cpu_bndl[reg]);
6449                     }
6450                 } else if (a.base == -1) {
6451                     /* no base register has lower bound of 0 */
6452                     tcg_gen_movi_i64(cpu_bndl[reg], 0);
6453                 } else {
6454                     /* rip-relative generates #ud */
6455                     goto illegal_op;
6456                 }
6457                 tcg_gen_not_tl(s->A0, gen_lea_modrm_1(s, a, false));
6458                 if (!CODE64(s)) {
6459                     tcg_gen_ext32u_tl(s->A0, s->A0);
6460                 }
6461                 tcg_gen_extu_tl_i64(cpu_bndu[reg], s->A0);
6462                 /* bnd registers are now in-use */
6463                 gen_set_hflag(s, HF_MPX_IU_MASK);
6464                 break;
6465             } else if (prefixes & PREFIX_REPNZ) {
6466                 /* bndcn */
6467                 if (reg >= 4
6468                     || (prefixes & PREFIX_LOCK)
6469                     || s->aflag == MO_16) {
6470                     goto illegal_op;
6471                 }
6472                 gen_bndck(env, s, modrm, TCG_COND_GTU, cpu_bndu[reg]);
6473             } else if (prefixes & PREFIX_DATA) {
6474                 /* bndmov -- to reg/mem */
6475                 if (reg >= 4 || s->aflag == MO_16) {
6476                     goto illegal_op;
6477                 }
6478                 if (mod == 3) {
6479                     int reg2 = (modrm & 7) | REX_B(s);
6480                     if (reg2 >= 4 || (prefixes & PREFIX_LOCK)) {
6481                         goto illegal_op;
6482                     }
6483                     if (s->flags & HF_MPX_IU_MASK) {
6484                         tcg_gen_mov_i64(cpu_bndl[reg2], cpu_bndl[reg]);
6485                         tcg_gen_mov_i64(cpu_bndu[reg2], cpu_bndu[reg]);
6486                     }
6487                 } else {
6488                     gen_lea_modrm(env, s, modrm);
6489                     if (CODE64(s)) {
6490                         tcg_gen_qemu_st_i64(cpu_bndl[reg], s->A0,
6491                                             s->mem_index, MO_LEUQ);
6492                         tcg_gen_addi_tl(s->A0, s->A0, 8);
6493                         tcg_gen_qemu_st_i64(cpu_bndu[reg], s->A0,
6494                                             s->mem_index, MO_LEUQ);
6495                     } else {
6496                         tcg_gen_qemu_st_i64(cpu_bndl[reg], s->A0,
6497                                             s->mem_index, MO_LEUL);
6498                         tcg_gen_addi_tl(s->A0, s->A0, 4);
6499                         tcg_gen_qemu_st_i64(cpu_bndu[reg], s->A0,
6500                                             s->mem_index, MO_LEUL);
6501                     }
6502                 }
6503             } else if (mod != 3) {
6504                 /* bndstx */
6505                 AddressParts a = gen_lea_modrm_0(env, s, modrm);
6506                 if (reg >= 4
6507                     || (prefixes & PREFIX_LOCK)
6508                     || s->aflag == MO_16
6509                     || a.base < -1) {
6510                     goto illegal_op;
6511                 }
6512                 if (a.base >= 0) {
6513                     tcg_gen_addi_tl(s->A0, cpu_regs[a.base], a.disp);
6514                 } else {
6515                     tcg_gen_movi_tl(s->A0, 0);
6516                 }
6517                 gen_lea_v_seg(s, s->aflag, s->A0, a.def_seg, s->override);
6518                 if (a.index >= 0) {
6519                     tcg_gen_mov_tl(s->T0, cpu_regs[a.index]);
6520                 } else {
6521                     tcg_gen_movi_tl(s->T0, 0);
6522                 }
6523                 if (CODE64(s)) {
6524                     gen_helper_bndstx64(cpu_env, s->A0, s->T0,
6525                                         cpu_bndl[reg], cpu_bndu[reg]);
6526                 } else {
6527                     gen_helper_bndstx32(cpu_env, s->A0, s->T0,
6528                                         cpu_bndl[reg], cpu_bndu[reg]);
6529                 }
6530             }
6531         }
6532         gen_nop_modrm(env, s, modrm);
6533         break;
6534     case 0x119: case 0x11c ... 0x11f: /* nop (multi byte) */
6535         modrm = x86_ldub_code(env, s);
6536         gen_nop_modrm(env, s, modrm);
6537         break;
6538 
6539     case 0x120: /* mov reg, crN */
6540     case 0x122: /* mov crN, reg */
6541         if (!check_cpl0(s)) {
6542             break;
6543         }
6544         modrm = x86_ldub_code(env, s);
6545         /*
6546          * Ignore the mod bits (assume (modrm&0xc0)==0xc0).
6547          * AMD documentation (24594.pdf) and testing of Intel 386 and 486
6548          * processors all show that the mod bits are assumed to be 1's,
6549          * regardless of actual values.
6550          */
6551         rm = (modrm & 7) | REX_B(s);
6552         reg = ((modrm >> 3) & 7) | REX_R(s);
6553         switch (reg) {
6554         case 0:
6555             if ((prefixes & PREFIX_LOCK) &&
6556                 (s->cpuid_ext3_features & CPUID_EXT3_CR8LEG)) {
6557                 reg = 8;
6558             }
6559             break;
6560         case 2:
6561         case 3:
6562         case 4:
6563         case 8:
6564             break;
6565         default:
6566             goto unknown_op;
6567         }
6568         ot  = (CODE64(s) ? MO_64 : MO_32);
6569 
6570         if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
6571             gen_io_start();
6572             s->base.is_jmp = DISAS_TOO_MANY;
6573         }
6574         if (b & 2) {
6575             gen_svm_check_intercept(s, SVM_EXIT_WRITE_CR0 + reg);
6576             gen_op_mov_v_reg(s, ot, s->T0, rm);
6577             gen_helper_write_crN(cpu_env, tcg_constant_i32(reg), s->T0);
6578             s->base.is_jmp = DISAS_EOB_NEXT;
6579         } else {
6580             gen_svm_check_intercept(s, SVM_EXIT_READ_CR0 + reg);
6581             gen_helper_read_crN(s->T0, cpu_env, tcg_constant_i32(reg));
6582             gen_op_mov_reg_v(s, ot, rm, s->T0);
6583         }
6584         break;
6585 
6586     case 0x121: /* mov reg, drN */
6587     case 0x123: /* mov drN, reg */
6588         if (check_cpl0(s)) {
6589             modrm = x86_ldub_code(env, s);
6590             /* Ignore the mod bits (assume (modrm&0xc0)==0xc0).
6591              * AMD documentation (24594.pdf) and testing of
6592              * intel 386 and 486 processors all show that the mod bits
6593              * are assumed to be 1's, regardless of actual values.
6594              */
6595             rm = (modrm & 7) | REX_B(s);
6596             reg = ((modrm >> 3) & 7) | REX_R(s);
6597             if (CODE64(s))
6598                 ot = MO_64;
6599             else
6600                 ot = MO_32;
6601             if (reg >= 8) {
6602                 goto illegal_op;
6603             }
6604             if (b & 2) {
6605                 gen_svm_check_intercept(s, SVM_EXIT_WRITE_DR0 + reg);
6606                 gen_op_mov_v_reg(s, ot, s->T0, rm);
6607                 tcg_gen_movi_i32(s->tmp2_i32, reg);
6608                 gen_helper_set_dr(cpu_env, s->tmp2_i32, s->T0);
6609                 s->base.is_jmp = DISAS_EOB_NEXT;
6610             } else {
6611                 gen_svm_check_intercept(s, SVM_EXIT_READ_DR0 + reg);
6612                 tcg_gen_movi_i32(s->tmp2_i32, reg);
6613                 gen_helper_get_dr(s->T0, cpu_env, s->tmp2_i32);
6614                 gen_op_mov_reg_v(s, ot, rm, s->T0);
6615             }
6616         }
6617         break;
6618     case 0x106: /* clts */
6619         if (check_cpl0(s)) {
6620             gen_svm_check_intercept(s, SVM_EXIT_WRITE_CR0);
6621             gen_helper_clts(cpu_env);
6622             /* abort block because static cpu state changed */
6623             s->base.is_jmp = DISAS_EOB_NEXT;
6624         }
6625         break;
6626     /* MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4 support */
6627     case 0x1c3: /* MOVNTI reg, mem */
6628         if (!(s->cpuid_features & CPUID_SSE2))
6629             goto illegal_op;
6630         ot = mo_64_32(dflag);
6631         modrm = x86_ldub_code(env, s);
6632         mod = (modrm >> 6) & 3;
6633         if (mod == 3)
6634             goto illegal_op;
6635         reg = ((modrm >> 3) & 7) | REX_R(s);
6636         /* generate a generic store */
6637         gen_ldst_modrm(env, s, modrm, ot, reg, 1);
6638         break;
6639     case 0x1ae:
6640         modrm = x86_ldub_code(env, s);
6641         switch (modrm) {
6642         CASE_MODRM_MEM_OP(0): /* fxsave */
6643             if (!(s->cpuid_features & CPUID_FXSR)
6644                 || (prefixes & PREFIX_LOCK)) {
6645                 goto illegal_op;
6646             }
6647             if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
6648                 gen_exception(s, EXCP07_PREX);
6649                 break;
6650             }
6651             gen_lea_modrm(env, s, modrm);
6652             gen_helper_fxsave(cpu_env, s->A0);
6653             break;
6654 
6655         CASE_MODRM_MEM_OP(1): /* fxrstor */
6656             if (!(s->cpuid_features & CPUID_FXSR)
6657                 || (prefixes & PREFIX_LOCK)) {
6658                 goto illegal_op;
6659             }
6660             if ((s->flags & HF_EM_MASK) || (s->flags & HF_TS_MASK)) {
6661                 gen_exception(s, EXCP07_PREX);
6662                 break;
6663             }
6664             gen_lea_modrm(env, s, modrm);
6665             gen_helper_fxrstor(cpu_env, s->A0);
6666             break;
6667 
6668         CASE_MODRM_MEM_OP(2): /* ldmxcsr */
6669             if ((s->flags & HF_EM_MASK) || !(s->flags & HF_OSFXSR_MASK)) {
6670                 goto illegal_op;
6671             }
6672             if (s->flags & HF_TS_MASK) {
6673                 gen_exception(s, EXCP07_PREX);
6674                 break;
6675             }
6676             gen_lea_modrm(env, s, modrm);
6677             tcg_gen_qemu_ld_i32(s->tmp2_i32, s->A0, s->mem_index, MO_LEUL);
6678             gen_helper_ldmxcsr(cpu_env, s->tmp2_i32);
6679             break;
6680 
6681         CASE_MODRM_MEM_OP(3): /* stmxcsr */
6682             if ((s->flags & HF_EM_MASK) || !(s->flags & HF_OSFXSR_MASK)) {
6683                 goto illegal_op;
6684             }
6685             if (s->flags & HF_TS_MASK) {
6686                 gen_exception(s, EXCP07_PREX);
6687                 break;
6688             }
6689             gen_helper_update_mxcsr(cpu_env);
6690             gen_lea_modrm(env, s, modrm);
6691             tcg_gen_ld32u_tl(s->T0, cpu_env, offsetof(CPUX86State, mxcsr));
6692             gen_op_st_v(s, MO_32, s->T0, s->A0);
6693             break;
6694 
6695         CASE_MODRM_MEM_OP(4): /* xsave */
6696             if ((s->cpuid_ext_features & CPUID_EXT_XSAVE) == 0
6697                 || (prefixes & (PREFIX_LOCK | PREFIX_DATA
6698                                 | PREFIX_REPZ | PREFIX_REPNZ))) {
6699                 goto illegal_op;
6700             }
6701             gen_lea_modrm(env, s, modrm);
6702             tcg_gen_concat_tl_i64(s->tmp1_i64, cpu_regs[R_EAX],
6703                                   cpu_regs[R_EDX]);
6704             gen_helper_xsave(cpu_env, s->A0, s->tmp1_i64);
6705             break;
6706 
6707         CASE_MODRM_MEM_OP(5): /* xrstor */
6708             if ((s->cpuid_ext_features & CPUID_EXT_XSAVE) == 0
6709                 || (prefixes & (PREFIX_LOCK | PREFIX_DATA
6710                                 | PREFIX_REPZ | PREFIX_REPNZ))) {
6711                 goto illegal_op;
6712             }
6713             gen_lea_modrm(env, s, modrm);
6714             tcg_gen_concat_tl_i64(s->tmp1_i64, cpu_regs[R_EAX],
6715                                   cpu_regs[R_EDX]);
6716             gen_helper_xrstor(cpu_env, s->A0, s->tmp1_i64);
6717             /* XRSTOR is how MPX is enabled, which changes how
6718                we translate.  Thus we need to end the TB.  */
6719             s->base.is_jmp = DISAS_EOB_NEXT;
6720             break;
6721 
6722         CASE_MODRM_MEM_OP(6): /* xsaveopt / clwb */
6723             if (prefixes & PREFIX_LOCK) {
6724                 goto illegal_op;
6725             }
6726             if (prefixes & PREFIX_DATA) {
6727                 /* clwb */
6728                 if (!(s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_CLWB)) {
6729                     goto illegal_op;
6730                 }
6731                 gen_nop_modrm(env, s, modrm);
6732             } else {
6733                 /* xsaveopt */
6734                 if ((s->cpuid_ext_features & CPUID_EXT_XSAVE) == 0
6735                     || (s->cpuid_xsave_features & CPUID_XSAVE_XSAVEOPT) == 0
6736                     || (prefixes & (PREFIX_REPZ | PREFIX_REPNZ))) {
6737                     goto illegal_op;
6738                 }
6739                 gen_lea_modrm(env, s, modrm);
6740                 tcg_gen_concat_tl_i64(s->tmp1_i64, cpu_regs[R_EAX],
6741                                       cpu_regs[R_EDX]);
6742                 gen_helper_xsaveopt(cpu_env, s->A0, s->tmp1_i64);
6743             }
6744             break;
6745 
6746         CASE_MODRM_MEM_OP(7): /* clflush / clflushopt */
6747             if (prefixes & PREFIX_LOCK) {
6748                 goto illegal_op;
6749             }
6750             if (prefixes & PREFIX_DATA) {
6751                 /* clflushopt */
6752                 if (!(s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_CLFLUSHOPT)) {
6753                     goto illegal_op;
6754                 }
6755             } else {
6756                 /* clflush */
6757                 if ((s->prefix & (PREFIX_REPZ | PREFIX_REPNZ))
6758                     || !(s->cpuid_features & CPUID_CLFLUSH)) {
6759                     goto illegal_op;
6760                 }
6761             }
6762             gen_nop_modrm(env, s, modrm);
6763             break;
6764 
6765         case 0xc0 ... 0xc7: /* rdfsbase (f3 0f ae /0) */
6766         case 0xc8 ... 0xcf: /* rdgsbase (f3 0f ae /1) */
6767         case 0xd0 ... 0xd7: /* wrfsbase (f3 0f ae /2) */
6768         case 0xd8 ... 0xdf: /* wrgsbase (f3 0f ae /3) */
6769             if (CODE64(s)
6770                 && (prefixes & PREFIX_REPZ)
6771                 && !(prefixes & PREFIX_LOCK)
6772                 && (s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_FSGSBASE)) {
6773                 TCGv base, treg, src, dst;
6774 
6775                 /* Preserve hflags bits by testing CR4 at runtime.  */
6776                 tcg_gen_movi_i32(s->tmp2_i32, CR4_FSGSBASE_MASK);
6777                 gen_helper_cr4_testbit(cpu_env, s->tmp2_i32);
6778 
6779                 base = cpu_seg_base[modrm & 8 ? R_GS : R_FS];
6780                 treg = cpu_regs[(modrm & 7) | REX_B(s)];
6781 
6782                 if (modrm & 0x10) {
6783                     /* wr*base */
6784                     dst = base, src = treg;
6785                 } else {
6786                     /* rd*base */
6787                     dst = treg, src = base;
6788                 }
6789 
6790                 if (s->dflag == MO_32) {
6791                     tcg_gen_ext32u_tl(dst, src);
6792                 } else {
6793                     tcg_gen_mov_tl(dst, src);
6794                 }
6795                 break;
6796             }
6797             goto unknown_op;
6798 
6799         case 0xf8: /* sfence / pcommit */
6800             if (prefixes & PREFIX_DATA) {
6801                 /* pcommit */
6802                 if (!(s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_PCOMMIT)
6803                     || (prefixes & PREFIX_LOCK)) {
6804                     goto illegal_op;
6805                 }
6806                 break;
6807             }
6808             /* fallthru */
6809         case 0xf9 ... 0xff: /* sfence */
6810             if (!(s->cpuid_features & CPUID_SSE)
6811                 || (prefixes & PREFIX_LOCK)) {
6812                 goto illegal_op;
6813             }
6814             tcg_gen_mb(TCG_MO_ST_ST | TCG_BAR_SC);
6815             break;
6816         case 0xe8 ... 0xef: /* lfence */
6817             if (!(s->cpuid_features & CPUID_SSE)
6818                 || (prefixes & PREFIX_LOCK)) {
6819                 goto illegal_op;
6820             }
6821             tcg_gen_mb(TCG_MO_LD_LD | TCG_BAR_SC);
6822             break;
6823         case 0xf0 ... 0xf7: /* mfence */
6824             if (!(s->cpuid_features & CPUID_SSE2)
6825                 || (prefixes & PREFIX_LOCK)) {
6826                 goto illegal_op;
6827             }
6828             tcg_gen_mb(TCG_MO_ALL | TCG_BAR_SC);
6829             break;
6830 
6831         default:
6832             goto unknown_op;
6833         }
6834         break;
6835 
6836     case 0x10d: /* 3DNow! prefetch(w) */
6837         modrm = x86_ldub_code(env, s);
6838         mod = (modrm >> 6) & 3;
6839         if (mod == 3)
6840             goto illegal_op;
6841         gen_nop_modrm(env, s, modrm);
6842         break;
6843     case 0x1aa: /* rsm */
6844         gen_svm_check_intercept(s, SVM_EXIT_RSM);
6845         if (!(s->flags & HF_SMM_MASK))
6846             goto illegal_op;
6847 #ifdef CONFIG_USER_ONLY
6848         /* we should not be in SMM mode */
6849         g_assert_not_reached();
6850 #else
6851         gen_update_cc_op(s);
6852         gen_update_eip_next(s);
6853         gen_helper_rsm(cpu_env);
6854 #endif /* CONFIG_USER_ONLY */
6855         s->base.is_jmp = DISAS_EOB_ONLY;
6856         break;
6857     case 0x1b8: /* SSE4.2 popcnt */
6858         if ((prefixes & (PREFIX_REPZ | PREFIX_LOCK | PREFIX_REPNZ)) !=
6859              PREFIX_REPZ)
6860             goto illegal_op;
6861         if (!(s->cpuid_ext_features & CPUID_EXT_POPCNT))
6862             goto illegal_op;
6863 
6864         modrm = x86_ldub_code(env, s);
6865         reg = ((modrm >> 3) & 7) | REX_R(s);
6866 
6867         if (s->prefix & PREFIX_DATA) {
6868             ot = MO_16;
6869         } else {
6870             ot = mo_64_32(dflag);
6871         }
6872 
6873         gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 0);
6874         gen_extu(ot, s->T0);
6875         tcg_gen_mov_tl(cpu_cc_src, s->T0);
6876         tcg_gen_ctpop_tl(s->T0, s->T0);
6877         gen_op_mov_reg_v(s, ot, reg, s->T0);
6878 
6879         set_cc_op(s, CC_OP_POPCNT);
6880         break;
6881     case 0x10e ... 0x117:
6882     case 0x128 ... 0x12f:
6883     case 0x138 ... 0x13a:
6884     case 0x150 ... 0x179:
6885     case 0x17c ... 0x17f:
6886     case 0x1c2:
6887     case 0x1c4 ... 0x1c6:
6888     case 0x1d0 ... 0x1fe:
6889         disas_insn_new(s, cpu, b);
6890         break;
6891     default:
6892         goto unknown_op;
6893     }
6894     return true;
6895  illegal_op:
6896     gen_illegal_opcode(s);
6897     return true;
6898  unknown_op:
6899     gen_unknown_opcode(env, s);
6900     return true;
6901 }
6902 
6903 void tcg_x86_init(void)
6904 {
6905     static const char reg_names[CPU_NB_REGS][4] = {
6906 #ifdef TARGET_X86_64
6907         [R_EAX] = "rax",
6908         [R_EBX] = "rbx",
6909         [R_ECX] = "rcx",
6910         [R_EDX] = "rdx",
6911         [R_ESI] = "rsi",
6912         [R_EDI] = "rdi",
6913         [R_EBP] = "rbp",
6914         [R_ESP] = "rsp",
6915         [8]  = "r8",
6916         [9]  = "r9",
6917         [10] = "r10",
6918         [11] = "r11",
6919         [12] = "r12",
6920         [13] = "r13",
6921         [14] = "r14",
6922         [15] = "r15",
6923 #else
6924         [R_EAX] = "eax",
6925         [R_EBX] = "ebx",
6926         [R_ECX] = "ecx",
6927         [R_EDX] = "edx",
6928         [R_ESI] = "esi",
6929         [R_EDI] = "edi",
6930         [R_EBP] = "ebp",
6931         [R_ESP] = "esp",
6932 #endif
6933     };
6934     static const char eip_name[] = {
6935 #ifdef TARGET_X86_64
6936         "rip"
6937 #else
6938         "eip"
6939 #endif
6940     };
6941     static const char seg_base_names[6][8] = {
6942         [R_CS] = "cs_base",
6943         [R_DS] = "ds_base",
6944         [R_ES] = "es_base",
6945         [R_FS] = "fs_base",
6946         [R_GS] = "gs_base",
6947         [R_SS] = "ss_base",
6948     };
6949     static const char bnd_regl_names[4][8] = {
6950         "bnd0_lb", "bnd1_lb", "bnd2_lb", "bnd3_lb"
6951     };
6952     static const char bnd_regu_names[4][8] = {
6953         "bnd0_ub", "bnd1_ub", "bnd2_ub", "bnd3_ub"
6954     };
6955     int i;
6956 
6957     cpu_cc_op = tcg_global_mem_new_i32(cpu_env,
6958                                        offsetof(CPUX86State, cc_op), "cc_op");
6959     cpu_cc_dst = tcg_global_mem_new(cpu_env, offsetof(CPUX86State, cc_dst),
6960                                     "cc_dst");
6961     cpu_cc_src = tcg_global_mem_new(cpu_env, offsetof(CPUX86State, cc_src),
6962                                     "cc_src");
6963     cpu_cc_src2 = tcg_global_mem_new(cpu_env, offsetof(CPUX86State, cc_src2),
6964                                      "cc_src2");
6965     cpu_eip = tcg_global_mem_new(cpu_env, offsetof(CPUX86State, eip), eip_name);
6966 
6967     for (i = 0; i < CPU_NB_REGS; ++i) {
6968         cpu_regs[i] = tcg_global_mem_new(cpu_env,
6969                                          offsetof(CPUX86State, regs[i]),
6970                                          reg_names[i]);
6971     }
6972 
6973     for (i = 0; i < 6; ++i) {
6974         cpu_seg_base[i]
6975             = tcg_global_mem_new(cpu_env,
6976                                  offsetof(CPUX86State, segs[i].base),
6977                                  seg_base_names[i]);
6978     }
6979 
6980     for (i = 0; i < 4; ++i) {
6981         cpu_bndl[i]
6982             = tcg_global_mem_new_i64(cpu_env,
6983                                      offsetof(CPUX86State, bnd_regs[i].lb),
6984                                      bnd_regl_names[i]);
6985         cpu_bndu[i]
6986             = tcg_global_mem_new_i64(cpu_env,
6987                                      offsetof(CPUX86State, bnd_regs[i].ub),
6988                                      bnd_regu_names[i]);
6989     }
6990 }
6991 
6992 static void i386_tr_init_disas_context(DisasContextBase *dcbase, CPUState *cpu)
6993 {
6994     DisasContext *dc = container_of(dcbase, DisasContext, base);
6995     CPUX86State *env = cpu->env_ptr;
6996     uint32_t flags = dc->base.tb->flags;
6997     uint32_t cflags = tb_cflags(dc->base.tb);
6998     int cpl = (flags >> HF_CPL_SHIFT) & 3;
6999     int iopl = (flags >> IOPL_SHIFT) & 3;
7000 
7001     dc->cs_base = dc->base.tb->cs_base;
7002     dc->pc_save = dc->base.pc_next;
7003     dc->flags = flags;
7004 #ifndef CONFIG_USER_ONLY
7005     dc->cpl = cpl;
7006     dc->iopl = iopl;
7007 #endif
7008 
7009     /* We make some simplifying assumptions; validate they're correct. */
7010     g_assert(PE(dc) == ((flags & HF_PE_MASK) != 0));
7011     g_assert(CPL(dc) == cpl);
7012     g_assert(IOPL(dc) == iopl);
7013     g_assert(VM86(dc) == ((flags & HF_VM_MASK) != 0));
7014     g_assert(CODE32(dc) == ((flags & HF_CS32_MASK) != 0));
7015     g_assert(CODE64(dc) == ((flags & HF_CS64_MASK) != 0));
7016     g_assert(SS32(dc) == ((flags & HF_SS32_MASK) != 0));
7017     g_assert(LMA(dc) == ((flags & HF_LMA_MASK) != 0));
7018     g_assert(ADDSEG(dc) == ((flags & HF_ADDSEG_MASK) != 0));
7019     g_assert(SVME(dc) == ((flags & HF_SVME_MASK) != 0));
7020     g_assert(GUEST(dc) == ((flags & HF_GUEST_MASK) != 0));
7021 
7022     dc->cc_op = CC_OP_DYNAMIC;
7023     dc->cc_op_dirty = false;
7024     dc->popl_esp_hack = 0;
7025     /* select memory access functions */
7026     dc->mem_index = 0;
7027 #ifdef CONFIG_SOFTMMU
7028     dc->mem_index = cpu_mmu_index(env, false);
7029 #endif
7030     dc->cpuid_features = env->features[FEAT_1_EDX];
7031     dc->cpuid_ext_features = env->features[FEAT_1_ECX];
7032     dc->cpuid_ext2_features = env->features[FEAT_8000_0001_EDX];
7033     dc->cpuid_ext3_features = env->features[FEAT_8000_0001_ECX];
7034     dc->cpuid_7_0_ebx_features = env->features[FEAT_7_0_EBX];
7035     dc->cpuid_7_0_ecx_features = env->features[FEAT_7_0_ECX];
7036     dc->cpuid_xsave_features = env->features[FEAT_XSAVE];
7037     dc->jmp_opt = !((cflags & CF_NO_GOTO_TB) ||
7038                     (flags & (HF_TF_MASK | HF_INHIBIT_IRQ_MASK)));
7039     /*
7040      * If jmp_opt, we want to handle each string instruction individually.
7041      * For icount also disable repz optimization so that each iteration
7042      * is accounted separately.
7043      */
7044     dc->repz_opt = !dc->jmp_opt && !(cflags & CF_USE_ICOUNT);
7045 
7046     dc->T0 = tcg_temp_new();
7047     dc->T1 = tcg_temp_new();
7048     dc->A0 = tcg_temp_new();
7049 
7050     dc->tmp0 = tcg_temp_new();
7051     dc->tmp1_i64 = tcg_temp_new_i64();
7052     dc->tmp2_i32 = tcg_temp_new_i32();
7053     dc->tmp3_i32 = tcg_temp_new_i32();
7054     dc->tmp4 = tcg_temp_new();
7055     dc->cc_srcT = tcg_temp_local_new();
7056 }
7057 
7058 static void i386_tr_tb_start(DisasContextBase *db, CPUState *cpu)
7059 {
7060 }
7061 
7062 static void i386_tr_insn_start(DisasContextBase *dcbase, CPUState *cpu)
7063 {
7064     DisasContext *dc = container_of(dcbase, DisasContext, base);
7065     target_ulong pc_arg = dc->base.pc_next;
7066 
7067     dc->prev_insn_end = tcg_last_op();
7068     if (TARGET_TB_PCREL) {
7069         pc_arg -= dc->cs_base;
7070         pc_arg &= ~TARGET_PAGE_MASK;
7071     }
7072     tcg_gen_insn_start(pc_arg, dc->cc_op);
7073 }
7074 
7075 static void i386_tr_translate_insn(DisasContextBase *dcbase, CPUState *cpu)
7076 {
7077     DisasContext *dc = container_of(dcbase, DisasContext, base);
7078 
7079 #ifdef TARGET_VSYSCALL_PAGE
7080     /*
7081      * Detect entry into the vsyscall page and invoke the syscall.
7082      */
7083     if ((dc->base.pc_next & TARGET_PAGE_MASK) == TARGET_VSYSCALL_PAGE) {
7084         gen_exception(dc, EXCP_VSYSCALL);
7085         dc->base.pc_next = dc->pc + 1;
7086         return;
7087     }
7088 #endif
7089 
7090     if (disas_insn(dc, cpu)) {
7091         target_ulong pc_next = dc->pc;
7092         dc->base.pc_next = pc_next;
7093 
7094         if (dc->base.is_jmp == DISAS_NEXT) {
7095             if (dc->flags & (HF_TF_MASK | HF_INHIBIT_IRQ_MASK)) {
7096                 /*
7097                  * If single step mode, we generate only one instruction and
7098                  * generate an exception.
7099                  * If irq were inhibited with HF_INHIBIT_IRQ_MASK, we clear
7100                  * the flag and abort the translation to give the irqs a
7101                  * chance to happen.
7102                  */
7103                 dc->base.is_jmp = DISAS_EOB_NEXT;
7104             } else if (!is_same_page(&dc->base, pc_next)) {
7105                 dc->base.is_jmp = DISAS_TOO_MANY;
7106             }
7107         }
7108     }
7109 }
7110 
7111 static void i386_tr_tb_stop(DisasContextBase *dcbase, CPUState *cpu)
7112 {
7113     DisasContext *dc = container_of(dcbase, DisasContext, base);
7114 
7115     switch (dc->base.is_jmp) {
7116     case DISAS_NORETURN:
7117         break;
7118     case DISAS_TOO_MANY:
7119         gen_update_cc_op(dc);
7120         gen_jmp_rel_csize(dc, 0, 0);
7121         break;
7122     case DISAS_EOB_NEXT:
7123         gen_update_cc_op(dc);
7124         gen_update_eip_cur(dc);
7125         /* fall through */
7126     case DISAS_EOB_ONLY:
7127         gen_eob(dc);
7128         break;
7129     case DISAS_EOB_INHIBIT_IRQ:
7130         gen_update_cc_op(dc);
7131         gen_update_eip_cur(dc);
7132         gen_eob_inhibit_irq(dc, true);
7133         break;
7134     case DISAS_JUMP:
7135         gen_jr(dc);
7136         break;
7137     default:
7138         g_assert_not_reached();
7139     }
7140 }
7141 
7142 static void i386_tr_disas_log(const DisasContextBase *dcbase,
7143                               CPUState *cpu, FILE *logfile)
7144 {
7145     DisasContext *dc = container_of(dcbase, DisasContext, base);
7146 
7147     fprintf(logfile, "IN: %s\n", lookup_symbol(dc->base.pc_first));
7148     target_disas(logfile, cpu, dc->base.pc_first, dc->base.tb->size);
7149 }
7150 
7151 static const TranslatorOps i386_tr_ops = {
7152     .init_disas_context = i386_tr_init_disas_context,
7153     .tb_start           = i386_tr_tb_start,
7154     .insn_start         = i386_tr_insn_start,
7155     .translate_insn     = i386_tr_translate_insn,
7156     .tb_stop            = i386_tr_tb_stop,
7157     .disas_log          = i386_tr_disas_log,
7158 };
7159 
7160 /* generate intermediate code for basic block 'tb'.  */
7161 void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int max_insns,
7162                            target_ulong pc, void *host_pc)
7163 {
7164     DisasContext dc;
7165 
7166     translator_loop(cpu, tb, max_insns, pc, host_pc, &i386_tr_ops, &dc.base);
7167 }
7168