xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision 260ea95c)
1 /******************************************************************************
2  * emulate.c
3  *
4  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
5  *
6  * Copyright (c) 2005 Keir Fraser
7  *
8  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
9  * privileged instructions:
10  *
11  * Copyright (C) 2006 Qumranet
12  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13  *
14  *   Avi Kivity <avi@qumranet.com>
15  *   Yaniv Kamay <yaniv@qumranet.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
21  */
22 
23 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
25 #include <asm/kvm_emulate.h>
26 #include <linux/stringify.h>
27 #include <asm/debugreg.h>
28 
29 #include "x86.h"
30 #include "tss.h"
31 
32 /*
33  * Operand types
34  */
35 #define OpNone             0ull
36 #define OpImplicit         1ull  /* No generic decode */
37 #define OpReg              2ull  /* Register */
38 #define OpMem              3ull  /* Memory */
39 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
40 #define OpDI               5ull  /* ES:DI/EDI/RDI */
41 #define OpMem64            6ull  /* Memory, 64-bit */
42 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
43 #define OpDX               8ull  /* DX register */
44 #define OpCL               9ull  /* CL register (for shifts) */
45 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
46 #define OpOne             11ull  /* Implied 1 */
47 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
48 #define OpMem16           13ull  /* Memory operand (16-bit). */
49 #define OpMem32           14ull  /* Memory operand (32-bit). */
50 #define OpImmU            15ull  /* Immediate operand, zero extended */
51 #define OpSI              16ull  /* SI/ESI/RSI */
52 #define OpImmFAddr        17ull  /* Immediate far address */
53 #define OpMemFAddr        18ull  /* Far address in memory */
54 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
55 #define OpES              20ull  /* ES */
56 #define OpCS              21ull  /* CS */
57 #define OpSS              22ull  /* SS */
58 #define OpDS              23ull  /* DS */
59 #define OpFS              24ull  /* FS */
60 #define OpGS              25ull  /* GS */
61 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
62 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
63 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
64 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
65 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
66 
67 #define OpBits             5  /* Width of operand field */
68 #define OpMask             ((1ull << OpBits) - 1)
69 
70 /*
71  * Opcode effective-address decode tables.
72  * Note that we only emulate instructions that have at least one memory
73  * operand (excluding implicit stack references). We assume that stack
74  * references and instruction fetches will never occur in special memory
75  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
76  * not be handled.
77  */
78 
79 /* Operand sizes: 8-bit operands or specified/overridden size. */
80 #define ByteOp      (1<<0)	/* 8-bit operands. */
81 /* Destination operand type. */
82 #define DstShift    1
83 #define ImplicitOps (OpImplicit << DstShift)
84 #define DstReg      (OpReg << DstShift)
85 #define DstMem      (OpMem << DstShift)
86 #define DstAcc      (OpAcc << DstShift)
87 #define DstDI       (OpDI << DstShift)
88 #define DstMem64    (OpMem64 << DstShift)
89 #define DstMem16    (OpMem16 << DstShift)
90 #define DstImmUByte (OpImmUByte << DstShift)
91 #define DstDX       (OpDX << DstShift)
92 #define DstAccLo    (OpAccLo << DstShift)
93 #define DstMask     (OpMask << DstShift)
94 /* Source operand type. */
95 #define SrcShift    6
96 #define SrcNone     (OpNone << SrcShift)
97 #define SrcReg      (OpReg << SrcShift)
98 #define SrcMem      (OpMem << SrcShift)
99 #define SrcMem16    (OpMem16 << SrcShift)
100 #define SrcMem32    (OpMem32 << SrcShift)
101 #define SrcImm      (OpImm << SrcShift)
102 #define SrcImmByte  (OpImmByte << SrcShift)
103 #define SrcOne      (OpOne << SrcShift)
104 #define SrcImmUByte (OpImmUByte << SrcShift)
105 #define SrcImmU     (OpImmU << SrcShift)
106 #define SrcSI       (OpSI << SrcShift)
107 #define SrcXLat     (OpXLat << SrcShift)
108 #define SrcImmFAddr (OpImmFAddr << SrcShift)
109 #define SrcMemFAddr (OpMemFAddr << SrcShift)
110 #define SrcAcc      (OpAcc << SrcShift)
111 #define SrcImmU16   (OpImmU16 << SrcShift)
112 #define SrcImm64    (OpImm64 << SrcShift)
113 #define SrcDX       (OpDX << SrcShift)
114 #define SrcMem8     (OpMem8 << SrcShift)
115 #define SrcAccHi    (OpAccHi << SrcShift)
116 #define SrcMask     (OpMask << SrcShift)
117 #define BitOp       (1<<11)
118 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
119 #define String      (1<<13)     /* String instruction (rep capable) */
120 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
121 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
122 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
123 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
124 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
125 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
126 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
127 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
128 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
129 #define Sse         (1<<18)     /* SSE Vector instruction */
130 /* Generic ModRM decode. */
131 #define ModRM       (1<<19)
132 /* Destination is only written; never read. */
133 #define Mov         (1<<20)
134 /* Misc flags */
135 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
136 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
137 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
138 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
139 #define Undefined   (1<<25) /* No Such Instruction */
140 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
141 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
142 #define No64	    (1<<28)
143 #define PageTable   (1 << 29)   /* instruction used to write page table */
144 #define NotImpl     (1 << 30)   /* instruction is not implemented */
145 /* Source 2 operand type */
146 #define Src2Shift   (31)
147 #define Src2None    (OpNone << Src2Shift)
148 #define Src2Mem     (OpMem << Src2Shift)
149 #define Src2CL      (OpCL << Src2Shift)
150 #define Src2ImmByte (OpImmByte << Src2Shift)
151 #define Src2One     (OpOne << Src2Shift)
152 #define Src2Imm     (OpImm << Src2Shift)
153 #define Src2ES      (OpES << Src2Shift)
154 #define Src2CS      (OpCS << Src2Shift)
155 #define Src2SS      (OpSS << Src2Shift)
156 #define Src2DS      (OpDS << Src2Shift)
157 #define Src2FS      (OpFS << Src2Shift)
158 #define Src2GS      (OpGS << Src2Shift)
159 #define Src2Mask    (OpMask << Src2Shift)
160 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
161 #define AlignMask   ((u64)7 << 41)
162 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
163 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
164 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
165 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
166 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
167 #define NoWrite     ((u64)1 << 45)  /* No writeback */
168 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
169 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
170 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
171 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
172 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
173 #define NearBranch  ((u64)1 << 52)  /* Near branches */
174 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
175 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
176 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
177 
178 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
179 
180 #define X2(x...) x, x
181 #define X3(x...) X2(x), x
182 #define X4(x...) X2(x), X2(x)
183 #define X5(x...) X4(x), x
184 #define X6(x...) X4(x), X2(x)
185 #define X7(x...) X4(x), X3(x)
186 #define X8(x...) X4(x), X4(x)
187 #define X16(x...) X8(x), X8(x)
188 
189 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1)
190 #define FASTOP_SIZE 8
191 
192 /*
193  * fastop functions have a special calling convention:
194  *
195  * dst:    rax        (in/out)
196  * src:    rdx        (in/out)
197  * src2:   rcx        (in)
198  * flags:  rflags     (in/out)
199  * ex:     rsi        (in:fastop pointer, out:zero if exception)
200  *
201  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
202  * different operand sizes can be reached by calculation, rather than a jump
203  * table (which would be bigger than the code).
204  *
205  * fastop functions are declared as taking a never-defined fastop parameter,
206  * so they can't be called from C directly.
207  */
208 
209 struct fastop;
210 
211 struct opcode {
212 	u64 flags : 56;
213 	u64 intercept : 8;
214 	union {
215 		int (*execute)(struct x86_emulate_ctxt *ctxt);
216 		const struct opcode *group;
217 		const struct group_dual *gdual;
218 		const struct gprefix *gprefix;
219 		const struct escape *esc;
220 		const struct instr_dual *idual;
221 		const struct mode_dual *mdual;
222 		void (*fastop)(struct fastop *fake);
223 	} u;
224 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
225 };
226 
227 struct group_dual {
228 	struct opcode mod012[8];
229 	struct opcode mod3[8];
230 };
231 
232 struct gprefix {
233 	struct opcode pfx_no;
234 	struct opcode pfx_66;
235 	struct opcode pfx_f2;
236 	struct opcode pfx_f3;
237 };
238 
239 struct escape {
240 	struct opcode op[8];
241 	struct opcode high[64];
242 };
243 
244 struct instr_dual {
245 	struct opcode mod012;
246 	struct opcode mod3;
247 };
248 
249 struct mode_dual {
250 	struct opcode mode32;
251 	struct opcode mode64;
252 };
253 
254 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
255 
256 enum x86_transfer_type {
257 	X86_TRANSFER_NONE,
258 	X86_TRANSFER_CALL_JMP,
259 	X86_TRANSFER_RET,
260 	X86_TRANSFER_TASK_SWITCH,
261 };
262 
263 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
264 {
265 	if (!(ctxt->regs_valid & (1 << nr))) {
266 		ctxt->regs_valid |= 1 << nr;
267 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
268 	}
269 	return ctxt->_regs[nr];
270 }
271 
272 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
273 {
274 	ctxt->regs_valid |= 1 << nr;
275 	ctxt->regs_dirty |= 1 << nr;
276 	return &ctxt->_regs[nr];
277 }
278 
279 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
280 {
281 	reg_read(ctxt, nr);
282 	return reg_write(ctxt, nr);
283 }
284 
285 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
286 {
287 	unsigned reg;
288 
289 	for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16)
290 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
291 }
292 
293 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
294 {
295 	ctxt->regs_dirty = 0;
296 	ctxt->regs_valid = 0;
297 }
298 
299 /*
300  * These EFLAGS bits are restored from saved value during emulation, and
301  * any changes are written back to the saved value after emulation.
302  */
303 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
304 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
305 
306 #ifdef CONFIG_X86_64
307 #define ON64(x) x
308 #else
309 #define ON64(x)
310 #endif
311 
312 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *));
313 
314 #define FOP_FUNC(name) \
315 	".align " __stringify(FASTOP_SIZE) " \n\t" \
316 	".type " name ", @function \n\t" \
317 	name ":\n\t"
318 
319 #define FOP_RET   "ret \n\t"
320 
321 #define FOP_START(op) \
322 	extern void em_##op(struct fastop *fake); \
323 	asm(".pushsection .text, \"ax\" \n\t" \
324 	    ".global em_" #op " \n\t" \
325 	    FOP_FUNC("em_" #op)
326 
327 #define FOP_END \
328 	    ".popsection")
329 
330 #define FOPNOP() \
331 	FOP_FUNC(__stringify(__UNIQUE_ID(nop))) \
332 	FOP_RET
333 
334 #define FOP1E(op,  dst) \
335 	FOP_FUNC(#op "_" #dst) \
336 	"10: " #op " %" #dst " \n\t" FOP_RET
337 
338 #define FOP1EEX(op,  dst) \
339 	FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception)
340 
341 #define FASTOP1(op) \
342 	FOP_START(op) \
343 	FOP1E(op##b, al) \
344 	FOP1E(op##w, ax) \
345 	FOP1E(op##l, eax) \
346 	ON64(FOP1E(op##q, rax))	\
347 	FOP_END
348 
349 /* 1-operand, using src2 (for MUL/DIV r/m) */
350 #define FASTOP1SRC2(op, name) \
351 	FOP_START(name) \
352 	FOP1E(op, cl) \
353 	FOP1E(op, cx) \
354 	FOP1E(op, ecx) \
355 	ON64(FOP1E(op, rcx)) \
356 	FOP_END
357 
358 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
359 #define FASTOP1SRC2EX(op, name) \
360 	FOP_START(name) \
361 	FOP1EEX(op, cl) \
362 	FOP1EEX(op, cx) \
363 	FOP1EEX(op, ecx) \
364 	ON64(FOP1EEX(op, rcx)) \
365 	FOP_END
366 
367 #define FOP2E(op,  dst, src)	   \
368 	FOP_FUNC(#op "_" #dst "_" #src) \
369 	#op " %" #src ", %" #dst " \n\t" FOP_RET
370 
371 #define FASTOP2(op) \
372 	FOP_START(op) \
373 	FOP2E(op##b, al, dl) \
374 	FOP2E(op##w, ax, dx) \
375 	FOP2E(op##l, eax, edx) \
376 	ON64(FOP2E(op##q, rax, rdx)) \
377 	FOP_END
378 
379 /* 2 operand, word only */
380 #define FASTOP2W(op) \
381 	FOP_START(op) \
382 	FOPNOP() \
383 	FOP2E(op##w, ax, dx) \
384 	FOP2E(op##l, eax, edx) \
385 	ON64(FOP2E(op##q, rax, rdx)) \
386 	FOP_END
387 
388 /* 2 operand, src is CL */
389 #define FASTOP2CL(op) \
390 	FOP_START(op) \
391 	FOP2E(op##b, al, cl) \
392 	FOP2E(op##w, ax, cl) \
393 	FOP2E(op##l, eax, cl) \
394 	ON64(FOP2E(op##q, rax, cl)) \
395 	FOP_END
396 
397 /* 2 operand, src and dest are reversed */
398 #define FASTOP2R(op, name) \
399 	FOP_START(name) \
400 	FOP2E(op##b, dl, al) \
401 	FOP2E(op##w, dx, ax) \
402 	FOP2E(op##l, edx, eax) \
403 	ON64(FOP2E(op##q, rdx, rax)) \
404 	FOP_END
405 
406 #define FOP3E(op,  dst, src, src2) \
407 	FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
408 	#op " %" #src2 ", %" #src ", %" #dst " \n\t" FOP_RET
409 
410 /* 3-operand, word-only, src2=cl */
411 #define FASTOP3WCL(op) \
412 	FOP_START(op) \
413 	FOPNOP() \
414 	FOP3E(op##w, ax, dx, cl) \
415 	FOP3E(op##l, eax, edx, cl) \
416 	ON64(FOP3E(op##q, rax, rdx, cl)) \
417 	FOP_END
418 
419 /* Special case for SETcc - 1 instruction per cc */
420 #define FOP_SETCC(op) \
421 	".align 4 \n\t" \
422 	".type " #op ", @function \n\t" \
423 	#op ": \n\t" \
424 	#op " %al \n\t" \
425 	FOP_RET
426 
427 asm(".global kvm_fastop_exception \n"
428     "kvm_fastop_exception: xor %esi, %esi; ret");
429 
430 FOP_START(setcc)
431 FOP_SETCC(seto)
432 FOP_SETCC(setno)
433 FOP_SETCC(setc)
434 FOP_SETCC(setnc)
435 FOP_SETCC(setz)
436 FOP_SETCC(setnz)
437 FOP_SETCC(setbe)
438 FOP_SETCC(setnbe)
439 FOP_SETCC(sets)
440 FOP_SETCC(setns)
441 FOP_SETCC(setp)
442 FOP_SETCC(setnp)
443 FOP_SETCC(setl)
444 FOP_SETCC(setnl)
445 FOP_SETCC(setle)
446 FOP_SETCC(setnle)
447 FOP_END;
448 
449 FOP_START(salc) "pushf; sbb %al, %al; popf \n\t" FOP_RET
450 FOP_END;
451 
452 /*
453  * XXX: inoutclob user must know where the argument is being expanded.
454  *      Relying on CC_HAVE_ASM_GOTO would allow us to remove _fault.
455  */
456 #define asm_safe(insn, inoutclob...) \
457 ({ \
458 	int _fault = 0; \
459  \
460 	asm volatile("1:" insn "\n" \
461 	             "2:\n" \
462 	             ".pushsection .fixup, \"ax\"\n" \
463 	             "3: movl $1, %[_fault]\n" \
464 	             "   jmp  2b\n" \
465 	             ".popsection\n" \
466 	             _ASM_EXTABLE(1b, 3b) \
467 	             : [_fault] "+qm"(_fault) inoutclob ); \
468  \
469 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
470 })
471 
472 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
473 				    enum x86_intercept intercept,
474 				    enum x86_intercept_stage stage)
475 {
476 	struct x86_instruction_info info = {
477 		.intercept  = intercept,
478 		.rep_prefix = ctxt->rep_prefix,
479 		.modrm_mod  = ctxt->modrm_mod,
480 		.modrm_reg  = ctxt->modrm_reg,
481 		.modrm_rm   = ctxt->modrm_rm,
482 		.src_val    = ctxt->src.val64,
483 		.dst_val    = ctxt->dst.val64,
484 		.src_bytes  = ctxt->src.bytes,
485 		.dst_bytes  = ctxt->dst.bytes,
486 		.ad_bytes   = ctxt->ad_bytes,
487 		.next_rip   = ctxt->eip,
488 	};
489 
490 	return ctxt->ops->intercept(ctxt, &info, stage);
491 }
492 
493 static void assign_masked(ulong *dest, ulong src, ulong mask)
494 {
495 	*dest = (*dest & ~mask) | (src & mask);
496 }
497 
498 static void assign_register(unsigned long *reg, u64 val, int bytes)
499 {
500 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
501 	switch (bytes) {
502 	case 1:
503 		*(u8 *)reg = (u8)val;
504 		break;
505 	case 2:
506 		*(u16 *)reg = (u16)val;
507 		break;
508 	case 4:
509 		*reg = (u32)val;
510 		break;	/* 64b: zero-extend */
511 	case 8:
512 		*reg = val;
513 		break;
514 	}
515 }
516 
517 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
518 {
519 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
520 }
521 
522 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
523 {
524 	u16 sel;
525 	struct desc_struct ss;
526 
527 	if (ctxt->mode == X86EMUL_MODE_PROT64)
528 		return ~0UL;
529 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
530 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
531 }
532 
533 static int stack_size(struct x86_emulate_ctxt *ctxt)
534 {
535 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
536 }
537 
538 /* Access/update address held in a register, based on addressing mode. */
539 static inline unsigned long
540 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
541 {
542 	if (ctxt->ad_bytes == sizeof(unsigned long))
543 		return reg;
544 	else
545 		return reg & ad_mask(ctxt);
546 }
547 
548 static inline unsigned long
549 register_address(struct x86_emulate_ctxt *ctxt, int reg)
550 {
551 	return address_mask(ctxt, reg_read(ctxt, reg));
552 }
553 
554 static void masked_increment(ulong *reg, ulong mask, int inc)
555 {
556 	assign_masked(reg, *reg + inc, mask);
557 }
558 
559 static inline void
560 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
561 {
562 	ulong *preg = reg_rmw(ctxt, reg);
563 
564 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
565 }
566 
567 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
568 {
569 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
570 }
571 
572 static u32 desc_limit_scaled(struct desc_struct *desc)
573 {
574 	u32 limit = get_desc_limit(desc);
575 
576 	return desc->g ? (limit << 12) | 0xfff : limit;
577 }
578 
579 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
580 {
581 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
582 		return 0;
583 
584 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
585 }
586 
587 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
588 			     u32 error, bool valid)
589 {
590 	WARN_ON(vec > 0x1f);
591 	ctxt->exception.vector = vec;
592 	ctxt->exception.error_code = error;
593 	ctxt->exception.error_code_valid = valid;
594 	return X86EMUL_PROPAGATE_FAULT;
595 }
596 
597 static int emulate_db(struct x86_emulate_ctxt *ctxt)
598 {
599 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
600 }
601 
602 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
603 {
604 	return emulate_exception(ctxt, GP_VECTOR, err, true);
605 }
606 
607 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
608 {
609 	return emulate_exception(ctxt, SS_VECTOR, err, true);
610 }
611 
612 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
613 {
614 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
615 }
616 
617 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
618 {
619 	return emulate_exception(ctxt, TS_VECTOR, err, true);
620 }
621 
622 static int emulate_de(struct x86_emulate_ctxt *ctxt)
623 {
624 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
625 }
626 
627 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
628 {
629 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
630 }
631 
632 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
633 {
634 	u16 selector;
635 	struct desc_struct desc;
636 
637 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
638 	return selector;
639 }
640 
641 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
642 				 unsigned seg)
643 {
644 	u16 dummy;
645 	u32 base3;
646 	struct desc_struct desc;
647 
648 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
649 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
650 }
651 
652 /*
653  * x86 defines three classes of vector instructions: explicitly
654  * aligned, explicitly unaligned, and the rest, which change behaviour
655  * depending on whether they're AVX encoded or not.
656  *
657  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
658  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
659  * 512 bytes of data must be aligned to a 16 byte boundary.
660  */
661 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
662 {
663 	u64 alignment = ctxt->d & AlignMask;
664 
665 	if (likely(size < 16))
666 		return 1;
667 
668 	switch (alignment) {
669 	case Unaligned:
670 	case Avx:
671 		return 1;
672 	case Aligned16:
673 		return 16;
674 	case Aligned:
675 	default:
676 		return size;
677 	}
678 }
679 
680 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
681 				       struct segmented_address addr,
682 				       unsigned *max_size, unsigned size,
683 				       bool write, bool fetch,
684 				       enum x86emul_mode mode, ulong *linear)
685 {
686 	struct desc_struct desc;
687 	bool usable;
688 	ulong la;
689 	u32 lim;
690 	u16 sel;
691 
692 	la = seg_base(ctxt, addr.seg) + addr.ea;
693 	*max_size = 0;
694 	switch (mode) {
695 	case X86EMUL_MODE_PROT64:
696 		*linear = la;
697 		if (is_noncanonical_address(la))
698 			goto bad;
699 
700 		*max_size = min_t(u64, ~0u, (1ull << 48) - la);
701 		if (size > *max_size)
702 			goto bad;
703 		break;
704 	default:
705 		*linear = la = (u32)la;
706 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
707 						addr.seg);
708 		if (!usable)
709 			goto bad;
710 		/* code segment in protected mode or read-only data segment */
711 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
712 					|| !(desc.type & 2)) && write)
713 			goto bad;
714 		/* unreadable code segment */
715 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
716 			goto bad;
717 		lim = desc_limit_scaled(&desc);
718 		if (!(desc.type & 8) && (desc.type & 4)) {
719 			/* expand-down segment */
720 			if (addr.ea <= lim)
721 				goto bad;
722 			lim = desc.d ? 0xffffffff : 0xffff;
723 		}
724 		if (addr.ea > lim)
725 			goto bad;
726 		if (lim == 0xffffffff)
727 			*max_size = ~0u;
728 		else {
729 			*max_size = (u64)lim + 1 - addr.ea;
730 			if (size > *max_size)
731 				goto bad;
732 		}
733 		break;
734 	}
735 	if (la & (insn_alignment(ctxt, size) - 1))
736 		return emulate_gp(ctxt, 0);
737 	return X86EMUL_CONTINUE;
738 bad:
739 	if (addr.seg == VCPU_SREG_SS)
740 		return emulate_ss(ctxt, 0);
741 	else
742 		return emulate_gp(ctxt, 0);
743 }
744 
745 static int linearize(struct x86_emulate_ctxt *ctxt,
746 		     struct segmented_address addr,
747 		     unsigned size, bool write,
748 		     ulong *linear)
749 {
750 	unsigned max_size;
751 	return __linearize(ctxt, addr, &max_size, size, write, false,
752 			   ctxt->mode, linear);
753 }
754 
755 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
756 			     enum x86emul_mode mode)
757 {
758 	ulong linear;
759 	int rc;
760 	unsigned max_size;
761 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
762 					   .ea = dst };
763 
764 	if (ctxt->op_bytes != sizeof(unsigned long))
765 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
766 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
767 	if (rc == X86EMUL_CONTINUE)
768 		ctxt->_eip = addr.ea;
769 	return rc;
770 }
771 
772 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
773 {
774 	return assign_eip(ctxt, dst, ctxt->mode);
775 }
776 
777 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
778 			  const struct desc_struct *cs_desc)
779 {
780 	enum x86emul_mode mode = ctxt->mode;
781 	int rc;
782 
783 #ifdef CONFIG_X86_64
784 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
785 		if (cs_desc->l) {
786 			u64 efer = 0;
787 
788 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
789 			if (efer & EFER_LMA)
790 				mode = X86EMUL_MODE_PROT64;
791 		} else
792 			mode = X86EMUL_MODE_PROT32; /* temporary value */
793 	}
794 #endif
795 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
796 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
797 	rc = assign_eip(ctxt, dst, mode);
798 	if (rc == X86EMUL_CONTINUE)
799 		ctxt->mode = mode;
800 	return rc;
801 }
802 
803 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
804 {
805 	return assign_eip_near(ctxt, ctxt->_eip + rel);
806 }
807 
808 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
809 			      struct segmented_address addr,
810 			      void *data,
811 			      unsigned size)
812 {
813 	int rc;
814 	ulong linear;
815 
816 	rc = linearize(ctxt, addr, size, false, &linear);
817 	if (rc != X86EMUL_CONTINUE)
818 		return rc;
819 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception);
820 }
821 
822 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
823 			       struct segmented_address addr,
824 			       void *data,
825 			       unsigned int size)
826 {
827 	int rc;
828 	ulong linear;
829 
830 	rc = linearize(ctxt, addr, size, true, &linear);
831 	if (rc != X86EMUL_CONTINUE)
832 		return rc;
833 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception);
834 }
835 
836 /*
837  * Prefetch the remaining bytes of the instruction without crossing page
838  * boundary if they are not in fetch_cache yet.
839  */
840 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
841 {
842 	int rc;
843 	unsigned size, max_size;
844 	unsigned long linear;
845 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
846 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
847 					   .ea = ctxt->eip + cur_size };
848 
849 	/*
850 	 * We do not know exactly how many bytes will be needed, and
851 	 * __linearize is expensive, so fetch as much as possible.  We
852 	 * just have to avoid going beyond the 15 byte limit, the end
853 	 * of the segment, or the end of the page.
854 	 *
855 	 * __linearize is called with size 0 so that it does not do any
856 	 * boundary check itself.  Instead, we use max_size to check
857 	 * against op_size.
858 	 */
859 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
860 			 &linear);
861 	if (unlikely(rc != X86EMUL_CONTINUE))
862 		return rc;
863 
864 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
865 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
866 
867 	/*
868 	 * One instruction can only straddle two pages,
869 	 * and one has been loaded at the beginning of
870 	 * x86_decode_insn.  So, if not enough bytes
871 	 * still, we must have hit the 15-byte boundary.
872 	 */
873 	if (unlikely(size < op_size))
874 		return emulate_gp(ctxt, 0);
875 
876 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
877 			      size, &ctxt->exception);
878 	if (unlikely(rc != X86EMUL_CONTINUE))
879 		return rc;
880 	ctxt->fetch.end += size;
881 	return X86EMUL_CONTINUE;
882 }
883 
884 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
885 					       unsigned size)
886 {
887 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
888 
889 	if (unlikely(done_size < size))
890 		return __do_insn_fetch_bytes(ctxt, size - done_size);
891 	else
892 		return X86EMUL_CONTINUE;
893 }
894 
895 /* Fetch next part of the instruction being emulated. */
896 #define insn_fetch(_type, _ctxt)					\
897 ({	_type _x;							\
898 									\
899 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
900 	if (rc != X86EMUL_CONTINUE)					\
901 		goto done;						\
902 	ctxt->_eip += sizeof(_type);					\
903 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
904 	ctxt->fetch.ptr += sizeof(_type);				\
905 	_x;								\
906 })
907 
908 #define insn_fetch_arr(_arr, _size, _ctxt)				\
909 ({									\
910 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
911 	if (rc != X86EMUL_CONTINUE)					\
912 		goto done;						\
913 	ctxt->_eip += (_size);						\
914 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
915 	ctxt->fetch.ptr += (_size);					\
916 })
917 
918 /*
919  * Given the 'reg' portion of a ModRM byte, and a register block, return a
920  * pointer into the block that addresses the relevant register.
921  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
922  */
923 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
924 			     int byteop)
925 {
926 	void *p;
927 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
928 
929 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
930 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
931 	else
932 		p = reg_rmw(ctxt, modrm_reg);
933 	return p;
934 }
935 
936 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
937 			   struct segmented_address addr,
938 			   u16 *size, unsigned long *address, int op_bytes)
939 {
940 	int rc;
941 
942 	if (op_bytes == 2)
943 		op_bytes = 3;
944 	*address = 0;
945 	rc = segmented_read_std(ctxt, addr, size, 2);
946 	if (rc != X86EMUL_CONTINUE)
947 		return rc;
948 	addr.ea += 2;
949 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
950 	return rc;
951 }
952 
953 FASTOP2(add);
954 FASTOP2(or);
955 FASTOP2(adc);
956 FASTOP2(sbb);
957 FASTOP2(and);
958 FASTOP2(sub);
959 FASTOP2(xor);
960 FASTOP2(cmp);
961 FASTOP2(test);
962 
963 FASTOP1SRC2(mul, mul_ex);
964 FASTOP1SRC2(imul, imul_ex);
965 FASTOP1SRC2EX(div, div_ex);
966 FASTOP1SRC2EX(idiv, idiv_ex);
967 
968 FASTOP3WCL(shld);
969 FASTOP3WCL(shrd);
970 
971 FASTOP2W(imul);
972 
973 FASTOP1(not);
974 FASTOP1(neg);
975 FASTOP1(inc);
976 FASTOP1(dec);
977 
978 FASTOP2CL(rol);
979 FASTOP2CL(ror);
980 FASTOP2CL(rcl);
981 FASTOP2CL(rcr);
982 FASTOP2CL(shl);
983 FASTOP2CL(shr);
984 FASTOP2CL(sar);
985 
986 FASTOP2W(bsf);
987 FASTOP2W(bsr);
988 FASTOP2W(bt);
989 FASTOP2W(bts);
990 FASTOP2W(btr);
991 FASTOP2W(btc);
992 
993 FASTOP2(xadd);
994 
995 FASTOP2R(cmp, cmp_r);
996 
997 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
998 {
999 	/* If src is zero, do not writeback, but update flags */
1000 	if (ctxt->src.val == 0)
1001 		ctxt->dst.type = OP_NONE;
1002 	return fastop(ctxt, em_bsf);
1003 }
1004 
1005 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1006 {
1007 	/* If src is zero, do not writeback, but update flags */
1008 	if (ctxt->src.val == 0)
1009 		ctxt->dst.type = OP_NONE;
1010 	return fastop(ctxt, em_bsr);
1011 }
1012 
1013 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1014 {
1015 	u8 rc;
1016 	void (*fop)(void) = (void *)em_setcc + 4 * (condition & 0xf);
1017 
1018 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1019 	asm("push %[flags]; popf; call *%[fastop]"
1020 	    : "=a"(rc) : [fastop]"r"(fop), [flags]"r"(flags));
1021 	return rc;
1022 }
1023 
1024 static void fetch_register_operand(struct operand *op)
1025 {
1026 	switch (op->bytes) {
1027 	case 1:
1028 		op->val = *(u8 *)op->addr.reg;
1029 		break;
1030 	case 2:
1031 		op->val = *(u16 *)op->addr.reg;
1032 		break;
1033 	case 4:
1034 		op->val = *(u32 *)op->addr.reg;
1035 		break;
1036 	case 8:
1037 		op->val = *(u64 *)op->addr.reg;
1038 		break;
1039 	}
1040 }
1041 
1042 static void read_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data, int reg)
1043 {
1044 	ctxt->ops->get_fpu(ctxt);
1045 	switch (reg) {
1046 	case 0: asm("movdqa %%xmm0, %0" : "=m"(*data)); break;
1047 	case 1: asm("movdqa %%xmm1, %0" : "=m"(*data)); break;
1048 	case 2: asm("movdqa %%xmm2, %0" : "=m"(*data)); break;
1049 	case 3: asm("movdqa %%xmm3, %0" : "=m"(*data)); break;
1050 	case 4: asm("movdqa %%xmm4, %0" : "=m"(*data)); break;
1051 	case 5: asm("movdqa %%xmm5, %0" : "=m"(*data)); break;
1052 	case 6: asm("movdqa %%xmm6, %0" : "=m"(*data)); break;
1053 	case 7: asm("movdqa %%xmm7, %0" : "=m"(*data)); break;
1054 #ifdef CONFIG_X86_64
1055 	case 8: asm("movdqa %%xmm8, %0" : "=m"(*data)); break;
1056 	case 9: asm("movdqa %%xmm9, %0" : "=m"(*data)); break;
1057 	case 10: asm("movdqa %%xmm10, %0" : "=m"(*data)); break;
1058 	case 11: asm("movdqa %%xmm11, %0" : "=m"(*data)); break;
1059 	case 12: asm("movdqa %%xmm12, %0" : "=m"(*data)); break;
1060 	case 13: asm("movdqa %%xmm13, %0" : "=m"(*data)); break;
1061 	case 14: asm("movdqa %%xmm14, %0" : "=m"(*data)); break;
1062 	case 15: asm("movdqa %%xmm15, %0" : "=m"(*data)); break;
1063 #endif
1064 	default: BUG();
1065 	}
1066 	ctxt->ops->put_fpu(ctxt);
1067 }
1068 
1069 static void write_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data,
1070 			  int reg)
1071 {
1072 	ctxt->ops->get_fpu(ctxt);
1073 	switch (reg) {
1074 	case 0: asm("movdqa %0, %%xmm0" : : "m"(*data)); break;
1075 	case 1: asm("movdqa %0, %%xmm1" : : "m"(*data)); break;
1076 	case 2: asm("movdqa %0, %%xmm2" : : "m"(*data)); break;
1077 	case 3: asm("movdqa %0, %%xmm3" : : "m"(*data)); break;
1078 	case 4: asm("movdqa %0, %%xmm4" : : "m"(*data)); break;
1079 	case 5: asm("movdqa %0, %%xmm5" : : "m"(*data)); break;
1080 	case 6: asm("movdqa %0, %%xmm6" : : "m"(*data)); break;
1081 	case 7: asm("movdqa %0, %%xmm7" : : "m"(*data)); break;
1082 #ifdef CONFIG_X86_64
1083 	case 8: asm("movdqa %0, %%xmm8" : : "m"(*data)); break;
1084 	case 9: asm("movdqa %0, %%xmm9" : : "m"(*data)); break;
1085 	case 10: asm("movdqa %0, %%xmm10" : : "m"(*data)); break;
1086 	case 11: asm("movdqa %0, %%xmm11" : : "m"(*data)); break;
1087 	case 12: asm("movdqa %0, %%xmm12" : : "m"(*data)); break;
1088 	case 13: asm("movdqa %0, %%xmm13" : : "m"(*data)); break;
1089 	case 14: asm("movdqa %0, %%xmm14" : : "m"(*data)); break;
1090 	case 15: asm("movdqa %0, %%xmm15" : : "m"(*data)); break;
1091 #endif
1092 	default: BUG();
1093 	}
1094 	ctxt->ops->put_fpu(ctxt);
1095 }
1096 
1097 static void read_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1098 {
1099 	ctxt->ops->get_fpu(ctxt);
1100 	switch (reg) {
1101 	case 0: asm("movq %%mm0, %0" : "=m"(*data)); break;
1102 	case 1: asm("movq %%mm1, %0" : "=m"(*data)); break;
1103 	case 2: asm("movq %%mm2, %0" : "=m"(*data)); break;
1104 	case 3: asm("movq %%mm3, %0" : "=m"(*data)); break;
1105 	case 4: asm("movq %%mm4, %0" : "=m"(*data)); break;
1106 	case 5: asm("movq %%mm5, %0" : "=m"(*data)); break;
1107 	case 6: asm("movq %%mm6, %0" : "=m"(*data)); break;
1108 	case 7: asm("movq %%mm7, %0" : "=m"(*data)); break;
1109 	default: BUG();
1110 	}
1111 	ctxt->ops->put_fpu(ctxt);
1112 }
1113 
1114 static void write_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1115 {
1116 	ctxt->ops->get_fpu(ctxt);
1117 	switch (reg) {
1118 	case 0: asm("movq %0, %%mm0" : : "m"(*data)); break;
1119 	case 1: asm("movq %0, %%mm1" : : "m"(*data)); break;
1120 	case 2: asm("movq %0, %%mm2" : : "m"(*data)); break;
1121 	case 3: asm("movq %0, %%mm3" : : "m"(*data)); break;
1122 	case 4: asm("movq %0, %%mm4" : : "m"(*data)); break;
1123 	case 5: asm("movq %0, %%mm5" : : "m"(*data)); break;
1124 	case 6: asm("movq %0, %%mm6" : : "m"(*data)); break;
1125 	case 7: asm("movq %0, %%mm7" : : "m"(*data)); break;
1126 	default: BUG();
1127 	}
1128 	ctxt->ops->put_fpu(ctxt);
1129 }
1130 
1131 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1132 {
1133 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1134 		return emulate_nm(ctxt);
1135 
1136 	ctxt->ops->get_fpu(ctxt);
1137 	asm volatile("fninit");
1138 	ctxt->ops->put_fpu(ctxt);
1139 	return X86EMUL_CONTINUE;
1140 }
1141 
1142 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1143 {
1144 	u16 fcw;
1145 
1146 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1147 		return emulate_nm(ctxt);
1148 
1149 	ctxt->ops->get_fpu(ctxt);
1150 	asm volatile("fnstcw %0": "+m"(fcw));
1151 	ctxt->ops->put_fpu(ctxt);
1152 
1153 	ctxt->dst.val = fcw;
1154 
1155 	return X86EMUL_CONTINUE;
1156 }
1157 
1158 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1159 {
1160 	u16 fsw;
1161 
1162 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1163 		return emulate_nm(ctxt);
1164 
1165 	ctxt->ops->get_fpu(ctxt);
1166 	asm volatile("fnstsw %0": "+m"(fsw));
1167 	ctxt->ops->put_fpu(ctxt);
1168 
1169 	ctxt->dst.val = fsw;
1170 
1171 	return X86EMUL_CONTINUE;
1172 }
1173 
1174 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1175 				    struct operand *op)
1176 {
1177 	unsigned reg = ctxt->modrm_reg;
1178 
1179 	if (!(ctxt->d & ModRM))
1180 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1181 
1182 	if (ctxt->d & Sse) {
1183 		op->type = OP_XMM;
1184 		op->bytes = 16;
1185 		op->addr.xmm = reg;
1186 		read_sse_reg(ctxt, &op->vec_val, reg);
1187 		return;
1188 	}
1189 	if (ctxt->d & Mmx) {
1190 		reg &= 7;
1191 		op->type = OP_MM;
1192 		op->bytes = 8;
1193 		op->addr.mm = reg;
1194 		return;
1195 	}
1196 
1197 	op->type = OP_REG;
1198 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1199 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1200 
1201 	fetch_register_operand(op);
1202 	op->orig_val = op->val;
1203 }
1204 
1205 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1206 {
1207 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1208 		ctxt->modrm_seg = VCPU_SREG_SS;
1209 }
1210 
1211 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1212 			struct operand *op)
1213 {
1214 	u8 sib;
1215 	int index_reg, base_reg, scale;
1216 	int rc = X86EMUL_CONTINUE;
1217 	ulong modrm_ea = 0;
1218 
1219 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1220 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1221 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1222 
1223 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1224 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1225 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1226 	ctxt->modrm_seg = VCPU_SREG_DS;
1227 
1228 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1229 		op->type = OP_REG;
1230 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1231 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1232 				ctxt->d & ByteOp);
1233 		if (ctxt->d & Sse) {
1234 			op->type = OP_XMM;
1235 			op->bytes = 16;
1236 			op->addr.xmm = ctxt->modrm_rm;
1237 			read_sse_reg(ctxt, &op->vec_val, ctxt->modrm_rm);
1238 			return rc;
1239 		}
1240 		if (ctxt->d & Mmx) {
1241 			op->type = OP_MM;
1242 			op->bytes = 8;
1243 			op->addr.mm = ctxt->modrm_rm & 7;
1244 			return rc;
1245 		}
1246 		fetch_register_operand(op);
1247 		return rc;
1248 	}
1249 
1250 	op->type = OP_MEM;
1251 
1252 	if (ctxt->ad_bytes == 2) {
1253 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1254 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1255 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1256 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1257 
1258 		/* 16-bit ModR/M decode. */
1259 		switch (ctxt->modrm_mod) {
1260 		case 0:
1261 			if (ctxt->modrm_rm == 6)
1262 				modrm_ea += insn_fetch(u16, ctxt);
1263 			break;
1264 		case 1:
1265 			modrm_ea += insn_fetch(s8, ctxt);
1266 			break;
1267 		case 2:
1268 			modrm_ea += insn_fetch(u16, ctxt);
1269 			break;
1270 		}
1271 		switch (ctxt->modrm_rm) {
1272 		case 0:
1273 			modrm_ea += bx + si;
1274 			break;
1275 		case 1:
1276 			modrm_ea += bx + di;
1277 			break;
1278 		case 2:
1279 			modrm_ea += bp + si;
1280 			break;
1281 		case 3:
1282 			modrm_ea += bp + di;
1283 			break;
1284 		case 4:
1285 			modrm_ea += si;
1286 			break;
1287 		case 5:
1288 			modrm_ea += di;
1289 			break;
1290 		case 6:
1291 			if (ctxt->modrm_mod != 0)
1292 				modrm_ea += bp;
1293 			break;
1294 		case 7:
1295 			modrm_ea += bx;
1296 			break;
1297 		}
1298 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1299 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1300 			ctxt->modrm_seg = VCPU_SREG_SS;
1301 		modrm_ea = (u16)modrm_ea;
1302 	} else {
1303 		/* 32/64-bit ModR/M decode. */
1304 		if ((ctxt->modrm_rm & 7) == 4) {
1305 			sib = insn_fetch(u8, ctxt);
1306 			index_reg |= (sib >> 3) & 7;
1307 			base_reg |= sib & 7;
1308 			scale = sib >> 6;
1309 
1310 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1311 				modrm_ea += insn_fetch(s32, ctxt);
1312 			else {
1313 				modrm_ea += reg_read(ctxt, base_reg);
1314 				adjust_modrm_seg(ctxt, base_reg);
1315 				/* Increment ESP on POP [ESP] */
1316 				if ((ctxt->d & IncSP) &&
1317 				    base_reg == VCPU_REGS_RSP)
1318 					modrm_ea += ctxt->op_bytes;
1319 			}
1320 			if (index_reg != 4)
1321 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1322 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1323 			modrm_ea += insn_fetch(s32, ctxt);
1324 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1325 				ctxt->rip_relative = 1;
1326 		} else {
1327 			base_reg = ctxt->modrm_rm;
1328 			modrm_ea += reg_read(ctxt, base_reg);
1329 			adjust_modrm_seg(ctxt, base_reg);
1330 		}
1331 		switch (ctxt->modrm_mod) {
1332 		case 1:
1333 			modrm_ea += insn_fetch(s8, ctxt);
1334 			break;
1335 		case 2:
1336 			modrm_ea += insn_fetch(s32, ctxt);
1337 			break;
1338 		}
1339 	}
1340 	op->addr.mem.ea = modrm_ea;
1341 	if (ctxt->ad_bytes != 8)
1342 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1343 
1344 done:
1345 	return rc;
1346 }
1347 
1348 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1349 		      struct operand *op)
1350 {
1351 	int rc = X86EMUL_CONTINUE;
1352 
1353 	op->type = OP_MEM;
1354 	switch (ctxt->ad_bytes) {
1355 	case 2:
1356 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1357 		break;
1358 	case 4:
1359 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1360 		break;
1361 	case 8:
1362 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1363 		break;
1364 	}
1365 done:
1366 	return rc;
1367 }
1368 
1369 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1370 {
1371 	long sv = 0, mask;
1372 
1373 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1374 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1375 
1376 		if (ctxt->src.bytes == 2)
1377 			sv = (s16)ctxt->src.val & (s16)mask;
1378 		else if (ctxt->src.bytes == 4)
1379 			sv = (s32)ctxt->src.val & (s32)mask;
1380 		else
1381 			sv = (s64)ctxt->src.val & (s64)mask;
1382 
1383 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1384 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1385 	}
1386 
1387 	/* only subword offset */
1388 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1389 }
1390 
1391 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1392 			 unsigned long addr, void *dest, unsigned size)
1393 {
1394 	int rc;
1395 	struct read_cache *mc = &ctxt->mem_read;
1396 
1397 	if (mc->pos < mc->end)
1398 		goto read_cached;
1399 
1400 	WARN_ON((mc->end + size) >= sizeof(mc->data));
1401 
1402 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1403 				      &ctxt->exception);
1404 	if (rc != X86EMUL_CONTINUE)
1405 		return rc;
1406 
1407 	mc->end += size;
1408 
1409 read_cached:
1410 	memcpy(dest, mc->data + mc->pos, size);
1411 	mc->pos += size;
1412 	return X86EMUL_CONTINUE;
1413 }
1414 
1415 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1416 			  struct segmented_address addr,
1417 			  void *data,
1418 			  unsigned size)
1419 {
1420 	int rc;
1421 	ulong linear;
1422 
1423 	rc = linearize(ctxt, addr, size, false, &linear);
1424 	if (rc != X86EMUL_CONTINUE)
1425 		return rc;
1426 	return read_emulated(ctxt, linear, data, size);
1427 }
1428 
1429 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1430 			   struct segmented_address addr,
1431 			   const void *data,
1432 			   unsigned size)
1433 {
1434 	int rc;
1435 	ulong linear;
1436 
1437 	rc = linearize(ctxt, addr, size, true, &linear);
1438 	if (rc != X86EMUL_CONTINUE)
1439 		return rc;
1440 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1441 					 &ctxt->exception);
1442 }
1443 
1444 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1445 			     struct segmented_address addr,
1446 			     const void *orig_data, const void *data,
1447 			     unsigned size)
1448 {
1449 	int rc;
1450 	ulong linear;
1451 
1452 	rc = linearize(ctxt, addr, size, true, &linear);
1453 	if (rc != X86EMUL_CONTINUE)
1454 		return rc;
1455 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1456 					   size, &ctxt->exception);
1457 }
1458 
1459 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1460 			   unsigned int size, unsigned short port,
1461 			   void *dest)
1462 {
1463 	struct read_cache *rc = &ctxt->io_read;
1464 
1465 	if (rc->pos == rc->end) { /* refill pio read ahead */
1466 		unsigned int in_page, n;
1467 		unsigned int count = ctxt->rep_prefix ?
1468 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1469 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1470 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1471 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1472 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1473 		if (n == 0)
1474 			n = 1;
1475 		rc->pos = rc->end = 0;
1476 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1477 			return 0;
1478 		rc->end = n * size;
1479 	}
1480 
1481 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1482 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1483 		ctxt->dst.data = rc->data + rc->pos;
1484 		ctxt->dst.type = OP_MEM_STR;
1485 		ctxt->dst.count = (rc->end - rc->pos) / size;
1486 		rc->pos = rc->end;
1487 	} else {
1488 		memcpy(dest, rc->data + rc->pos, size);
1489 		rc->pos += size;
1490 	}
1491 	return 1;
1492 }
1493 
1494 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1495 				     u16 index, struct desc_struct *desc)
1496 {
1497 	struct desc_ptr dt;
1498 	ulong addr;
1499 
1500 	ctxt->ops->get_idt(ctxt, &dt);
1501 
1502 	if (dt.size < index * 8 + 7)
1503 		return emulate_gp(ctxt, index << 3 | 0x2);
1504 
1505 	addr = dt.address + index * 8;
1506 	return ctxt->ops->read_std(ctxt, addr, desc, sizeof *desc,
1507 				   &ctxt->exception);
1508 }
1509 
1510 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1511 				     u16 selector, struct desc_ptr *dt)
1512 {
1513 	const struct x86_emulate_ops *ops = ctxt->ops;
1514 	u32 base3 = 0;
1515 
1516 	if (selector & 1 << 2) {
1517 		struct desc_struct desc;
1518 		u16 sel;
1519 
1520 		memset (dt, 0, sizeof *dt);
1521 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1522 				      VCPU_SREG_LDTR))
1523 			return;
1524 
1525 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1526 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1527 	} else
1528 		ops->get_gdt(ctxt, dt);
1529 }
1530 
1531 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1532 			      u16 selector, ulong *desc_addr_p)
1533 {
1534 	struct desc_ptr dt;
1535 	u16 index = selector >> 3;
1536 	ulong addr;
1537 
1538 	get_descriptor_table_ptr(ctxt, selector, &dt);
1539 
1540 	if (dt.size < index * 8 + 7)
1541 		return emulate_gp(ctxt, selector & 0xfffc);
1542 
1543 	addr = dt.address + index * 8;
1544 
1545 #ifdef CONFIG_X86_64
1546 	if (addr >> 32 != 0) {
1547 		u64 efer = 0;
1548 
1549 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1550 		if (!(efer & EFER_LMA))
1551 			addr &= (u32)-1;
1552 	}
1553 #endif
1554 
1555 	*desc_addr_p = addr;
1556 	return X86EMUL_CONTINUE;
1557 }
1558 
1559 /* allowed just for 8 bytes segments */
1560 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1561 				   u16 selector, struct desc_struct *desc,
1562 				   ulong *desc_addr_p)
1563 {
1564 	int rc;
1565 
1566 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1567 	if (rc != X86EMUL_CONTINUE)
1568 		return rc;
1569 
1570 	return ctxt->ops->read_std(ctxt, *desc_addr_p, desc, sizeof(*desc),
1571 				   &ctxt->exception);
1572 }
1573 
1574 /* allowed just for 8 bytes segments */
1575 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1576 				    u16 selector, struct desc_struct *desc)
1577 {
1578 	int rc;
1579 	ulong addr;
1580 
1581 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1582 	if (rc != X86EMUL_CONTINUE)
1583 		return rc;
1584 
1585 	return ctxt->ops->write_std(ctxt, addr, desc, sizeof *desc,
1586 				    &ctxt->exception);
1587 }
1588 
1589 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1590 				     u16 selector, int seg, u8 cpl,
1591 				     enum x86_transfer_type transfer,
1592 				     struct desc_struct *desc)
1593 {
1594 	struct desc_struct seg_desc, old_desc;
1595 	u8 dpl, rpl;
1596 	unsigned err_vec = GP_VECTOR;
1597 	u32 err_code = 0;
1598 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1599 	ulong desc_addr;
1600 	int ret;
1601 	u16 dummy;
1602 	u32 base3 = 0;
1603 
1604 	memset(&seg_desc, 0, sizeof seg_desc);
1605 
1606 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1607 		/* set real mode segment descriptor (keep limit etc. for
1608 		 * unreal mode) */
1609 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1610 		set_desc_base(&seg_desc, selector << 4);
1611 		goto load;
1612 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1613 		/* VM86 needs a clean new segment descriptor */
1614 		set_desc_base(&seg_desc, selector << 4);
1615 		set_desc_limit(&seg_desc, 0xffff);
1616 		seg_desc.type = 3;
1617 		seg_desc.p = 1;
1618 		seg_desc.s = 1;
1619 		seg_desc.dpl = 3;
1620 		goto load;
1621 	}
1622 
1623 	rpl = selector & 3;
1624 
1625 	/* TR should be in GDT only */
1626 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1627 		goto exception;
1628 
1629 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1630 	if (null_selector) {
1631 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1632 			goto exception;
1633 
1634 		if (seg == VCPU_SREG_SS) {
1635 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1636 				goto exception;
1637 
1638 			/*
1639 			 * ctxt->ops->set_segment expects the CPL to be in
1640 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1641 			 */
1642 			seg_desc.type = 3;
1643 			seg_desc.p = 1;
1644 			seg_desc.s = 1;
1645 			seg_desc.dpl = cpl;
1646 			seg_desc.d = 1;
1647 			seg_desc.g = 1;
1648 		}
1649 
1650 		/* Skip all following checks */
1651 		goto load;
1652 	}
1653 
1654 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1655 	if (ret != X86EMUL_CONTINUE)
1656 		return ret;
1657 
1658 	err_code = selector & 0xfffc;
1659 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1660 							   GP_VECTOR;
1661 
1662 	/* can't load system descriptor into segment selector */
1663 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1664 		if (transfer == X86_TRANSFER_CALL_JMP)
1665 			return X86EMUL_UNHANDLEABLE;
1666 		goto exception;
1667 	}
1668 
1669 	if (!seg_desc.p) {
1670 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1671 		goto exception;
1672 	}
1673 
1674 	dpl = seg_desc.dpl;
1675 
1676 	switch (seg) {
1677 	case VCPU_SREG_SS:
1678 		/*
1679 		 * segment is not a writable data segment or segment
1680 		 * selector's RPL != CPL or segment selector's RPL != CPL
1681 		 */
1682 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1683 			goto exception;
1684 		break;
1685 	case VCPU_SREG_CS:
1686 		if (!(seg_desc.type & 8))
1687 			goto exception;
1688 
1689 		if (seg_desc.type & 4) {
1690 			/* conforming */
1691 			if (dpl > cpl)
1692 				goto exception;
1693 		} else {
1694 			/* nonconforming */
1695 			if (rpl > cpl || dpl != cpl)
1696 				goto exception;
1697 		}
1698 		/* in long-mode d/b must be clear if l is set */
1699 		if (seg_desc.d && seg_desc.l) {
1700 			u64 efer = 0;
1701 
1702 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1703 			if (efer & EFER_LMA)
1704 				goto exception;
1705 		}
1706 
1707 		/* CS(RPL) <- CPL */
1708 		selector = (selector & 0xfffc) | cpl;
1709 		break;
1710 	case VCPU_SREG_TR:
1711 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1712 			goto exception;
1713 		old_desc = seg_desc;
1714 		seg_desc.type |= 2; /* busy */
1715 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1716 						  sizeof(seg_desc), &ctxt->exception);
1717 		if (ret != X86EMUL_CONTINUE)
1718 			return ret;
1719 		break;
1720 	case VCPU_SREG_LDTR:
1721 		if (seg_desc.s || seg_desc.type != 2)
1722 			goto exception;
1723 		break;
1724 	default: /*  DS, ES, FS, or GS */
1725 		/*
1726 		 * segment is not a data or readable code segment or
1727 		 * ((segment is a data or nonconforming code segment)
1728 		 * and (both RPL and CPL > DPL))
1729 		 */
1730 		if ((seg_desc.type & 0xa) == 0x8 ||
1731 		    (((seg_desc.type & 0xc) != 0xc) &&
1732 		     (rpl > dpl && cpl > dpl)))
1733 			goto exception;
1734 		break;
1735 	}
1736 
1737 	if (seg_desc.s) {
1738 		/* mark segment as accessed */
1739 		if (!(seg_desc.type & 1)) {
1740 			seg_desc.type |= 1;
1741 			ret = write_segment_descriptor(ctxt, selector,
1742 						       &seg_desc);
1743 			if (ret != X86EMUL_CONTINUE)
1744 				return ret;
1745 		}
1746 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1747 		ret = ctxt->ops->read_std(ctxt, desc_addr+8, &base3,
1748 				sizeof(base3), &ctxt->exception);
1749 		if (ret != X86EMUL_CONTINUE)
1750 			return ret;
1751 		if (is_noncanonical_address(get_desc_base(&seg_desc) |
1752 					     ((u64)base3 << 32)))
1753 			return emulate_gp(ctxt, 0);
1754 	}
1755 load:
1756 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1757 	if (desc)
1758 		*desc = seg_desc;
1759 	return X86EMUL_CONTINUE;
1760 exception:
1761 	return emulate_exception(ctxt, err_vec, err_code, true);
1762 }
1763 
1764 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1765 				   u16 selector, int seg)
1766 {
1767 	u8 cpl = ctxt->ops->cpl(ctxt);
1768 
1769 	/*
1770 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1771 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1772 	 * but it's wrong).
1773 	 *
1774 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1775 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1776 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1777 	 * and only forbid it here.
1778 	 */
1779 	if (seg == VCPU_SREG_SS && selector == 3 &&
1780 	    ctxt->mode == X86EMUL_MODE_PROT64)
1781 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1782 
1783 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1784 					 X86_TRANSFER_NONE, NULL);
1785 }
1786 
1787 static void write_register_operand(struct operand *op)
1788 {
1789 	return assign_register(op->addr.reg, op->val, op->bytes);
1790 }
1791 
1792 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1793 {
1794 	switch (op->type) {
1795 	case OP_REG:
1796 		write_register_operand(op);
1797 		break;
1798 	case OP_MEM:
1799 		if (ctxt->lock_prefix)
1800 			return segmented_cmpxchg(ctxt,
1801 						 op->addr.mem,
1802 						 &op->orig_val,
1803 						 &op->val,
1804 						 op->bytes);
1805 		else
1806 			return segmented_write(ctxt,
1807 					       op->addr.mem,
1808 					       &op->val,
1809 					       op->bytes);
1810 		break;
1811 	case OP_MEM_STR:
1812 		return segmented_write(ctxt,
1813 				       op->addr.mem,
1814 				       op->data,
1815 				       op->bytes * op->count);
1816 		break;
1817 	case OP_XMM:
1818 		write_sse_reg(ctxt, &op->vec_val, op->addr.xmm);
1819 		break;
1820 	case OP_MM:
1821 		write_mmx_reg(ctxt, &op->mm_val, op->addr.mm);
1822 		break;
1823 	case OP_NONE:
1824 		/* no writeback */
1825 		break;
1826 	default:
1827 		break;
1828 	}
1829 	return X86EMUL_CONTINUE;
1830 }
1831 
1832 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1833 {
1834 	struct segmented_address addr;
1835 
1836 	rsp_increment(ctxt, -bytes);
1837 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1838 	addr.seg = VCPU_SREG_SS;
1839 
1840 	return segmented_write(ctxt, addr, data, bytes);
1841 }
1842 
1843 static int em_push(struct x86_emulate_ctxt *ctxt)
1844 {
1845 	/* Disable writeback. */
1846 	ctxt->dst.type = OP_NONE;
1847 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1848 }
1849 
1850 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1851 		       void *dest, int len)
1852 {
1853 	int rc;
1854 	struct segmented_address addr;
1855 
1856 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1857 	addr.seg = VCPU_SREG_SS;
1858 	rc = segmented_read(ctxt, addr, dest, len);
1859 	if (rc != X86EMUL_CONTINUE)
1860 		return rc;
1861 
1862 	rsp_increment(ctxt, len);
1863 	return rc;
1864 }
1865 
1866 static int em_pop(struct x86_emulate_ctxt *ctxt)
1867 {
1868 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1869 }
1870 
1871 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1872 			void *dest, int len)
1873 {
1874 	int rc;
1875 	unsigned long val, change_mask;
1876 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1877 	int cpl = ctxt->ops->cpl(ctxt);
1878 
1879 	rc = emulate_pop(ctxt, &val, len);
1880 	if (rc != X86EMUL_CONTINUE)
1881 		return rc;
1882 
1883 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1884 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1885 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1886 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1887 
1888 	switch(ctxt->mode) {
1889 	case X86EMUL_MODE_PROT64:
1890 	case X86EMUL_MODE_PROT32:
1891 	case X86EMUL_MODE_PROT16:
1892 		if (cpl == 0)
1893 			change_mask |= X86_EFLAGS_IOPL;
1894 		if (cpl <= iopl)
1895 			change_mask |= X86_EFLAGS_IF;
1896 		break;
1897 	case X86EMUL_MODE_VM86:
1898 		if (iopl < 3)
1899 			return emulate_gp(ctxt, 0);
1900 		change_mask |= X86_EFLAGS_IF;
1901 		break;
1902 	default: /* real mode */
1903 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1904 		break;
1905 	}
1906 
1907 	*(unsigned long *)dest =
1908 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1909 
1910 	return rc;
1911 }
1912 
1913 static int em_popf(struct x86_emulate_ctxt *ctxt)
1914 {
1915 	ctxt->dst.type = OP_REG;
1916 	ctxt->dst.addr.reg = &ctxt->eflags;
1917 	ctxt->dst.bytes = ctxt->op_bytes;
1918 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1919 }
1920 
1921 static int em_enter(struct x86_emulate_ctxt *ctxt)
1922 {
1923 	int rc;
1924 	unsigned frame_size = ctxt->src.val;
1925 	unsigned nesting_level = ctxt->src2.val & 31;
1926 	ulong rbp;
1927 
1928 	if (nesting_level)
1929 		return X86EMUL_UNHANDLEABLE;
1930 
1931 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1932 	rc = push(ctxt, &rbp, stack_size(ctxt));
1933 	if (rc != X86EMUL_CONTINUE)
1934 		return rc;
1935 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1936 		      stack_mask(ctxt));
1937 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1938 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1939 		      stack_mask(ctxt));
1940 	return X86EMUL_CONTINUE;
1941 }
1942 
1943 static int em_leave(struct x86_emulate_ctxt *ctxt)
1944 {
1945 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1946 		      stack_mask(ctxt));
1947 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1948 }
1949 
1950 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1951 {
1952 	int seg = ctxt->src2.val;
1953 
1954 	ctxt->src.val = get_segment_selector(ctxt, seg);
1955 	if (ctxt->op_bytes == 4) {
1956 		rsp_increment(ctxt, -2);
1957 		ctxt->op_bytes = 2;
1958 	}
1959 
1960 	return em_push(ctxt);
1961 }
1962 
1963 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1964 {
1965 	int seg = ctxt->src2.val;
1966 	unsigned long selector;
1967 	int rc;
1968 
1969 	rc = emulate_pop(ctxt, &selector, 2);
1970 	if (rc != X86EMUL_CONTINUE)
1971 		return rc;
1972 
1973 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1974 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1975 	if (ctxt->op_bytes > 2)
1976 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1977 
1978 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1979 	return rc;
1980 }
1981 
1982 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1983 {
1984 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1985 	int rc = X86EMUL_CONTINUE;
1986 	int reg = VCPU_REGS_RAX;
1987 
1988 	while (reg <= VCPU_REGS_RDI) {
1989 		(reg == VCPU_REGS_RSP) ?
1990 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1991 
1992 		rc = em_push(ctxt);
1993 		if (rc != X86EMUL_CONTINUE)
1994 			return rc;
1995 
1996 		++reg;
1997 	}
1998 
1999 	return rc;
2000 }
2001 
2002 static int em_pushf(struct x86_emulate_ctxt *ctxt)
2003 {
2004 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
2005 	return em_push(ctxt);
2006 }
2007 
2008 static int em_popa(struct x86_emulate_ctxt *ctxt)
2009 {
2010 	int rc = X86EMUL_CONTINUE;
2011 	int reg = VCPU_REGS_RDI;
2012 	u32 val;
2013 
2014 	while (reg >= VCPU_REGS_RAX) {
2015 		if (reg == VCPU_REGS_RSP) {
2016 			rsp_increment(ctxt, ctxt->op_bytes);
2017 			--reg;
2018 		}
2019 
2020 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2021 		if (rc != X86EMUL_CONTINUE)
2022 			break;
2023 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2024 		--reg;
2025 	}
2026 	return rc;
2027 }
2028 
2029 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2030 {
2031 	const struct x86_emulate_ops *ops = ctxt->ops;
2032 	int rc;
2033 	struct desc_ptr dt;
2034 	gva_t cs_addr;
2035 	gva_t eip_addr;
2036 	u16 cs, eip;
2037 
2038 	/* TODO: Add limit checks */
2039 	ctxt->src.val = ctxt->eflags;
2040 	rc = em_push(ctxt);
2041 	if (rc != X86EMUL_CONTINUE)
2042 		return rc;
2043 
2044 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2045 
2046 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2047 	rc = em_push(ctxt);
2048 	if (rc != X86EMUL_CONTINUE)
2049 		return rc;
2050 
2051 	ctxt->src.val = ctxt->_eip;
2052 	rc = em_push(ctxt);
2053 	if (rc != X86EMUL_CONTINUE)
2054 		return rc;
2055 
2056 	ops->get_idt(ctxt, &dt);
2057 
2058 	eip_addr = dt.address + (irq << 2);
2059 	cs_addr = dt.address + (irq << 2) + 2;
2060 
2061 	rc = ops->read_std(ctxt, cs_addr, &cs, 2, &ctxt->exception);
2062 	if (rc != X86EMUL_CONTINUE)
2063 		return rc;
2064 
2065 	rc = ops->read_std(ctxt, eip_addr, &eip, 2, &ctxt->exception);
2066 	if (rc != X86EMUL_CONTINUE)
2067 		return rc;
2068 
2069 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2070 	if (rc != X86EMUL_CONTINUE)
2071 		return rc;
2072 
2073 	ctxt->_eip = eip;
2074 
2075 	return rc;
2076 }
2077 
2078 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2079 {
2080 	int rc;
2081 
2082 	invalidate_registers(ctxt);
2083 	rc = __emulate_int_real(ctxt, irq);
2084 	if (rc == X86EMUL_CONTINUE)
2085 		writeback_registers(ctxt);
2086 	return rc;
2087 }
2088 
2089 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2090 {
2091 	switch(ctxt->mode) {
2092 	case X86EMUL_MODE_REAL:
2093 		return __emulate_int_real(ctxt, irq);
2094 	case X86EMUL_MODE_VM86:
2095 	case X86EMUL_MODE_PROT16:
2096 	case X86EMUL_MODE_PROT32:
2097 	case X86EMUL_MODE_PROT64:
2098 	default:
2099 		/* Protected mode interrupts unimplemented yet */
2100 		return X86EMUL_UNHANDLEABLE;
2101 	}
2102 }
2103 
2104 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2105 {
2106 	int rc = X86EMUL_CONTINUE;
2107 	unsigned long temp_eip = 0;
2108 	unsigned long temp_eflags = 0;
2109 	unsigned long cs = 0;
2110 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2111 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2112 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2113 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2114 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2115 			     X86_EFLAGS_FIXED;
2116 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2117 				  X86_EFLAGS_VIP;
2118 
2119 	/* TODO: Add stack limit check */
2120 
2121 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2122 
2123 	if (rc != X86EMUL_CONTINUE)
2124 		return rc;
2125 
2126 	if (temp_eip & ~0xffff)
2127 		return emulate_gp(ctxt, 0);
2128 
2129 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2130 
2131 	if (rc != X86EMUL_CONTINUE)
2132 		return rc;
2133 
2134 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2135 
2136 	if (rc != X86EMUL_CONTINUE)
2137 		return rc;
2138 
2139 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2140 
2141 	if (rc != X86EMUL_CONTINUE)
2142 		return rc;
2143 
2144 	ctxt->_eip = temp_eip;
2145 
2146 	if (ctxt->op_bytes == 4)
2147 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2148 	else if (ctxt->op_bytes == 2) {
2149 		ctxt->eflags &= ~0xffff;
2150 		ctxt->eflags |= temp_eflags;
2151 	}
2152 
2153 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2154 	ctxt->eflags |= X86_EFLAGS_FIXED;
2155 	ctxt->ops->set_nmi_mask(ctxt, false);
2156 
2157 	return rc;
2158 }
2159 
2160 static int em_iret(struct x86_emulate_ctxt *ctxt)
2161 {
2162 	switch(ctxt->mode) {
2163 	case X86EMUL_MODE_REAL:
2164 		return emulate_iret_real(ctxt);
2165 	case X86EMUL_MODE_VM86:
2166 	case X86EMUL_MODE_PROT16:
2167 	case X86EMUL_MODE_PROT32:
2168 	case X86EMUL_MODE_PROT64:
2169 	default:
2170 		/* iret from protected mode unimplemented yet */
2171 		return X86EMUL_UNHANDLEABLE;
2172 	}
2173 }
2174 
2175 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2176 {
2177 	int rc;
2178 	unsigned short sel;
2179 	struct desc_struct new_desc;
2180 	u8 cpl = ctxt->ops->cpl(ctxt);
2181 
2182 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2183 
2184 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2185 				       X86_TRANSFER_CALL_JMP,
2186 				       &new_desc);
2187 	if (rc != X86EMUL_CONTINUE)
2188 		return rc;
2189 
2190 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2191 	/* Error handling is not implemented. */
2192 	if (rc != X86EMUL_CONTINUE)
2193 		return X86EMUL_UNHANDLEABLE;
2194 
2195 	return rc;
2196 }
2197 
2198 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2199 {
2200 	return assign_eip_near(ctxt, ctxt->src.val);
2201 }
2202 
2203 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2204 {
2205 	int rc;
2206 	long int old_eip;
2207 
2208 	old_eip = ctxt->_eip;
2209 	rc = assign_eip_near(ctxt, ctxt->src.val);
2210 	if (rc != X86EMUL_CONTINUE)
2211 		return rc;
2212 	ctxt->src.val = old_eip;
2213 	rc = em_push(ctxt);
2214 	return rc;
2215 }
2216 
2217 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2218 {
2219 	u64 old = ctxt->dst.orig_val64;
2220 
2221 	if (ctxt->dst.bytes == 16)
2222 		return X86EMUL_UNHANDLEABLE;
2223 
2224 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2225 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2226 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2227 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2228 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2229 	} else {
2230 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2231 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2232 
2233 		ctxt->eflags |= X86_EFLAGS_ZF;
2234 	}
2235 	return X86EMUL_CONTINUE;
2236 }
2237 
2238 static int em_ret(struct x86_emulate_ctxt *ctxt)
2239 {
2240 	int rc;
2241 	unsigned long eip;
2242 
2243 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2244 	if (rc != X86EMUL_CONTINUE)
2245 		return rc;
2246 
2247 	return assign_eip_near(ctxt, eip);
2248 }
2249 
2250 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2251 {
2252 	int rc;
2253 	unsigned long eip, cs;
2254 	int cpl = ctxt->ops->cpl(ctxt);
2255 	struct desc_struct new_desc;
2256 
2257 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2258 	if (rc != X86EMUL_CONTINUE)
2259 		return rc;
2260 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2261 	if (rc != X86EMUL_CONTINUE)
2262 		return rc;
2263 	/* Outer-privilege level return is not implemented */
2264 	if (ctxt->mode >= X86EMUL_MODE_PROT16 && (cs & 3) > cpl)
2265 		return X86EMUL_UNHANDLEABLE;
2266 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2267 				       X86_TRANSFER_RET,
2268 				       &new_desc);
2269 	if (rc != X86EMUL_CONTINUE)
2270 		return rc;
2271 	rc = assign_eip_far(ctxt, eip, &new_desc);
2272 	/* Error handling is not implemented. */
2273 	if (rc != X86EMUL_CONTINUE)
2274 		return X86EMUL_UNHANDLEABLE;
2275 
2276 	return rc;
2277 }
2278 
2279 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2280 {
2281         int rc;
2282 
2283         rc = em_ret_far(ctxt);
2284         if (rc != X86EMUL_CONTINUE)
2285                 return rc;
2286         rsp_increment(ctxt, ctxt->src.val);
2287         return X86EMUL_CONTINUE;
2288 }
2289 
2290 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2291 {
2292 	/* Save real source value, then compare EAX against destination. */
2293 	ctxt->dst.orig_val = ctxt->dst.val;
2294 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2295 	ctxt->src.orig_val = ctxt->src.val;
2296 	ctxt->src.val = ctxt->dst.orig_val;
2297 	fastop(ctxt, em_cmp);
2298 
2299 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2300 		/* Success: write back to memory; no update of EAX */
2301 		ctxt->src.type = OP_NONE;
2302 		ctxt->dst.val = ctxt->src.orig_val;
2303 	} else {
2304 		/* Failure: write the value we saw to EAX. */
2305 		ctxt->src.type = OP_REG;
2306 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2307 		ctxt->src.val = ctxt->dst.orig_val;
2308 		/* Create write-cycle to dest by writing the same value */
2309 		ctxt->dst.val = ctxt->dst.orig_val;
2310 	}
2311 	return X86EMUL_CONTINUE;
2312 }
2313 
2314 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2315 {
2316 	int seg = ctxt->src2.val;
2317 	unsigned short sel;
2318 	int rc;
2319 
2320 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2321 
2322 	rc = load_segment_descriptor(ctxt, sel, seg);
2323 	if (rc != X86EMUL_CONTINUE)
2324 		return rc;
2325 
2326 	ctxt->dst.val = ctxt->src.val;
2327 	return rc;
2328 }
2329 
2330 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2331 {
2332 	u32 eax, ebx, ecx, edx;
2333 
2334 	eax = 0x80000001;
2335 	ecx = 0;
2336 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
2337 	return edx & bit(X86_FEATURE_LM);
2338 }
2339 
2340 #define GET_SMSTATE(type, smbase, offset)				  \
2341 	({								  \
2342 	 type __val;							  \
2343 	 int r = ctxt->ops->read_phys(ctxt, smbase + offset, &__val,      \
2344 				      sizeof(__val));			  \
2345 	 if (r != X86EMUL_CONTINUE)					  \
2346 		 return X86EMUL_UNHANDLEABLE;				  \
2347 	 __val;								  \
2348 	})
2349 
2350 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2351 {
2352 	desc->g    = (flags >> 23) & 1;
2353 	desc->d    = (flags >> 22) & 1;
2354 	desc->l    = (flags >> 21) & 1;
2355 	desc->avl  = (flags >> 20) & 1;
2356 	desc->p    = (flags >> 15) & 1;
2357 	desc->dpl  = (flags >> 13) & 3;
2358 	desc->s    = (flags >> 12) & 1;
2359 	desc->type = (flags >>  8) & 15;
2360 }
2361 
2362 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, u64 smbase, int n)
2363 {
2364 	struct desc_struct desc;
2365 	int offset;
2366 	u16 selector;
2367 
2368 	selector = GET_SMSTATE(u32, smbase, 0x7fa8 + n * 4);
2369 
2370 	if (n < 3)
2371 		offset = 0x7f84 + n * 12;
2372 	else
2373 		offset = 0x7f2c + (n - 3) * 12;
2374 
2375 	set_desc_base(&desc,      GET_SMSTATE(u32, smbase, offset + 8));
2376 	set_desc_limit(&desc,     GET_SMSTATE(u32, smbase, offset + 4));
2377 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smbase, offset));
2378 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2379 	return X86EMUL_CONTINUE;
2380 }
2381 
2382 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, u64 smbase, int n)
2383 {
2384 	struct desc_struct desc;
2385 	int offset;
2386 	u16 selector;
2387 	u32 base3;
2388 
2389 	offset = 0x7e00 + n * 16;
2390 
2391 	selector =                GET_SMSTATE(u16, smbase, offset);
2392 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smbase, offset + 2) << 8);
2393 	set_desc_limit(&desc,     GET_SMSTATE(u32, smbase, offset + 4));
2394 	set_desc_base(&desc,      GET_SMSTATE(u32, smbase, offset + 8));
2395 	base3 =                   GET_SMSTATE(u32, smbase, offset + 12);
2396 
2397 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2398 	return X86EMUL_CONTINUE;
2399 }
2400 
2401 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2402 				     u64 cr0, u64 cr4)
2403 {
2404 	int bad;
2405 
2406 	/*
2407 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2408 	 * Then enable protected mode.	However, PCID cannot be enabled
2409 	 * if EFER.LMA=0, so set it separately.
2410 	 */
2411 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2412 	if (bad)
2413 		return X86EMUL_UNHANDLEABLE;
2414 
2415 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2416 	if (bad)
2417 		return X86EMUL_UNHANDLEABLE;
2418 
2419 	if (cr4 & X86_CR4_PCIDE) {
2420 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2421 		if (bad)
2422 			return X86EMUL_UNHANDLEABLE;
2423 	}
2424 
2425 	return X86EMUL_CONTINUE;
2426 }
2427 
2428 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt, u64 smbase)
2429 {
2430 	struct desc_struct desc;
2431 	struct desc_ptr dt;
2432 	u16 selector;
2433 	u32 val, cr0, cr4;
2434 	int i;
2435 
2436 	cr0 =                      GET_SMSTATE(u32, smbase, 0x7ffc);
2437 	ctxt->ops->set_cr(ctxt, 3, GET_SMSTATE(u32, smbase, 0x7ff8));
2438 	ctxt->eflags =             GET_SMSTATE(u32, smbase, 0x7ff4) | X86_EFLAGS_FIXED;
2439 	ctxt->_eip =               GET_SMSTATE(u32, smbase, 0x7ff0);
2440 
2441 	for (i = 0; i < 8; i++)
2442 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smbase, 0x7fd0 + i * 4);
2443 
2444 	val = GET_SMSTATE(u32, smbase, 0x7fcc);
2445 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2446 	val = GET_SMSTATE(u32, smbase, 0x7fc8);
2447 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2448 
2449 	selector =                 GET_SMSTATE(u32, smbase, 0x7fc4);
2450 	set_desc_base(&desc,       GET_SMSTATE(u32, smbase, 0x7f64));
2451 	set_desc_limit(&desc,      GET_SMSTATE(u32, smbase, 0x7f60));
2452 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smbase, 0x7f5c));
2453 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2454 
2455 	selector =                 GET_SMSTATE(u32, smbase, 0x7fc0);
2456 	set_desc_base(&desc,       GET_SMSTATE(u32, smbase, 0x7f80));
2457 	set_desc_limit(&desc,      GET_SMSTATE(u32, smbase, 0x7f7c));
2458 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smbase, 0x7f78));
2459 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2460 
2461 	dt.address =               GET_SMSTATE(u32, smbase, 0x7f74);
2462 	dt.size =                  GET_SMSTATE(u32, smbase, 0x7f70);
2463 	ctxt->ops->set_gdt(ctxt, &dt);
2464 
2465 	dt.address =               GET_SMSTATE(u32, smbase, 0x7f58);
2466 	dt.size =                  GET_SMSTATE(u32, smbase, 0x7f54);
2467 	ctxt->ops->set_idt(ctxt, &dt);
2468 
2469 	for (i = 0; i < 6; i++) {
2470 		int r = rsm_load_seg_32(ctxt, smbase, i);
2471 		if (r != X86EMUL_CONTINUE)
2472 			return r;
2473 	}
2474 
2475 	cr4 = GET_SMSTATE(u32, smbase, 0x7f14);
2476 
2477 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smbase, 0x7ef8));
2478 
2479 	return rsm_enter_protected_mode(ctxt, cr0, cr4);
2480 }
2481 
2482 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt, u64 smbase)
2483 {
2484 	struct desc_struct desc;
2485 	struct desc_ptr dt;
2486 	u64 val, cr0, cr4;
2487 	u32 base3;
2488 	u16 selector;
2489 	int i, r;
2490 
2491 	for (i = 0; i < 16; i++)
2492 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smbase, 0x7ff8 - i * 8);
2493 
2494 	ctxt->_eip   = GET_SMSTATE(u64, smbase, 0x7f78);
2495 	ctxt->eflags = GET_SMSTATE(u32, smbase, 0x7f70) | X86_EFLAGS_FIXED;
2496 
2497 	val = GET_SMSTATE(u32, smbase, 0x7f68);
2498 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2499 	val = GET_SMSTATE(u32, smbase, 0x7f60);
2500 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2501 
2502 	cr0 =                       GET_SMSTATE(u64, smbase, 0x7f58);
2503 	ctxt->ops->set_cr(ctxt, 3,  GET_SMSTATE(u64, smbase, 0x7f50));
2504 	cr4 =                       GET_SMSTATE(u64, smbase, 0x7f48);
2505 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smbase, 0x7f00));
2506 	val =                       GET_SMSTATE(u64, smbase, 0x7ed0);
2507 	ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA);
2508 
2509 	selector =                  GET_SMSTATE(u32, smbase, 0x7e90);
2510 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smbase, 0x7e92) << 8);
2511 	set_desc_limit(&desc,       GET_SMSTATE(u32, smbase, 0x7e94));
2512 	set_desc_base(&desc,        GET_SMSTATE(u32, smbase, 0x7e98));
2513 	base3 =                     GET_SMSTATE(u32, smbase, 0x7e9c);
2514 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2515 
2516 	dt.size =                   GET_SMSTATE(u32, smbase, 0x7e84);
2517 	dt.address =                GET_SMSTATE(u64, smbase, 0x7e88);
2518 	ctxt->ops->set_idt(ctxt, &dt);
2519 
2520 	selector =                  GET_SMSTATE(u32, smbase, 0x7e70);
2521 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smbase, 0x7e72) << 8);
2522 	set_desc_limit(&desc,       GET_SMSTATE(u32, smbase, 0x7e74));
2523 	set_desc_base(&desc,        GET_SMSTATE(u32, smbase, 0x7e78));
2524 	base3 =                     GET_SMSTATE(u32, smbase, 0x7e7c);
2525 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2526 
2527 	dt.size =                   GET_SMSTATE(u32, smbase, 0x7e64);
2528 	dt.address =                GET_SMSTATE(u64, smbase, 0x7e68);
2529 	ctxt->ops->set_gdt(ctxt, &dt);
2530 
2531 	r = rsm_enter_protected_mode(ctxt, cr0, cr4);
2532 	if (r != X86EMUL_CONTINUE)
2533 		return r;
2534 
2535 	for (i = 0; i < 6; i++) {
2536 		r = rsm_load_seg_64(ctxt, smbase, i);
2537 		if (r != X86EMUL_CONTINUE)
2538 			return r;
2539 	}
2540 
2541 	return X86EMUL_CONTINUE;
2542 }
2543 
2544 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2545 {
2546 	unsigned long cr0, cr4, efer;
2547 	u64 smbase;
2548 	int ret;
2549 
2550 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2551 		return emulate_ud(ctxt);
2552 
2553 	/*
2554 	 * Get back to real mode, to prepare a safe state in which to load
2555 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2556 	 * supports long mode.
2557 	 */
2558 	cr4 = ctxt->ops->get_cr(ctxt, 4);
2559 	if (emulator_has_longmode(ctxt)) {
2560 		struct desc_struct cs_desc;
2561 
2562 		/* Zero CR4.PCIDE before CR0.PG.  */
2563 		if (cr4 & X86_CR4_PCIDE) {
2564 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2565 			cr4 &= ~X86_CR4_PCIDE;
2566 		}
2567 
2568 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2569 		memset(&cs_desc, 0, sizeof(cs_desc));
2570 		cs_desc.type = 0xb;
2571 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2572 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2573 	}
2574 
2575 	/* For the 64-bit case, this will clear EFER.LMA.  */
2576 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2577 	if (cr0 & X86_CR0_PE)
2578 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2579 
2580 	/* Now clear CR4.PAE (which must be done before clearing EFER.LME).  */
2581 	if (cr4 & X86_CR4_PAE)
2582 		ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2583 
2584 	/* And finally go back to 32-bit mode.  */
2585 	efer = 0;
2586 	ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2587 
2588 	smbase = ctxt->ops->get_smbase(ctxt);
2589 	if (emulator_has_longmode(ctxt))
2590 		ret = rsm_load_state_64(ctxt, smbase + 0x8000);
2591 	else
2592 		ret = rsm_load_state_32(ctxt, smbase + 0x8000);
2593 
2594 	if (ret != X86EMUL_CONTINUE) {
2595 		/* FIXME: should triple fault */
2596 		return X86EMUL_UNHANDLEABLE;
2597 	}
2598 
2599 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2600 		ctxt->ops->set_nmi_mask(ctxt, false);
2601 
2602 	ctxt->ops->set_hflags(ctxt, ctxt->ops->get_hflags(ctxt) &
2603 		~(X86EMUL_SMM_INSIDE_NMI_MASK | X86EMUL_SMM_MASK));
2604 	return X86EMUL_CONTINUE;
2605 }
2606 
2607 static void
2608 setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
2609 			struct desc_struct *cs, struct desc_struct *ss)
2610 {
2611 	cs->l = 0;		/* will be adjusted later */
2612 	set_desc_base(cs, 0);	/* flat segment */
2613 	cs->g = 1;		/* 4kb granularity */
2614 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2615 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2616 	cs->s = 1;
2617 	cs->dpl = 0;		/* will be adjusted later */
2618 	cs->p = 1;
2619 	cs->d = 1;
2620 	cs->avl = 0;
2621 
2622 	set_desc_base(ss, 0);	/* flat segment */
2623 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2624 	ss->g = 1;		/* 4kb granularity */
2625 	ss->s = 1;
2626 	ss->type = 0x03;	/* Read/Write, Accessed */
2627 	ss->d = 1;		/* 32bit stack segment */
2628 	ss->dpl = 0;
2629 	ss->p = 1;
2630 	ss->l = 0;
2631 	ss->avl = 0;
2632 }
2633 
2634 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2635 {
2636 	u32 eax, ebx, ecx, edx;
2637 
2638 	eax = ecx = 0;
2639 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
2640 	return ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx
2641 		&& ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx
2642 		&& edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx;
2643 }
2644 
2645 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2646 {
2647 	const struct x86_emulate_ops *ops = ctxt->ops;
2648 	u32 eax, ebx, ecx, edx;
2649 
2650 	/*
2651 	 * syscall should always be enabled in longmode - so only become
2652 	 * vendor specific (cpuid) if other modes are active...
2653 	 */
2654 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2655 		return true;
2656 
2657 	eax = 0x00000000;
2658 	ecx = 0x00000000;
2659 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
2660 	/*
2661 	 * Intel ("GenuineIntel")
2662 	 * remark: Intel CPUs only support "syscall" in 64bit
2663 	 * longmode. Also an 64bit guest with a
2664 	 * 32bit compat-app running will #UD !! While this
2665 	 * behaviour can be fixed (by emulating) into AMD
2666 	 * response - CPUs of AMD can't behave like Intel.
2667 	 */
2668 	if (ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx &&
2669 	    ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx &&
2670 	    edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx)
2671 		return false;
2672 
2673 	/* AMD ("AuthenticAMD") */
2674 	if (ebx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ebx &&
2675 	    ecx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ecx &&
2676 	    edx == X86EMUL_CPUID_VENDOR_AuthenticAMD_edx)
2677 		return true;
2678 
2679 	/* AMD ("AMDisbetter!") */
2680 	if (ebx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ebx &&
2681 	    ecx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ecx &&
2682 	    edx == X86EMUL_CPUID_VENDOR_AMDisbetterI_edx)
2683 		return true;
2684 
2685 	/* default: (not Intel, not AMD), apply Intel's stricter rules... */
2686 	return false;
2687 }
2688 
2689 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2690 {
2691 	const struct x86_emulate_ops *ops = ctxt->ops;
2692 	struct desc_struct cs, ss;
2693 	u64 msr_data;
2694 	u16 cs_sel, ss_sel;
2695 	u64 efer = 0;
2696 
2697 	/* syscall is not available in real mode */
2698 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2699 	    ctxt->mode == X86EMUL_MODE_VM86)
2700 		return emulate_ud(ctxt);
2701 
2702 	if (!(em_syscall_is_enabled(ctxt)))
2703 		return emulate_ud(ctxt);
2704 
2705 	ops->get_msr(ctxt, MSR_EFER, &efer);
2706 	setup_syscalls_segments(ctxt, &cs, &ss);
2707 
2708 	if (!(efer & EFER_SCE))
2709 		return emulate_ud(ctxt);
2710 
2711 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2712 	msr_data >>= 32;
2713 	cs_sel = (u16)(msr_data & 0xfffc);
2714 	ss_sel = (u16)(msr_data + 8);
2715 
2716 	if (efer & EFER_LMA) {
2717 		cs.d = 0;
2718 		cs.l = 1;
2719 	}
2720 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2721 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2722 
2723 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2724 	if (efer & EFER_LMA) {
2725 #ifdef CONFIG_X86_64
2726 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2727 
2728 		ops->get_msr(ctxt,
2729 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2730 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2731 		ctxt->_eip = msr_data;
2732 
2733 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2734 		ctxt->eflags &= ~msr_data;
2735 		ctxt->eflags |= X86_EFLAGS_FIXED;
2736 #endif
2737 	} else {
2738 		/* legacy mode */
2739 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2740 		ctxt->_eip = (u32)msr_data;
2741 
2742 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2743 	}
2744 
2745 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2746 	return X86EMUL_CONTINUE;
2747 }
2748 
2749 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2750 {
2751 	const struct x86_emulate_ops *ops = ctxt->ops;
2752 	struct desc_struct cs, ss;
2753 	u64 msr_data;
2754 	u16 cs_sel, ss_sel;
2755 	u64 efer = 0;
2756 
2757 	ops->get_msr(ctxt, MSR_EFER, &efer);
2758 	/* inject #GP if in real mode */
2759 	if (ctxt->mode == X86EMUL_MODE_REAL)
2760 		return emulate_gp(ctxt, 0);
2761 
2762 	/*
2763 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2764 	 * mode).
2765 	 */
2766 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2767 	    && !vendor_intel(ctxt))
2768 		return emulate_ud(ctxt);
2769 
2770 	/* sysenter/sysexit have not been tested in 64bit mode. */
2771 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2772 		return X86EMUL_UNHANDLEABLE;
2773 
2774 	setup_syscalls_segments(ctxt, &cs, &ss);
2775 
2776 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2777 	if ((msr_data & 0xfffc) == 0x0)
2778 		return emulate_gp(ctxt, 0);
2779 
2780 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2781 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2782 	ss_sel = cs_sel + 8;
2783 	if (efer & EFER_LMA) {
2784 		cs.d = 0;
2785 		cs.l = 1;
2786 	}
2787 
2788 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2789 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2790 
2791 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2792 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2793 
2794 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2795 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2796 							      (u32)msr_data;
2797 
2798 	return X86EMUL_CONTINUE;
2799 }
2800 
2801 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2802 {
2803 	const struct x86_emulate_ops *ops = ctxt->ops;
2804 	struct desc_struct cs, ss;
2805 	u64 msr_data, rcx, rdx;
2806 	int usermode;
2807 	u16 cs_sel = 0, ss_sel = 0;
2808 
2809 	/* inject #GP if in real mode or Virtual 8086 mode */
2810 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2811 	    ctxt->mode == X86EMUL_MODE_VM86)
2812 		return emulate_gp(ctxt, 0);
2813 
2814 	setup_syscalls_segments(ctxt, &cs, &ss);
2815 
2816 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2817 		usermode = X86EMUL_MODE_PROT64;
2818 	else
2819 		usermode = X86EMUL_MODE_PROT32;
2820 
2821 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2822 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2823 
2824 	cs.dpl = 3;
2825 	ss.dpl = 3;
2826 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2827 	switch (usermode) {
2828 	case X86EMUL_MODE_PROT32:
2829 		cs_sel = (u16)(msr_data + 16);
2830 		if ((msr_data & 0xfffc) == 0x0)
2831 			return emulate_gp(ctxt, 0);
2832 		ss_sel = (u16)(msr_data + 24);
2833 		rcx = (u32)rcx;
2834 		rdx = (u32)rdx;
2835 		break;
2836 	case X86EMUL_MODE_PROT64:
2837 		cs_sel = (u16)(msr_data + 32);
2838 		if (msr_data == 0x0)
2839 			return emulate_gp(ctxt, 0);
2840 		ss_sel = cs_sel + 8;
2841 		cs.d = 0;
2842 		cs.l = 1;
2843 		if (is_noncanonical_address(rcx) ||
2844 		    is_noncanonical_address(rdx))
2845 			return emulate_gp(ctxt, 0);
2846 		break;
2847 	}
2848 	cs_sel |= SEGMENT_RPL_MASK;
2849 	ss_sel |= SEGMENT_RPL_MASK;
2850 
2851 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2852 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2853 
2854 	ctxt->_eip = rdx;
2855 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2856 
2857 	return X86EMUL_CONTINUE;
2858 }
2859 
2860 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2861 {
2862 	int iopl;
2863 	if (ctxt->mode == X86EMUL_MODE_REAL)
2864 		return false;
2865 	if (ctxt->mode == X86EMUL_MODE_VM86)
2866 		return true;
2867 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2868 	return ctxt->ops->cpl(ctxt) > iopl;
2869 }
2870 
2871 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2872 					    u16 port, u16 len)
2873 {
2874 	const struct x86_emulate_ops *ops = ctxt->ops;
2875 	struct desc_struct tr_seg;
2876 	u32 base3;
2877 	int r;
2878 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2879 	unsigned mask = (1 << len) - 1;
2880 	unsigned long base;
2881 
2882 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2883 	if (!tr_seg.p)
2884 		return false;
2885 	if (desc_limit_scaled(&tr_seg) < 103)
2886 		return false;
2887 	base = get_desc_base(&tr_seg);
2888 #ifdef CONFIG_X86_64
2889 	base |= ((u64)base3) << 32;
2890 #endif
2891 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL);
2892 	if (r != X86EMUL_CONTINUE)
2893 		return false;
2894 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2895 		return false;
2896 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL);
2897 	if (r != X86EMUL_CONTINUE)
2898 		return false;
2899 	if ((perm >> bit_idx) & mask)
2900 		return false;
2901 	return true;
2902 }
2903 
2904 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2905 				 u16 port, u16 len)
2906 {
2907 	if (ctxt->perm_ok)
2908 		return true;
2909 
2910 	if (emulator_bad_iopl(ctxt))
2911 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2912 			return false;
2913 
2914 	ctxt->perm_ok = true;
2915 
2916 	return true;
2917 }
2918 
2919 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2920 {
2921 	/*
2922 	 * Intel CPUs mask the counter and pointers in quite strange
2923 	 * manner when ECX is zero due to REP-string optimizations.
2924 	 */
2925 #ifdef CONFIG_X86_64
2926 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2927 		return;
2928 
2929 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2930 
2931 	switch (ctxt->b) {
2932 	case 0xa4:	/* movsb */
2933 	case 0xa5:	/* movsd/w */
2934 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2935 		/* fall through */
2936 	case 0xaa:	/* stosb */
2937 	case 0xab:	/* stosd/w */
2938 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2939 	}
2940 #endif
2941 }
2942 
2943 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2944 				struct tss_segment_16 *tss)
2945 {
2946 	tss->ip = ctxt->_eip;
2947 	tss->flag = ctxt->eflags;
2948 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2949 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2950 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2951 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2952 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2953 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2954 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2955 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2956 
2957 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2958 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2959 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2960 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2961 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2962 }
2963 
2964 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2965 				 struct tss_segment_16 *tss)
2966 {
2967 	int ret;
2968 	u8 cpl;
2969 
2970 	ctxt->_eip = tss->ip;
2971 	ctxt->eflags = tss->flag | 2;
2972 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2973 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2974 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2975 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2976 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2977 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2978 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2979 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2980 
2981 	/*
2982 	 * SDM says that segment selectors are loaded before segment
2983 	 * descriptors
2984 	 */
2985 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
2986 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2987 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2988 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2989 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2990 
2991 	cpl = tss->cs & 3;
2992 
2993 	/*
2994 	 * Now load segment descriptors. If fault happens at this stage
2995 	 * it is handled in a context of new task
2996 	 */
2997 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
2998 					X86_TRANSFER_TASK_SWITCH, NULL);
2999 	if (ret != X86EMUL_CONTINUE)
3000 		return ret;
3001 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3002 					X86_TRANSFER_TASK_SWITCH, NULL);
3003 	if (ret != X86EMUL_CONTINUE)
3004 		return ret;
3005 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3006 					X86_TRANSFER_TASK_SWITCH, NULL);
3007 	if (ret != X86EMUL_CONTINUE)
3008 		return ret;
3009 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3010 					X86_TRANSFER_TASK_SWITCH, NULL);
3011 	if (ret != X86EMUL_CONTINUE)
3012 		return ret;
3013 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3014 					X86_TRANSFER_TASK_SWITCH, NULL);
3015 	if (ret != X86EMUL_CONTINUE)
3016 		return ret;
3017 
3018 	return X86EMUL_CONTINUE;
3019 }
3020 
3021 static int task_switch_16(struct x86_emulate_ctxt *ctxt,
3022 			  u16 tss_selector, u16 old_tss_sel,
3023 			  ulong old_tss_base, struct desc_struct *new_desc)
3024 {
3025 	const struct x86_emulate_ops *ops = ctxt->ops;
3026 	struct tss_segment_16 tss_seg;
3027 	int ret;
3028 	u32 new_tss_base = get_desc_base(new_desc);
3029 
3030 	ret = ops->read_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3031 			    &ctxt->exception);
3032 	if (ret != X86EMUL_CONTINUE)
3033 		return ret;
3034 
3035 	save_state_to_tss16(ctxt, &tss_seg);
3036 
3037 	ret = ops->write_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3038 			     &ctxt->exception);
3039 	if (ret != X86EMUL_CONTINUE)
3040 		return ret;
3041 
3042 	ret = ops->read_std(ctxt, new_tss_base, &tss_seg, sizeof tss_seg,
3043 			    &ctxt->exception);
3044 	if (ret != X86EMUL_CONTINUE)
3045 		return ret;
3046 
3047 	if (old_tss_sel != 0xffff) {
3048 		tss_seg.prev_task_link = old_tss_sel;
3049 
3050 		ret = ops->write_std(ctxt, new_tss_base,
3051 				     &tss_seg.prev_task_link,
3052 				     sizeof tss_seg.prev_task_link,
3053 				     &ctxt->exception);
3054 		if (ret != X86EMUL_CONTINUE)
3055 			return ret;
3056 	}
3057 
3058 	return load_state_from_tss16(ctxt, &tss_seg);
3059 }
3060 
3061 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3062 				struct tss_segment_32 *tss)
3063 {
3064 	/* CR3 and ldt selector are not saved intentionally */
3065 	tss->eip = ctxt->_eip;
3066 	tss->eflags = ctxt->eflags;
3067 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3068 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3069 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3070 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3071 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3072 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3073 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3074 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3075 
3076 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3077 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3078 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3079 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3080 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3081 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3082 }
3083 
3084 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3085 				 struct tss_segment_32 *tss)
3086 {
3087 	int ret;
3088 	u8 cpl;
3089 
3090 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3091 		return emulate_gp(ctxt, 0);
3092 	ctxt->_eip = tss->eip;
3093 	ctxt->eflags = tss->eflags | 2;
3094 
3095 	/* General purpose registers */
3096 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3097 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3098 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3099 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3100 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3101 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3102 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3103 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3104 
3105 	/*
3106 	 * SDM says that segment selectors are loaded before segment
3107 	 * descriptors.  This is important because CPL checks will
3108 	 * use CS.RPL.
3109 	 */
3110 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3111 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3112 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3113 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3114 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3115 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3116 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3117 
3118 	/*
3119 	 * If we're switching between Protected Mode and VM86, we need to make
3120 	 * sure to update the mode before loading the segment descriptors so
3121 	 * that the selectors are interpreted correctly.
3122 	 */
3123 	if (ctxt->eflags & X86_EFLAGS_VM) {
3124 		ctxt->mode = X86EMUL_MODE_VM86;
3125 		cpl = 3;
3126 	} else {
3127 		ctxt->mode = X86EMUL_MODE_PROT32;
3128 		cpl = tss->cs & 3;
3129 	}
3130 
3131 	/*
3132 	 * Now load segment descriptors. If fault happenes at this stage
3133 	 * it is handled in a context of new task
3134 	 */
3135 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3136 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3137 	if (ret != X86EMUL_CONTINUE)
3138 		return ret;
3139 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3140 					X86_TRANSFER_TASK_SWITCH, NULL);
3141 	if (ret != X86EMUL_CONTINUE)
3142 		return ret;
3143 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3144 					X86_TRANSFER_TASK_SWITCH, NULL);
3145 	if (ret != X86EMUL_CONTINUE)
3146 		return ret;
3147 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3148 					X86_TRANSFER_TASK_SWITCH, NULL);
3149 	if (ret != X86EMUL_CONTINUE)
3150 		return ret;
3151 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3152 					X86_TRANSFER_TASK_SWITCH, NULL);
3153 	if (ret != X86EMUL_CONTINUE)
3154 		return ret;
3155 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3156 					X86_TRANSFER_TASK_SWITCH, NULL);
3157 	if (ret != X86EMUL_CONTINUE)
3158 		return ret;
3159 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3160 					X86_TRANSFER_TASK_SWITCH, NULL);
3161 
3162 	return ret;
3163 }
3164 
3165 static int task_switch_32(struct x86_emulate_ctxt *ctxt,
3166 			  u16 tss_selector, u16 old_tss_sel,
3167 			  ulong old_tss_base, struct desc_struct *new_desc)
3168 {
3169 	const struct x86_emulate_ops *ops = ctxt->ops;
3170 	struct tss_segment_32 tss_seg;
3171 	int ret;
3172 	u32 new_tss_base = get_desc_base(new_desc);
3173 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3174 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3175 
3176 	ret = ops->read_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3177 			    &ctxt->exception);
3178 	if (ret != X86EMUL_CONTINUE)
3179 		return ret;
3180 
3181 	save_state_to_tss32(ctxt, &tss_seg);
3182 
3183 	/* Only GP registers and segment selectors are saved */
3184 	ret = ops->write_std(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3185 			     ldt_sel_offset - eip_offset, &ctxt->exception);
3186 	if (ret != X86EMUL_CONTINUE)
3187 		return ret;
3188 
3189 	ret = ops->read_std(ctxt, new_tss_base, &tss_seg, sizeof tss_seg,
3190 			    &ctxt->exception);
3191 	if (ret != X86EMUL_CONTINUE)
3192 		return ret;
3193 
3194 	if (old_tss_sel != 0xffff) {
3195 		tss_seg.prev_task_link = old_tss_sel;
3196 
3197 		ret = ops->write_std(ctxt, new_tss_base,
3198 				     &tss_seg.prev_task_link,
3199 				     sizeof tss_seg.prev_task_link,
3200 				     &ctxt->exception);
3201 		if (ret != X86EMUL_CONTINUE)
3202 			return ret;
3203 	}
3204 
3205 	return load_state_from_tss32(ctxt, &tss_seg);
3206 }
3207 
3208 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3209 				   u16 tss_selector, int idt_index, int reason,
3210 				   bool has_error_code, u32 error_code)
3211 {
3212 	const struct x86_emulate_ops *ops = ctxt->ops;
3213 	struct desc_struct curr_tss_desc, next_tss_desc;
3214 	int ret;
3215 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3216 	ulong old_tss_base =
3217 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3218 	u32 desc_limit;
3219 	ulong desc_addr, dr7;
3220 
3221 	/* FIXME: old_tss_base == ~0 ? */
3222 
3223 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3224 	if (ret != X86EMUL_CONTINUE)
3225 		return ret;
3226 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3227 	if (ret != X86EMUL_CONTINUE)
3228 		return ret;
3229 
3230 	/* FIXME: check that next_tss_desc is tss */
3231 
3232 	/*
3233 	 * Check privileges. The three cases are task switch caused by...
3234 	 *
3235 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3236 	 * 2. Exception/IRQ/iret: No check is performed
3237 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3238 	 *    hardware checks it before exiting.
3239 	 */
3240 	if (reason == TASK_SWITCH_GATE) {
3241 		if (idt_index != -1) {
3242 			/* Software interrupts */
3243 			struct desc_struct task_gate_desc;
3244 			int dpl;
3245 
3246 			ret = read_interrupt_descriptor(ctxt, idt_index,
3247 							&task_gate_desc);
3248 			if (ret != X86EMUL_CONTINUE)
3249 				return ret;
3250 
3251 			dpl = task_gate_desc.dpl;
3252 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3253 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3254 		}
3255 	}
3256 
3257 	desc_limit = desc_limit_scaled(&next_tss_desc);
3258 	if (!next_tss_desc.p ||
3259 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3260 	     desc_limit < 0x2b)) {
3261 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3262 	}
3263 
3264 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3265 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3266 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3267 	}
3268 
3269 	if (reason == TASK_SWITCH_IRET)
3270 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3271 
3272 	/* set back link to prev task only if NT bit is set in eflags
3273 	   note that old_tss_sel is not used after this point */
3274 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3275 		old_tss_sel = 0xffff;
3276 
3277 	if (next_tss_desc.type & 8)
3278 		ret = task_switch_32(ctxt, tss_selector, old_tss_sel,
3279 				     old_tss_base, &next_tss_desc);
3280 	else
3281 		ret = task_switch_16(ctxt, tss_selector, old_tss_sel,
3282 				     old_tss_base, &next_tss_desc);
3283 	if (ret != X86EMUL_CONTINUE)
3284 		return ret;
3285 
3286 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3287 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3288 
3289 	if (reason != TASK_SWITCH_IRET) {
3290 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3291 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3292 	}
3293 
3294 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3295 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3296 
3297 	if (has_error_code) {
3298 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3299 		ctxt->lock_prefix = 0;
3300 		ctxt->src.val = (unsigned long) error_code;
3301 		ret = em_push(ctxt);
3302 	}
3303 
3304 	ops->get_dr(ctxt, 7, &dr7);
3305 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3306 
3307 	return ret;
3308 }
3309 
3310 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3311 			 u16 tss_selector, int idt_index, int reason,
3312 			 bool has_error_code, u32 error_code)
3313 {
3314 	int rc;
3315 
3316 	invalidate_registers(ctxt);
3317 	ctxt->_eip = ctxt->eip;
3318 	ctxt->dst.type = OP_NONE;
3319 
3320 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3321 				     has_error_code, error_code);
3322 
3323 	if (rc == X86EMUL_CONTINUE) {
3324 		ctxt->eip = ctxt->_eip;
3325 		writeback_registers(ctxt);
3326 	}
3327 
3328 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3329 }
3330 
3331 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3332 		struct operand *op)
3333 {
3334 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3335 
3336 	register_address_increment(ctxt, reg, df * op->bytes);
3337 	op->addr.mem.ea = register_address(ctxt, reg);
3338 }
3339 
3340 static int em_das(struct x86_emulate_ctxt *ctxt)
3341 {
3342 	u8 al, old_al;
3343 	bool af, cf, old_cf;
3344 
3345 	cf = ctxt->eflags & X86_EFLAGS_CF;
3346 	al = ctxt->dst.val;
3347 
3348 	old_al = al;
3349 	old_cf = cf;
3350 	cf = false;
3351 	af = ctxt->eflags & X86_EFLAGS_AF;
3352 	if ((al & 0x0f) > 9 || af) {
3353 		al -= 6;
3354 		cf = old_cf | (al >= 250);
3355 		af = true;
3356 	} else {
3357 		af = false;
3358 	}
3359 	if (old_al > 0x99 || old_cf) {
3360 		al -= 0x60;
3361 		cf = true;
3362 	}
3363 
3364 	ctxt->dst.val = al;
3365 	/* Set PF, ZF, SF */
3366 	ctxt->src.type = OP_IMM;
3367 	ctxt->src.val = 0;
3368 	ctxt->src.bytes = 1;
3369 	fastop(ctxt, em_or);
3370 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3371 	if (cf)
3372 		ctxt->eflags |= X86_EFLAGS_CF;
3373 	if (af)
3374 		ctxt->eflags |= X86_EFLAGS_AF;
3375 	return X86EMUL_CONTINUE;
3376 }
3377 
3378 static int em_aam(struct x86_emulate_ctxt *ctxt)
3379 {
3380 	u8 al, ah;
3381 
3382 	if (ctxt->src.val == 0)
3383 		return emulate_de(ctxt);
3384 
3385 	al = ctxt->dst.val & 0xff;
3386 	ah = al / ctxt->src.val;
3387 	al %= ctxt->src.val;
3388 
3389 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3390 
3391 	/* Set PF, ZF, SF */
3392 	ctxt->src.type = OP_IMM;
3393 	ctxt->src.val = 0;
3394 	ctxt->src.bytes = 1;
3395 	fastop(ctxt, em_or);
3396 
3397 	return X86EMUL_CONTINUE;
3398 }
3399 
3400 static int em_aad(struct x86_emulate_ctxt *ctxt)
3401 {
3402 	u8 al = ctxt->dst.val & 0xff;
3403 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3404 
3405 	al = (al + (ah * ctxt->src.val)) & 0xff;
3406 
3407 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3408 
3409 	/* Set PF, ZF, SF */
3410 	ctxt->src.type = OP_IMM;
3411 	ctxt->src.val = 0;
3412 	ctxt->src.bytes = 1;
3413 	fastop(ctxt, em_or);
3414 
3415 	return X86EMUL_CONTINUE;
3416 }
3417 
3418 static int em_call(struct x86_emulate_ctxt *ctxt)
3419 {
3420 	int rc;
3421 	long rel = ctxt->src.val;
3422 
3423 	ctxt->src.val = (unsigned long)ctxt->_eip;
3424 	rc = jmp_rel(ctxt, rel);
3425 	if (rc != X86EMUL_CONTINUE)
3426 		return rc;
3427 	return em_push(ctxt);
3428 }
3429 
3430 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3431 {
3432 	u16 sel, old_cs;
3433 	ulong old_eip;
3434 	int rc;
3435 	struct desc_struct old_desc, new_desc;
3436 	const struct x86_emulate_ops *ops = ctxt->ops;
3437 	int cpl = ctxt->ops->cpl(ctxt);
3438 	enum x86emul_mode prev_mode = ctxt->mode;
3439 
3440 	old_eip = ctxt->_eip;
3441 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3442 
3443 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3444 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3445 				       X86_TRANSFER_CALL_JMP, &new_desc);
3446 	if (rc != X86EMUL_CONTINUE)
3447 		return rc;
3448 
3449 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3450 	if (rc != X86EMUL_CONTINUE)
3451 		goto fail;
3452 
3453 	ctxt->src.val = old_cs;
3454 	rc = em_push(ctxt);
3455 	if (rc != X86EMUL_CONTINUE)
3456 		goto fail;
3457 
3458 	ctxt->src.val = old_eip;
3459 	rc = em_push(ctxt);
3460 	/* If we failed, we tainted the memory, but the very least we should
3461 	   restore cs */
3462 	if (rc != X86EMUL_CONTINUE) {
3463 		pr_warn_once("faulting far call emulation tainted memory\n");
3464 		goto fail;
3465 	}
3466 	return rc;
3467 fail:
3468 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3469 	ctxt->mode = prev_mode;
3470 	return rc;
3471 
3472 }
3473 
3474 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3475 {
3476 	int rc;
3477 	unsigned long eip;
3478 
3479 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3480 	if (rc != X86EMUL_CONTINUE)
3481 		return rc;
3482 	rc = assign_eip_near(ctxt, eip);
3483 	if (rc != X86EMUL_CONTINUE)
3484 		return rc;
3485 	rsp_increment(ctxt, ctxt->src.val);
3486 	return X86EMUL_CONTINUE;
3487 }
3488 
3489 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3490 {
3491 	/* Write back the register source. */
3492 	ctxt->src.val = ctxt->dst.val;
3493 	write_register_operand(&ctxt->src);
3494 
3495 	/* Write back the memory destination with implicit LOCK prefix. */
3496 	ctxt->dst.val = ctxt->src.orig_val;
3497 	ctxt->lock_prefix = 1;
3498 	return X86EMUL_CONTINUE;
3499 }
3500 
3501 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3502 {
3503 	ctxt->dst.val = ctxt->src2.val;
3504 	return fastop(ctxt, em_imul);
3505 }
3506 
3507 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3508 {
3509 	ctxt->dst.type = OP_REG;
3510 	ctxt->dst.bytes = ctxt->src.bytes;
3511 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3512 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3513 
3514 	return X86EMUL_CONTINUE;
3515 }
3516 
3517 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3518 {
3519 	u64 tsc = 0;
3520 
3521 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3522 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3523 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3524 	return X86EMUL_CONTINUE;
3525 }
3526 
3527 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3528 {
3529 	u64 pmc;
3530 
3531 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3532 		return emulate_gp(ctxt, 0);
3533 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3534 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3535 	return X86EMUL_CONTINUE;
3536 }
3537 
3538 static int em_mov(struct x86_emulate_ctxt *ctxt)
3539 {
3540 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3541 	return X86EMUL_CONTINUE;
3542 }
3543 
3544 #define FFL(x) bit(X86_FEATURE_##x)
3545 
3546 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3547 {
3548 	u32 ebx, ecx, edx, eax = 1;
3549 	u16 tmp;
3550 
3551 	/*
3552 	 * Check MOVBE is set in the guest-visible CPUID leaf.
3553 	 */
3554 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
3555 	if (!(ecx & FFL(MOVBE)))
3556 		return emulate_ud(ctxt);
3557 
3558 	switch (ctxt->op_bytes) {
3559 	case 2:
3560 		/*
3561 		 * From MOVBE definition: "...When the operand size is 16 bits,
3562 		 * the upper word of the destination register remains unchanged
3563 		 * ..."
3564 		 *
3565 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3566 		 * rules so we have to do the operation almost per hand.
3567 		 */
3568 		tmp = (u16)ctxt->src.val;
3569 		ctxt->dst.val &= ~0xffffUL;
3570 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3571 		break;
3572 	case 4:
3573 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3574 		break;
3575 	case 8:
3576 		ctxt->dst.val = swab64(ctxt->src.val);
3577 		break;
3578 	default:
3579 		BUG();
3580 	}
3581 	return X86EMUL_CONTINUE;
3582 }
3583 
3584 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3585 {
3586 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3587 		return emulate_gp(ctxt, 0);
3588 
3589 	/* Disable writeback. */
3590 	ctxt->dst.type = OP_NONE;
3591 	return X86EMUL_CONTINUE;
3592 }
3593 
3594 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3595 {
3596 	unsigned long val;
3597 
3598 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3599 		val = ctxt->src.val & ~0ULL;
3600 	else
3601 		val = ctxt->src.val & ~0U;
3602 
3603 	/* #UD condition is already handled. */
3604 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3605 		return emulate_gp(ctxt, 0);
3606 
3607 	/* Disable writeback. */
3608 	ctxt->dst.type = OP_NONE;
3609 	return X86EMUL_CONTINUE;
3610 }
3611 
3612 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3613 {
3614 	u64 msr_data;
3615 
3616 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3617 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3618 	if (ctxt->ops->set_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), msr_data))
3619 		return emulate_gp(ctxt, 0);
3620 
3621 	return X86EMUL_CONTINUE;
3622 }
3623 
3624 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3625 {
3626 	u64 msr_data;
3627 
3628 	if (ctxt->ops->get_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &msr_data))
3629 		return emulate_gp(ctxt, 0);
3630 
3631 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3632 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3633 	return X86EMUL_CONTINUE;
3634 }
3635 
3636 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3637 {
3638 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3639 		return emulate_ud(ctxt);
3640 
3641 	ctxt->dst.val = get_segment_selector(ctxt, ctxt->modrm_reg);
3642 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3643 		ctxt->dst.bytes = 2;
3644 	return X86EMUL_CONTINUE;
3645 }
3646 
3647 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3648 {
3649 	u16 sel = ctxt->src.val;
3650 
3651 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3652 		return emulate_ud(ctxt);
3653 
3654 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3655 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3656 
3657 	/* Disable writeback. */
3658 	ctxt->dst.type = OP_NONE;
3659 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3660 }
3661 
3662 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3663 {
3664 	u16 sel = ctxt->src.val;
3665 
3666 	/* Disable writeback. */
3667 	ctxt->dst.type = OP_NONE;
3668 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3669 }
3670 
3671 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3672 {
3673 	u16 sel = ctxt->src.val;
3674 
3675 	/* Disable writeback. */
3676 	ctxt->dst.type = OP_NONE;
3677 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3678 }
3679 
3680 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3681 {
3682 	int rc;
3683 	ulong linear;
3684 
3685 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3686 	if (rc == X86EMUL_CONTINUE)
3687 		ctxt->ops->invlpg(ctxt, linear);
3688 	/* Disable writeback. */
3689 	ctxt->dst.type = OP_NONE;
3690 	return X86EMUL_CONTINUE;
3691 }
3692 
3693 static int em_clts(struct x86_emulate_ctxt *ctxt)
3694 {
3695 	ulong cr0;
3696 
3697 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3698 	cr0 &= ~X86_CR0_TS;
3699 	ctxt->ops->set_cr(ctxt, 0, cr0);
3700 	return X86EMUL_CONTINUE;
3701 }
3702 
3703 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3704 {
3705 	int rc = ctxt->ops->fix_hypercall(ctxt);
3706 
3707 	if (rc != X86EMUL_CONTINUE)
3708 		return rc;
3709 
3710 	/* Let the processor re-execute the fixed hypercall */
3711 	ctxt->_eip = ctxt->eip;
3712 	/* Disable writeback. */
3713 	ctxt->dst.type = OP_NONE;
3714 	return X86EMUL_CONTINUE;
3715 }
3716 
3717 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3718 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3719 					      struct desc_ptr *ptr))
3720 {
3721 	struct desc_ptr desc_ptr;
3722 
3723 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3724 		ctxt->op_bytes = 8;
3725 	get(ctxt, &desc_ptr);
3726 	if (ctxt->op_bytes == 2) {
3727 		ctxt->op_bytes = 4;
3728 		desc_ptr.address &= 0x00ffffff;
3729 	}
3730 	/* Disable writeback. */
3731 	ctxt->dst.type = OP_NONE;
3732 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3733 				   &desc_ptr, 2 + ctxt->op_bytes);
3734 }
3735 
3736 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3737 {
3738 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3739 }
3740 
3741 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3742 {
3743 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3744 }
3745 
3746 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3747 {
3748 	struct desc_ptr desc_ptr;
3749 	int rc;
3750 
3751 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3752 		ctxt->op_bytes = 8;
3753 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3754 			     &desc_ptr.size, &desc_ptr.address,
3755 			     ctxt->op_bytes);
3756 	if (rc != X86EMUL_CONTINUE)
3757 		return rc;
3758 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3759 	    is_noncanonical_address(desc_ptr.address))
3760 		return emulate_gp(ctxt, 0);
3761 	if (lgdt)
3762 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3763 	else
3764 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3765 	/* Disable writeback. */
3766 	ctxt->dst.type = OP_NONE;
3767 	return X86EMUL_CONTINUE;
3768 }
3769 
3770 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3771 {
3772 	return em_lgdt_lidt(ctxt, true);
3773 }
3774 
3775 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3776 {
3777 	return em_lgdt_lidt(ctxt, false);
3778 }
3779 
3780 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3781 {
3782 	if (ctxt->dst.type == OP_MEM)
3783 		ctxt->dst.bytes = 2;
3784 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3785 	return X86EMUL_CONTINUE;
3786 }
3787 
3788 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3789 {
3790 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3791 			  | (ctxt->src.val & 0x0f));
3792 	ctxt->dst.type = OP_NONE;
3793 	return X86EMUL_CONTINUE;
3794 }
3795 
3796 static int em_loop(struct x86_emulate_ctxt *ctxt)
3797 {
3798 	int rc = X86EMUL_CONTINUE;
3799 
3800 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3801 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3802 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3803 		rc = jmp_rel(ctxt, ctxt->src.val);
3804 
3805 	return rc;
3806 }
3807 
3808 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3809 {
3810 	int rc = X86EMUL_CONTINUE;
3811 
3812 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3813 		rc = jmp_rel(ctxt, ctxt->src.val);
3814 
3815 	return rc;
3816 }
3817 
3818 static int em_in(struct x86_emulate_ctxt *ctxt)
3819 {
3820 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3821 			     &ctxt->dst.val))
3822 		return X86EMUL_IO_NEEDED;
3823 
3824 	return X86EMUL_CONTINUE;
3825 }
3826 
3827 static int em_out(struct x86_emulate_ctxt *ctxt)
3828 {
3829 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3830 				    &ctxt->src.val, 1);
3831 	/* Disable writeback. */
3832 	ctxt->dst.type = OP_NONE;
3833 	return X86EMUL_CONTINUE;
3834 }
3835 
3836 static int em_cli(struct x86_emulate_ctxt *ctxt)
3837 {
3838 	if (emulator_bad_iopl(ctxt))
3839 		return emulate_gp(ctxt, 0);
3840 
3841 	ctxt->eflags &= ~X86_EFLAGS_IF;
3842 	return X86EMUL_CONTINUE;
3843 }
3844 
3845 static int em_sti(struct x86_emulate_ctxt *ctxt)
3846 {
3847 	if (emulator_bad_iopl(ctxt))
3848 		return emulate_gp(ctxt, 0);
3849 
3850 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3851 	ctxt->eflags |= X86_EFLAGS_IF;
3852 	return X86EMUL_CONTINUE;
3853 }
3854 
3855 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3856 {
3857 	u32 eax, ebx, ecx, edx;
3858 	u64 msr = 0;
3859 
3860 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3861 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3862 	    ctxt->ops->cpl(ctxt)) {
3863 		return emulate_gp(ctxt, 0);
3864 	}
3865 
3866 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3867 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3868 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
3869 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3870 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3871 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3872 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3873 	return X86EMUL_CONTINUE;
3874 }
3875 
3876 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3877 {
3878 	u32 flags;
3879 
3880 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3881 		X86_EFLAGS_SF;
3882 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3883 
3884 	ctxt->eflags &= ~0xffUL;
3885 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3886 	return X86EMUL_CONTINUE;
3887 }
3888 
3889 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3890 {
3891 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3892 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3893 	return X86EMUL_CONTINUE;
3894 }
3895 
3896 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3897 {
3898 	switch (ctxt->op_bytes) {
3899 #ifdef CONFIG_X86_64
3900 	case 8:
3901 		asm("bswap %0" : "+r"(ctxt->dst.val));
3902 		break;
3903 #endif
3904 	default:
3905 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3906 		break;
3907 	}
3908 	return X86EMUL_CONTINUE;
3909 }
3910 
3911 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3912 {
3913 	/* emulating clflush regardless of cpuid */
3914 	return X86EMUL_CONTINUE;
3915 }
3916 
3917 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3918 {
3919 	ctxt->dst.val = (s32) ctxt->src.val;
3920 	return X86EMUL_CONTINUE;
3921 }
3922 
3923 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3924 {
3925 	u32 eax = 1, ebx, ecx = 0, edx;
3926 
3927 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
3928 	if (!(edx & FFL(FXSR)))
3929 		return emulate_ud(ctxt);
3930 
3931 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3932 		return emulate_nm(ctxt);
3933 
3934 	/*
3935 	 * Don't emulate a case that should never be hit, instead of working
3936 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3937 	 */
3938 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3939 		return X86EMUL_UNHANDLEABLE;
3940 
3941 	return X86EMUL_CONTINUE;
3942 }
3943 
3944 /*
3945  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
3946  * and restore MXCSR.
3947  */
3948 static size_t __fxstate_size(int nregs)
3949 {
3950 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
3951 }
3952 
3953 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
3954 {
3955 	bool cr4_osfxsr;
3956 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3957 		return __fxstate_size(16);
3958 
3959 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
3960 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
3961 }
3962 
3963 /*
3964  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
3965  *  1) 16 bit mode
3966  *  2) 32 bit mode
3967  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
3968  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
3969  *       save and restore
3970  *  3) 64-bit mode with REX.W prefix
3971  *     - like (2), but XMM 8-15 are being saved and restored
3972  *  4) 64-bit mode without REX.W prefix
3973  *     - like (3), but FIP and FDP are 64 bit
3974  *
3975  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
3976  * desired result.  (4) is not emulated.
3977  *
3978  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
3979  * and FPU DS) should match.
3980  */
3981 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
3982 {
3983 	struct fxregs_state fx_state;
3984 	int rc;
3985 
3986 	rc = check_fxsr(ctxt);
3987 	if (rc != X86EMUL_CONTINUE)
3988 		return rc;
3989 
3990 	ctxt->ops->get_fpu(ctxt);
3991 
3992 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
3993 
3994 	ctxt->ops->put_fpu(ctxt);
3995 
3996 	if (rc != X86EMUL_CONTINUE)
3997 		return rc;
3998 
3999 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4000 		                   fxstate_size(ctxt));
4001 }
4002 
4003 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4004 {
4005 	struct fxregs_state fx_state;
4006 	int rc;
4007 	size_t size;
4008 
4009 	rc = check_fxsr(ctxt);
4010 	if (rc != X86EMUL_CONTINUE)
4011 		return rc;
4012 
4013 	ctxt->ops->get_fpu(ctxt);
4014 
4015 	size = fxstate_size(ctxt);
4016 	if (size < __fxstate_size(16)) {
4017 		rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4018 		if (rc != X86EMUL_CONTINUE)
4019 			goto out;
4020 	}
4021 
4022 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4023 	if (rc != X86EMUL_CONTINUE)
4024 		goto out;
4025 
4026 	if (fx_state.mxcsr >> 16) {
4027 		rc = emulate_gp(ctxt, 0);
4028 		goto out;
4029 	}
4030 
4031 	if (rc == X86EMUL_CONTINUE)
4032 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4033 
4034 out:
4035 	ctxt->ops->put_fpu(ctxt);
4036 
4037 	return rc;
4038 }
4039 
4040 static bool valid_cr(int nr)
4041 {
4042 	switch (nr) {
4043 	case 0:
4044 	case 2 ... 4:
4045 	case 8:
4046 		return true;
4047 	default:
4048 		return false;
4049 	}
4050 }
4051 
4052 static int check_cr_read(struct x86_emulate_ctxt *ctxt)
4053 {
4054 	if (!valid_cr(ctxt->modrm_reg))
4055 		return emulate_ud(ctxt);
4056 
4057 	return X86EMUL_CONTINUE;
4058 }
4059 
4060 static int check_cr_write(struct x86_emulate_ctxt *ctxt)
4061 {
4062 	u64 new_val = ctxt->src.val64;
4063 	int cr = ctxt->modrm_reg;
4064 	u64 efer = 0;
4065 
4066 	static u64 cr_reserved_bits[] = {
4067 		0xffffffff00000000ULL,
4068 		0, 0, 0, /* CR3 checked later */
4069 		CR4_RESERVED_BITS,
4070 		0, 0, 0,
4071 		CR8_RESERVED_BITS,
4072 	};
4073 
4074 	if (!valid_cr(cr))
4075 		return emulate_ud(ctxt);
4076 
4077 	if (new_val & cr_reserved_bits[cr])
4078 		return emulate_gp(ctxt, 0);
4079 
4080 	switch (cr) {
4081 	case 0: {
4082 		u64 cr4;
4083 		if (((new_val & X86_CR0_PG) && !(new_val & X86_CR0_PE)) ||
4084 		    ((new_val & X86_CR0_NW) && !(new_val & X86_CR0_CD)))
4085 			return emulate_gp(ctxt, 0);
4086 
4087 		cr4 = ctxt->ops->get_cr(ctxt, 4);
4088 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4089 
4090 		if ((new_val & X86_CR0_PG) && (efer & EFER_LME) &&
4091 		    !(cr4 & X86_CR4_PAE))
4092 			return emulate_gp(ctxt, 0);
4093 
4094 		break;
4095 		}
4096 	case 3: {
4097 		u64 rsvd = 0;
4098 
4099 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4100 		if (efer & EFER_LMA)
4101 			rsvd = CR3_L_MODE_RESERVED_BITS & ~CR3_PCID_INVD;
4102 
4103 		if (new_val & rsvd)
4104 			return emulate_gp(ctxt, 0);
4105 
4106 		break;
4107 		}
4108 	case 4: {
4109 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4110 
4111 		if ((efer & EFER_LMA) && !(new_val & X86_CR4_PAE))
4112 			return emulate_gp(ctxt, 0);
4113 
4114 		break;
4115 		}
4116 	}
4117 
4118 	return X86EMUL_CONTINUE;
4119 }
4120 
4121 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4122 {
4123 	unsigned long dr7;
4124 
4125 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4126 
4127 	/* Check if DR7.Global_Enable is set */
4128 	return dr7 & (1 << 13);
4129 }
4130 
4131 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4132 {
4133 	int dr = ctxt->modrm_reg;
4134 	u64 cr4;
4135 
4136 	if (dr > 7)
4137 		return emulate_ud(ctxt);
4138 
4139 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4140 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4141 		return emulate_ud(ctxt);
4142 
4143 	if (check_dr7_gd(ctxt)) {
4144 		ulong dr6;
4145 
4146 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4147 		dr6 &= ~15;
4148 		dr6 |= DR6_BD | DR6_RTM;
4149 		ctxt->ops->set_dr(ctxt, 6, dr6);
4150 		return emulate_db(ctxt);
4151 	}
4152 
4153 	return X86EMUL_CONTINUE;
4154 }
4155 
4156 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4157 {
4158 	u64 new_val = ctxt->src.val64;
4159 	int dr = ctxt->modrm_reg;
4160 
4161 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4162 		return emulate_gp(ctxt, 0);
4163 
4164 	return check_dr_read(ctxt);
4165 }
4166 
4167 static int check_svme(struct x86_emulate_ctxt *ctxt)
4168 {
4169 	u64 efer = 0;
4170 
4171 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4172 
4173 	if (!(efer & EFER_SVME))
4174 		return emulate_ud(ctxt);
4175 
4176 	return X86EMUL_CONTINUE;
4177 }
4178 
4179 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4180 {
4181 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4182 
4183 	/* Valid physical address? */
4184 	if (rax & 0xffff000000000000ULL)
4185 		return emulate_gp(ctxt, 0);
4186 
4187 	return check_svme(ctxt);
4188 }
4189 
4190 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4191 {
4192 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4193 
4194 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4195 		return emulate_ud(ctxt);
4196 
4197 	return X86EMUL_CONTINUE;
4198 }
4199 
4200 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4201 {
4202 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4203 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4204 
4205 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4206 	    ctxt->ops->check_pmc(ctxt, rcx))
4207 		return emulate_gp(ctxt, 0);
4208 
4209 	return X86EMUL_CONTINUE;
4210 }
4211 
4212 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4213 {
4214 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4215 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4216 		return emulate_gp(ctxt, 0);
4217 
4218 	return X86EMUL_CONTINUE;
4219 }
4220 
4221 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4222 {
4223 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4224 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4225 		return emulate_gp(ctxt, 0);
4226 
4227 	return X86EMUL_CONTINUE;
4228 }
4229 
4230 #define D(_y) { .flags = (_y) }
4231 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4232 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4233 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4234 #define N    D(NotImpl)
4235 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4236 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4237 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4238 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4239 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4240 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4241 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4242 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4243 #define II(_f, _e, _i) \
4244 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4245 #define IIP(_f, _e, _i, _p) \
4246 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4247 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4248 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4249 
4250 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4251 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4252 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4253 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4254 #define I2bvIP(_f, _e, _i, _p) \
4255 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4256 
4257 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4258 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4259 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4260 
4261 static const struct opcode group7_rm0[] = {
4262 	N,
4263 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4264 	N, N, N, N, N, N,
4265 };
4266 
4267 static const struct opcode group7_rm1[] = {
4268 	DI(SrcNone | Priv, monitor),
4269 	DI(SrcNone | Priv, mwait),
4270 	N, N, N, N, N, N,
4271 };
4272 
4273 static const struct opcode group7_rm3[] = {
4274 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4275 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4276 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4277 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4278 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4279 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4280 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4281 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4282 };
4283 
4284 static const struct opcode group7_rm7[] = {
4285 	N,
4286 	DIP(SrcNone, rdtscp, check_rdtsc),
4287 	N, N, N, N, N, N,
4288 };
4289 
4290 static const struct opcode group1[] = {
4291 	F(Lock, em_add),
4292 	F(Lock | PageTable, em_or),
4293 	F(Lock, em_adc),
4294 	F(Lock, em_sbb),
4295 	F(Lock | PageTable, em_and),
4296 	F(Lock, em_sub),
4297 	F(Lock, em_xor),
4298 	F(NoWrite, em_cmp),
4299 };
4300 
4301 static const struct opcode group1A[] = {
4302 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4303 };
4304 
4305 static const struct opcode group2[] = {
4306 	F(DstMem | ModRM, em_rol),
4307 	F(DstMem | ModRM, em_ror),
4308 	F(DstMem | ModRM, em_rcl),
4309 	F(DstMem | ModRM, em_rcr),
4310 	F(DstMem | ModRM, em_shl),
4311 	F(DstMem | ModRM, em_shr),
4312 	F(DstMem | ModRM, em_shl),
4313 	F(DstMem | ModRM, em_sar),
4314 };
4315 
4316 static const struct opcode group3[] = {
4317 	F(DstMem | SrcImm | NoWrite, em_test),
4318 	F(DstMem | SrcImm | NoWrite, em_test),
4319 	F(DstMem | SrcNone | Lock, em_not),
4320 	F(DstMem | SrcNone | Lock, em_neg),
4321 	F(DstXacc | Src2Mem, em_mul_ex),
4322 	F(DstXacc | Src2Mem, em_imul_ex),
4323 	F(DstXacc | Src2Mem, em_div_ex),
4324 	F(DstXacc | Src2Mem, em_idiv_ex),
4325 };
4326 
4327 static const struct opcode group4[] = {
4328 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4329 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4330 	N, N, N, N, N, N,
4331 };
4332 
4333 static const struct opcode group5[] = {
4334 	F(DstMem | SrcNone | Lock,		em_inc),
4335 	F(DstMem | SrcNone | Lock,		em_dec),
4336 	I(SrcMem | NearBranch,			em_call_near_abs),
4337 	I(SrcMemFAddr | ImplicitOps,		em_call_far),
4338 	I(SrcMem | NearBranch,			em_jmp_abs),
4339 	I(SrcMemFAddr | ImplicitOps,		em_jmp_far),
4340 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4341 };
4342 
4343 static const struct opcode group6[] = {
4344 	DI(Prot | DstMem,	sldt),
4345 	DI(Prot | DstMem,	str),
4346 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4347 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4348 	N, N, N, N,
4349 };
4350 
4351 static const struct group_dual group7 = { {
4352 	II(Mov | DstMem,			em_sgdt, sgdt),
4353 	II(Mov | DstMem,			em_sidt, sidt),
4354 	II(SrcMem | Priv,			em_lgdt, lgdt),
4355 	II(SrcMem | Priv,			em_lidt, lidt),
4356 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4357 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4358 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4359 }, {
4360 	EXT(0, group7_rm0),
4361 	EXT(0, group7_rm1),
4362 	N, EXT(0, group7_rm3),
4363 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4364 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4365 	EXT(0, group7_rm7),
4366 } };
4367 
4368 static const struct opcode group8[] = {
4369 	N, N, N, N,
4370 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4371 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4372 	F(DstMem | SrcImmByte | Lock,			em_btr),
4373 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4374 };
4375 
4376 static const struct group_dual group9 = { {
4377 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4378 }, {
4379 	N, N, N, N, N, N, N, N,
4380 } };
4381 
4382 static const struct opcode group11[] = {
4383 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4384 	X7(D(Undefined)),
4385 };
4386 
4387 static const struct gprefix pfx_0f_ae_7 = {
4388 	I(SrcMem | ByteOp, em_clflush), N, N, N,
4389 };
4390 
4391 static const struct group_dual group15 = { {
4392 	I(ModRM | Aligned16, em_fxsave),
4393 	I(ModRM | Aligned16, em_fxrstor),
4394 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4395 }, {
4396 	N, N, N, N, N, N, N, N,
4397 } };
4398 
4399 static const struct gprefix pfx_0f_6f_0f_7f = {
4400 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4401 };
4402 
4403 static const struct instr_dual instr_dual_0f_2b = {
4404 	I(0, em_mov), N
4405 };
4406 
4407 static const struct gprefix pfx_0f_2b = {
4408 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4409 };
4410 
4411 static const struct gprefix pfx_0f_28_0f_29 = {
4412 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4413 };
4414 
4415 static const struct gprefix pfx_0f_e7 = {
4416 	N, I(Sse, em_mov), N, N,
4417 };
4418 
4419 static const struct escape escape_d9 = { {
4420 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4421 }, {
4422 	/* 0xC0 - 0xC7 */
4423 	N, N, N, N, N, N, N, N,
4424 	/* 0xC8 - 0xCF */
4425 	N, N, N, N, N, N, N, N,
4426 	/* 0xD0 - 0xC7 */
4427 	N, N, N, N, N, N, N, N,
4428 	/* 0xD8 - 0xDF */
4429 	N, N, N, N, N, N, N, N,
4430 	/* 0xE0 - 0xE7 */
4431 	N, N, N, N, N, N, N, N,
4432 	/* 0xE8 - 0xEF */
4433 	N, N, N, N, N, N, N, N,
4434 	/* 0xF0 - 0xF7 */
4435 	N, N, N, N, N, N, N, N,
4436 	/* 0xF8 - 0xFF */
4437 	N, N, N, N, N, N, N, N,
4438 } };
4439 
4440 static const struct escape escape_db = { {
4441 	N, N, N, N, N, N, N, N,
4442 }, {
4443 	/* 0xC0 - 0xC7 */
4444 	N, N, N, N, N, N, N, N,
4445 	/* 0xC8 - 0xCF */
4446 	N, N, N, N, N, N, N, N,
4447 	/* 0xD0 - 0xC7 */
4448 	N, N, N, N, N, N, N, N,
4449 	/* 0xD8 - 0xDF */
4450 	N, N, N, N, N, N, N, N,
4451 	/* 0xE0 - 0xE7 */
4452 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4453 	/* 0xE8 - 0xEF */
4454 	N, N, N, N, N, N, N, N,
4455 	/* 0xF0 - 0xF7 */
4456 	N, N, N, N, N, N, N, N,
4457 	/* 0xF8 - 0xFF */
4458 	N, N, N, N, N, N, N, N,
4459 } };
4460 
4461 static const struct escape escape_dd = { {
4462 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4463 }, {
4464 	/* 0xC0 - 0xC7 */
4465 	N, N, N, N, N, N, N, N,
4466 	/* 0xC8 - 0xCF */
4467 	N, N, N, N, N, N, N, N,
4468 	/* 0xD0 - 0xC7 */
4469 	N, N, N, N, N, N, N, N,
4470 	/* 0xD8 - 0xDF */
4471 	N, N, N, N, N, N, N, N,
4472 	/* 0xE0 - 0xE7 */
4473 	N, N, N, N, N, N, N, N,
4474 	/* 0xE8 - 0xEF */
4475 	N, N, N, N, N, N, N, N,
4476 	/* 0xF0 - 0xF7 */
4477 	N, N, N, N, N, N, N, N,
4478 	/* 0xF8 - 0xFF */
4479 	N, N, N, N, N, N, N, N,
4480 } };
4481 
4482 static const struct instr_dual instr_dual_0f_c3 = {
4483 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4484 };
4485 
4486 static const struct mode_dual mode_dual_63 = {
4487 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4488 };
4489 
4490 static const struct opcode opcode_table[256] = {
4491 	/* 0x00 - 0x07 */
4492 	F6ALU(Lock, em_add),
4493 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4494 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4495 	/* 0x08 - 0x0F */
4496 	F6ALU(Lock | PageTable, em_or),
4497 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4498 	N,
4499 	/* 0x10 - 0x17 */
4500 	F6ALU(Lock, em_adc),
4501 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4502 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4503 	/* 0x18 - 0x1F */
4504 	F6ALU(Lock, em_sbb),
4505 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4506 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4507 	/* 0x20 - 0x27 */
4508 	F6ALU(Lock | PageTable, em_and), N, N,
4509 	/* 0x28 - 0x2F */
4510 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4511 	/* 0x30 - 0x37 */
4512 	F6ALU(Lock, em_xor), N, N,
4513 	/* 0x38 - 0x3F */
4514 	F6ALU(NoWrite, em_cmp), N, N,
4515 	/* 0x40 - 0x4F */
4516 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4517 	/* 0x50 - 0x57 */
4518 	X8(I(SrcReg | Stack, em_push)),
4519 	/* 0x58 - 0x5F */
4520 	X8(I(DstReg | Stack, em_pop)),
4521 	/* 0x60 - 0x67 */
4522 	I(ImplicitOps | Stack | No64, em_pusha),
4523 	I(ImplicitOps | Stack | No64, em_popa),
4524 	N, MD(ModRM, &mode_dual_63),
4525 	N, N, N, N,
4526 	/* 0x68 - 0x6F */
4527 	I(SrcImm | Mov | Stack, em_push),
4528 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4529 	I(SrcImmByte | Mov | Stack, em_push),
4530 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4531 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4532 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4533 	/* 0x70 - 0x7F */
4534 	X16(D(SrcImmByte | NearBranch)),
4535 	/* 0x80 - 0x87 */
4536 	G(ByteOp | DstMem | SrcImm, group1),
4537 	G(DstMem | SrcImm, group1),
4538 	G(ByteOp | DstMem | SrcImm | No64, group1),
4539 	G(DstMem | SrcImmByte, group1),
4540 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4541 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4542 	/* 0x88 - 0x8F */
4543 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4544 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4545 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4546 	D(ModRM | SrcMem | NoAccess | DstReg),
4547 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4548 	G(0, group1A),
4549 	/* 0x90 - 0x97 */
4550 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4551 	/* 0x98 - 0x9F */
4552 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4553 	I(SrcImmFAddr | No64, em_call_far), N,
4554 	II(ImplicitOps | Stack, em_pushf, pushf),
4555 	II(ImplicitOps | Stack, em_popf, popf),
4556 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4557 	/* 0xA0 - 0xA7 */
4558 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4559 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4560 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4561 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4562 	/* 0xA8 - 0xAF */
4563 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4564 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4565 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4566 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4567 	/* 0xB0 - 0xB7 */
4568 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4569 	/* 0xB8 - 0xBF */
4570 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4571 	/* 0xC0 - 0xC7 */
4572 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4573 	I(ImplicitOps | NearBranch | SrcImmU16, em_ret_near_imm),
4574 	I(ImplicitOps | NearBranch, em_ret),
4575 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4576 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4577 	G(ByteOp, group11), G(0, group11),
4578 	/* 0xC8 - 0xCF */
4579 	I(Stack | SrcImmU16 | Src2ImmByte, em_enter), I(Stack, em_leave),
4580 	I(ImplicitOps | SrcImmU16, em_ret_far_imm),
4581 	I(ImplicitOps, em_ret_far),
4582 	D(ImplicitOps), DI(SrcImmByte, intn),
4583 	D(ImplicitOps | No64), II(ImplicitOps, em_iret, iret),
4584 	/* 0xD0 - 0xD7 */
4585 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4586 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4587 	I(DstAcc | SrcImmUByte | No64, em_aam),
4588 	I(DstAcc | SrcImmUByte | No64, em_aad),
4589 	F(DstAcc | ByteOp | No64, em_salc),
4590 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4591 	/* 0xD8 - 0xDF */
4592 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4593 	/* 0xE0 - 0xE7 */
4594 	X3(I(SrcImmByte | NearBranch, em_loop)),
4595 	I(SrcImmByte | NearBranch, em_jcxz),
4596 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4597 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4598 	/* 0xE8 - 0xEF */
4599 	I(SrcImm | NearBranch, em_call), D(SrcImm | ImplicitOps | NearBranch),
4600 	I(SrcImmFAddr | No64, em_jmp_far),
4601 	D(SrcImmByte | ImplicitOps | NearBranch),
4602 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4603 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4604 	/* 0xF0 - 0xF7 */
4605 	N, DI(ImplicitOps, icebp), N, N,
4606 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4607 	G(ByteOp, group3), G(0, group3),
4608 	/* 0xF8 - 0xFF */
4609 	D(ImplicitOps), D(ImplicitOps),
4610 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4611 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4612 };
4613 
4614 static const struct opcode twobyte_table[256] = {
4615 	/* 0x00 - 0x0F */
4616 	G(0, group6), GD(0, &group7), N, N,
4617 	N, I(ImplicitOps | EmulateOnUD, em_syscall),
4618 	II(ImplicitOps | Priv, em_clts, clts), N,
4619 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4620 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4621 	/* 0x10 - 0x1F */
4622 	N, N, N, N, N, N, N, N,
4623 	D(ImplicitOps | ModRM | SrcMem | NoAccess),
4624 	N, N, N, N, N, N, D(ImplicitOps | ModRM | SrcMem | NoAccess),
4625 	/* 0x20 - 0x2F */
4626 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_read),
4627 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4628 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4629 						check_cr_write),
4630 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4631 						check_dr_write),
4632 	N, N, N, N,
4633 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4634 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4635 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4636 	N, N, N, N,
4637 	/* 0x30 - 0x3F */
4638 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4639 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4640 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4641 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4642 	I(ImplicitOps | EmulateOnUD, em_sysenter),
4643 	I(ImplicitOps | Priv | EmulateOnUD, em_sysexit),
4644 	N, N,
4645 	N, N, N, N, N, N, N, N,
4646 	/* 0x40 - 0x4F */
4647 	X16(D(DstReg | SrcMem | ModRM)),
4648 	/* 0x50 - 0x5F */
4649 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4650 	/* 0x60 - 0x6F */
4651 	N, N, N, N,
4652 	N, N, N, N,
4653 	N, N, N, N,
4654 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4655 	/* 0x70 - 0x7F */
4656 	N, N, N, N,
4657 	N, N, N, N,
4658 	N, N, N, N,
4659 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4660 	/* 0x80 - 0x8F */
4661 	X16(D(SrcImm | NearBranch)),
4662 	/* 0x90 - 0x9F */
4663 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4664 	/* 0xA0 - 0xA7 */
4665 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4666 	II(ImplicitOps, em_cpuid, cpuid),
4667 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4668 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4669 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4670 	/* 0xA8 - 0xAF */
4671 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4672 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4673 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4674 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4675 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4676 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4677 	/* 0xB0 - 0xB7 */
4678 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4679 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4680 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4681 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4682 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4683 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4684 	/* 0xB8 - 0xBF */
4685 	N, N,
4686 	G(BitOp, group8),
4687 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4688 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4689 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4690 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4691 	/* 0xC0 - 0xC7 */
4692 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4693 	N, ID(0, &instr_dual_0f_c3),
4694 	N, N, N, GD(0, &group9),
4695 	/* 0xC8 - 0xCF */
4696 	X8(I(DstReg, em_bswap)),
4697 	/* 0xD0 - 0xDF */
4698 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4699 	/* 0xE0 - 0xEF */
4700 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4701 	N, N, N, N, N, N, N, N,
4702 	/* 0xF0 - 0xFF */
4703 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4704 };
4705 
4706 static const struct instr_dual instr_dual_0f_38_f0 = {
4707 	I(DstReg | SrcMem | Mov, em_movbe), N
4708 };
4709 
4710 static const struct instr_dual instr_dual_0f_38_f1 = {
4711 	I(DstMem | SrcReg | Mov, em_movbe), N
4712 };
4713 
4714 static const struct gprefix three_byte_0f_38_f0 = {
4715 	ID(0, &instr_dual_0f_38_f0), N, N, N
4716 };
4717 
4718 static const struct gprefix three_byte_0f_38_f1 = {
4719 	ID(0, &instr_dual_0f_38_f1), N, N, N
4720 };
4721 
4722 /*
4723  * Insns below are selected by the prefix which indexed by the third opcode
4724  * byte.
4725  */
4726 static const struct opcode opcode_map_0f_38[256] = {
4727 	/* 0x00 - 0x7f */
4728 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4729 	/* 0x80 - 0xef */
4730 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4731 	/* 0xf0 - 0xf1 */
4732 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4733 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4734 	/* 0xf2 - 0xff */
4735 	N, N, X4(N), X8(N)
4736 };
4737 
4738 #undef D
4739 #undef N
4740 #undef G
4741 #undef GD
4742 #undef I
4743 #undef GP
4744 #undef EXT
4745 #undef MD
4746 #undef ID
4747 
4748 #undef D2bv
4749 #undef D2bvIP
4750 #undef I2bv
4751 #undef I2bvIP
4752 #undef I6ALU
4753 
4754 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4755 {
4756 	unsigned size;
4757 
4758 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4759 	if (size == 8)
4760 		size = 4;
4761 	return size;
4762 }
4763 
4764 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4765 		      unsigned size, bool sign_extension)
4766 {
4767 	int rc = X86EMUL_CONTINUE;
4768 
4769 	op->type = OP_IMM;
4770 	op->bytes = size;
4771 	op->addr.mem.ea = ctxt->_eip;
4772 	/* NB. Immediates are sign-extended as necessary. */
4773 	switch (op->bytes) {
4774 	case 1:
4775 		op->val = insn_fetch(s8, ctxt);
4776 		break;
4777 	case 2:
4778 		op->val = insn_fetch(s16, ctxt);
4779 		break;
4780 	case 4:
4781 		op->val = insn_fetch(s32, ctxt);
4782 		break;
4783 	case 8:
4784 		op->val = insn_fetch(s64, ctxt);
4785 		break;
4786 	}
4787 	if (!sign_extension) {
4788 		switch (op->bytes) {
4789 		case 1:
4790 			op->val &= 0xff;
4791 			break;
4792 		case 2:
4793 			op->val &= 0xffff;
4794 			break;
4795 		case 4:
4796 			op->val &= 0xffffffff;
4797 			break;
4798 		}
4799 	}
4800 done:
4801 	return rc;
4802 }
4803 
4804 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4805 			  unsigned d)
4806 {
4807 	int rc = X86EMUL_CONTINUE;
4808 
4809 	switch (d) {
4810 	case OpReg:
4811 		decode_register_operand(ctxt, op);
4812 		break;
4813 	case OpImmUByte:
4814 		rc = decode_imm(ctxt, op, 1, false);
4815 		break;
4816 	case OpMem:
4817 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4818 	mem_common:
4819 		*op = ctxt->memop;
4820 		ctxt->memopp = op;
4821 		if (ctxt->d & BitOp)
4822 			fetch_bit_operand(ctxt);
4823 		op->orig_val = op->val;
4824 		break;
4825 	case OpMem64:
4826 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4827 		goto mem_common;
4828 	case OpAcc:
4829 		op->type = OP_REG;
4830 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4831 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4832 		fetch_register_operand(op);
4833 		op->orig_val = op->val;
4834 		break;
4835 	case OpAccLo:
4836 		op->type = OP_REG;
4837 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4838 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4839 		fetch_register_operand(op);
4840 		op->orig_val = op->val;
4841 		break;
4842 	case OpAccHi:
4843 		if (ctxt->d & ByteOp) {
4844 			op->type = OP_NONE;
4845 			break;
4846 		}
4847 		op->type = OP_REG;
4848 		op->bytes = ctxt->op_bytes;
4849 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4850 		fetch_register_operand(op);
4851 		op->orig_val = op->val;
4852 		break;
4853 	case OpDI:
4854 		op->type = OP_MEM;
4855 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4856 		op->addr.mem.ea =
4857 			register_address(ctxt, VCPU_REGS_RDI);
4858 		op->addr.mem.seg = VCPU_SREG_ES;
4859 		op->val = 0;
4860 		op->count = 1;
4861 		break;
4862 	case OpDX:
4863 		op->type = OP_REG;
4864 		op->bytes = 2;
4865 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4866 		fetch_register_operand(op);
4867 		break;
4868 	case OpCL:
4869 		op->type = OP_IMM;
4870 		op->bytes = 1;
4871 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4872 		break;
4873 	case OpImmByte:
4874 		rc = decode_imm(ctxt, op, 1, true);
4875 		break;
4876 	case OpOne:
4877 		op->type = OP_IMM;
4878 		op->bytes = 1;
4879 		op->val = 1;
4880 		break;
4881 	case OpImm:
4882 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4883 		break;
4884 	case OpImm64:
4885 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4886 		break;
4887 	case OpMem8:
4888 		ctxt->memop.bytes = 1;
4889 		if (ctxt->memop.type == OP_REG) {
4890 			ctxt->memop.addr.reg = decode_register(ctxt,
4891 					ctxt->modrm_rm, true);
4892 			fetch_register_operand(&ctxt->memop);
4893 		}
4894 		goto mem_common;
4895 	case OpMem16:
4896 		ctxt->memop.bytes = 2;
4897 		goto mem_common;
4898 	case OpMem32:
4899 		ctxt->memop.bytes = 4;
4900 		goto mem_common;
4901 	case OpImmU16:
4902 		rc = decode_imm(ctxt, op, 2, false);
4903 		break;
4904 	case OpImmU:
4905 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4906 		break;
4907 	case OpSI:
4908 		op->type = OP_MEM;
4909 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4910 		op->addr.mem.ea =
4911 			register_address(ctxt, VCPU_REGS_RSI);
4912 		op->addr.mem.seg = ctxt->seg_override;
4913 		op->val = 0;
4914 		op->count = 1;
4915 		break;
4916 	case OpXLat:
4917 		op->type = OP_MEM;
4918 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4919 		op->addr.mem.ea =
4920 			address_mask(ctxt,
4921 				reg_read(ctxt, VCPU_REGS_RBX) +
4922 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
4923 		op->addr.mem.seg = ctxt->seg_override;
4924 		op->val = 0;
4925 		break;
4926 	case OpImmFAddr:
4927 		op->type = OP_IMM;
4928 		op->addr.mem.ea = ctxt->_eip;
4929 		op->bytes = ctxt->op_bytes + 2;
4930 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
4931 		break;
4932 	case OpMemFAddr:
4933 		ctxt->memop.bytes = ctxt->op_bytes + 2;
4934 		goto mem_common;
4935 	case OpES:
4936 		op->type = OP_IMM;
4937 		op->val = VCPU_SREG_ES;
4938 		break;
4939 	case OpCS:
4940 		op->type = OP_IMM;
4941 		op->val = VCPU_SREG_CS;
4942 		break;
4943 	case OpSS:
4944 		op->type = OP_IMM;
4945 		op->val = VCPU_SREG_SS;
4946 		break;
4947 	case OpDS:
4948 		op->type = OP_IMM;
4949 		op->val = VCPU_SREG_DS;
4950 		break;
4951 	case OpFS:
4952 		op->type = OP_IMM;
4953 		op->val = VCPU_SREG_FS;
4954 		break;
4955 	case OpGS:
4956 		op->type = OP_IMM;
4957 		op->val = VCPU_SREG_GS;
4958 		break;
4959 	case OpImplicit:
4960 		/* Special instructions do their own operand decoding. */
4961 	default:
4962 		op->type = OP_NONE; /* Disable writeback. */
4963 		break;
4964 	}
4965 
4966 done:
4967 	return rc;
4968 }
4969 
4970 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len)
4971 {
4972 	int rc = X86EMUL_CONTINUE;
4973 	int mode = ctxt->mode;
4974 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
4975 	bool op_prefix = false;
4976 	bool has_seg_override = false;
4977 	struct opcode opcode;
4978 
4979 	ctxt->memop.type = OP_NONE;
4980 	ctxt->memopp = NULL;
4981 	ctxt->_eip = ctxt->eip;
4982 	ctxt->fetch.ptr = ctxt->fetch.data;
4983 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
4984 	ctxt->opcode_len = 1;
4985 	if (insn_len > 0)
4986 		memcpy(ctxt->fetch.data, insn, insn_len);
4987 	else {
4988 		rc = __do_insn_fetch_bytes(ctxt, 1);
4989 		if (rc != X86EMUL_CONTINUE)
4990 			return rc;
4991 	}
4992 
4993 	switch (mode) {
4994 	case X86EMUL_MODE_REAL:
4995 	case X86EMUL_MODE_VM86:
4996 	case X86EMUL_MODE_PROT16:
4997 		def_op_bytes = def_ad_bytes = 2;
4998 		break;
4999 	case X86EMUL_MODE_PROT32:
5000 		def_op_bytes = def_ad_bytes = 4;
5001 		break;
5002 #ifdef CONFIG_X86_64
5003 	case X86EMUL_MODE_PROT64:
5004 		def_op_bytes = 4;
5005 		def_ad_bytes = 8;
5006 		break;
5007 #endif
5008 	default:
5009 		return EMULATION_FAILED;
5010 	}
5011 
5012 	ctxt->op_bytes = def_op_bytes;
5013 	ctxt->ad_bytes = def_ad_bytes;
5014 
5015 	/* Legacy prefixes. */
5016 	for (;;) {
5017 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5018 		case 0x66:	/* operand-size override */
5019 			op_prefix = true;
5020 			/* switch between 2/4 bytes */
5021 			ctxt->op_bytes = def_op_bytes ^ 6;
5022 			break;
5023 		case 0x67:	/* address-size override */
5024 			if (mode == X86EMUL_MODE_PROT64)
5025 				/* switch between 4/8 bytes */
5026 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5027 			else
5028 				/* switch between 2/4 bytes */
5029 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5030 			break;
5031 		case 0x26:	/* ES override */
5032 		case 0x2e:	/* CS override */
5033 		case 0x36:	/* SS override */
5034 		case 0x3e:	/* DS override */
5035 			has_seg_override = true;
5036 			ctxt->seg_override = (ctxt->b >> 3) & 3;
5037 			break;
5038 		case 0x64:	/* FS override */
5039 		case 0x65:	/* GS override */
5040 			has_seg_override = true;
5041 			ctxt->seg_override = ctxt->b & 7;
5042 			break;
5043 		case 0x40 ... 0x4f: /* REX */
5044 			if (mode != X86EMUL_MODE_PROT64)
5045 				goto done_prefixes;
5046 			ctxt->rex_prefix = ctxt->b;
5047 			continue;
5048 		case 0xf0:	/* LOCK */
5049 			ctxt->lock_prefix = 1;
5050 			break;
5051 		case 0xf2:	/* REPNE/REPNZ */
5052 		case 0xf3:	/* REP/REPE/REPZ */
5053 			ctxt->rep_prefix = ctxt->b;
5054 			break;
5055 		default:
5056 			goto done_prefixes;
5057 		}
5058 
5059 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5060 
5061 		ctxt->rex_prefix = 0;
5062 	}
5063 
5064 done_prefixes:
5065 
5066 	/* REX prefix. */
5067 	if (ctxt->rex_prefix & 8)
5068 		ctxt->op_bytes = 8;	/* REX.W */
5069 
5070 	/* Opcode byte(s). */
5071 	opcode = opcode_table[ctxt->b];
5072 	/* Two-byte opcode? */
5073 	if (ctxt->b == 0x0f) {
5074 		ctxt->opcode_len = 2;
5075 		ctxt->b = insn_fetch(u8, ctxt);
5076 		opcode = twobyte_table[ctxt->b];
5077 
5078 		/* 0F_38 opcode map */
5079 		if (ctxt->b == 0x38) {
5080 			ctxt->opcode_len = 3;
5081 			ctxt->b = insn_fetch(u8, ctxt);
5082 			opcode = opcode_map_0f_38[ctxt->b];
5083 		}
5084 	}
5085 	ctxt->d = opcode.flags;
5086 
5087 	if (ctxt->d & ModRM)
5088 		ctxt->modrm = insn_fetch(u8, ctxt);
5089 
5090 	/* vex-prefix instructions are not implemented */
5091 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5092 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5093 		ctxt->d = NotImpl;
5094 	}
5095 
5096 	while (ctxt->d & GroupMask) {
5097 		switch (ctxt->d & GroupMask) {
5098 		case Group:
5099 			goffset = (ctxt->modrm >> 3) & 7;
5100 			opcode = opcode.u.group[goffset];
5101 			break;
5102 		case GroupDual:
5103 			goffset = (ctxt->modrm >> 3) & 7;
5104 			if ((ctxt->modrm >> 6) == 3)
5105 				opcode = opcode.u.gdual->mod3[goffset];
5106 			else
5107 				opcode = opcode.u.gdual->mod012[goffset];
5108 			break;
5109 		case RMExt:
5110 			goffset = ctxt->modrm & 7;
5111 			opcode = opcode.u.group[goffset];
5112 			break;
5113 		case Prefix:
5114 			if (ctxt->rep_prefix && op_prefix)
5115 				return EMULATION_FAILED;
5116 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5117 			switch (simd_prefix) {
5118 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5119 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5120 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5121 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5122 			}
5123 			break;
5124 		case Escape:
5125 			if (ctxt->modrm > 0xbf)
5126 				opcode = opcode.u.esc->high[ctxt->modrm - 0xc0];
5127 			else
5128 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5129 			break;
5130 		case InstrDual:
5131 			if ((ctxt->modrm >> 6) == 3)
5132 				opcode = opcode.u.idual->mod3;
5133 			else
5134 				opcode = opcode.u.idual->mod012;
5135 			break;
5136 		case ModeDual:
5137 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5138 				opcode = opcode.u.mdual->mode64;
5139 			else
5140 				opcode = opcode.u.mdual->mode32;
5141 			break;
5142 		default:
5143 			return EMULATION_FAILED;
5144 		}
5145 
5146 		ctxt->d &= ~(u64)GroupMask;
5147 		ctxt->d |= opcode.flags;
5148 	}
5149 
5150 	/* Unrecognised? */
5151 	if (ctxt->d == 0)
5152 		return EMULATION_FAILED;
5153 
5154 	ctxt->execute = opcode.u.execute;
5155 
5156 	if (unlikely(ctxt->ud) && likely(!(ctxt->d & EmulateOnUD)))
5157 		return EMULATION_FAILED;
5158 
5159 	if (unlikely(ctxt->d &
5160 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5161 	     No16))) {
5162 		/*
5163 		 * These are copied unconditionally here, and checked unconditionally
5164 		 * in x86_emulate_insn.
5165 		 */
5166 		ctxt->check_perm = opcode.check_perm;
5167 		ctxt->intercept = opcode.intercept;
5168 
5169 		if (ctxt->d & NotImpl)
5170 			return EMULATION_FAILED;
5171 
5172 		if (mode == X86EMUL_MODE_PROT64) {
5173 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5174 				ctxt->op_bytes = 8;
5175 			else if (ctxt->d & NearBranch)
5176 				ctxt->op_bytes = 8;
5177 		}
5178 
5179 		if (ctxt->d & Op3264) {
5180 			if (mode == X86EMUL_MODE_PROT64)
5181 				ctxt->op_bytes = 8;
5182 			else
5183 				ctxt->op_bytes = 4;
5184 		}
5185 
5186 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5187 			ctxt->op_bytes = 4;
5188 
5189 		if (ctxt->d & Sse)
5190 			ctxt->op_bytes = 16;
5191 		else if (ctxt->d & Mmx)
5192 			ctxt->op_bytes = 8;
5193 	}
5194 
5195 	/* ModRM and SIB bytes. */
5196 	if (ctxt->d & ModRM) {
5197 		rc = decode_modrm(ctxt, &ctxt->memop);
5198 		if (!has_seg_override) {
5199 			has_seg_override = true;
5200 			ctxt->seg_override = ctxt->modrm_seg;
5201 		}
5202 	} else if (ctxt->d & MemAbs)
5203 		rc = decode_abs(ctxt, &ctxt->memop);
5204 	if (rc != X86EMUL_CONTINUE)
5205 		goto done;
5206 
5207 	if (!has_seg_override)
5208 		ctxt->seg_override = VCPU_SREG_DS;
5209 
5210 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5211 
5212 	/*
5213 	 * Decode and fetch the source operand: register, memory
5214 	 * or immediate.
5215 	 */
5216 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5217 	if (rc != X86EMUL_CONTINUE)
5218 		goto done;
5219 
5220 	/*
5221 	 * Decode and fetch the second source operand: register, memory
5222 	 * or immediate.
5223 	 */
5224 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5225 	if (rc != X86EMUL_CONTINUE)
5226 		goto done;
5227 
5228 	/* Decode and fetch the destination operand: register or memory. */
5229 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5230 
5231 	if (ctxt->rip_relative && likely(ctxt->memopp))
5232 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5233 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5234 
5235 done:
5236 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5237 }
5238 
5239 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5240 {
5241 	return ctxt->d & PageTable;
5242 }
5243 
5244 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5245 {
5246 	/* The second termination condition only applies for REPE
5247 	 * and REPNE. Test if the repeat string operation prefix is
5248 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5249 	 * corresponding termination condition according to:
5250 	 * 	- if REPE/REPZ and ZF = 0 then done
5251 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5252 	 */
5253 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5254 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5255 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5256 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5257 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5258 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5259 		return true;
5260 
5261 	return false;
5262 }
5263 
5264 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5265 {
5266 	int rc;
5267 
5268 	ctxt->ops->get_fpu(ctxt);
5269 	rc = asm_safe("fwait");
5270 	ctxt->ops->put_fpu(ctxt);
5271 
5272 	if (unlikely(rc != X86EMUL_CONTINUE))
5273 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5274 
5275 	return X86EMUL_CONTINUE;
5276 }
5277 
5278 static void fetch_possible_mmx_operand(struct x86_emulate_ctxt *ctxt,
5279 				       struct operand *op)
5280 {
5281 	if (op->type == OP_MM)
5282 		read_mmx_reg(ctxt, &op->mm_val, op->addr.mm);
5283 }
5284 
5285 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *))
5286 {
5287 	register void *__sp asm(_ASM_SP);
5288 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5289 
5290 	if (!(ctxt->d & ByteOp))
5291 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5292 
5293 	asm("push %[flags]; popf; call *%[fastop]; pushf; pop %[flags]\n"
5294 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5295 	      [fastop]"+S"(fop), "+r"(__sp)
5296 	    : "c"(ctxt->src2.val));
5297 
5298 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5299 	if (!fop) /* exception is returned in fop variable */
5300 		return emulate_de(ctxt);
5301 	return X86EMUL_CONTINUE;
5302 }
5303 
5304 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5305 {
5306 	memset(&ctxt->rip_relative, 0,
5307 	       (void *)&ctxt->modrm - (void *)&ctxt->rip_relative);
5308 
5309 	ctxt->io_read.pos = 0;
5310 	ctxt->io_read.end = 0;
5311 	ctxt->mem_read.end = 0;
5312 }
5313 
5314 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5315 {
5316 	const struct x86_emulate_ops *ops = ctxt->ops;
5317 	int rc = X86EMUL_CONTINUE;
5318 	int saved_dst_type = ctxt->dst.type;
5319 	unsigned emul_flags;
5320 
5321 	ctxt->mem_read.pos = 0;
5322 
5323 	/* LOCK prefix is allowed only with some instructions */
5324 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5325 		rc = emulate_ud(ctxt);
5326 		goto done;
5327 	}
5328 
5329 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5330 		rc = emulate_ud(ctxt);
5331 		goto done;
5332 	}
5333 
5334 	emul_flags = ctxt->ops->get_hflags(ctxt);
5335 	if (unlikely(ctxt->d &
5336 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5337 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5338 				(ctxt->d & Undefined)) {
5339 			rc = emulate_ud(ctxt);
5340 			goto done;
5341 		}
5342 
5343 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5344 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5345 			rc = emulate_ud(ctxt);
5346 			goto done;
5347 		}
5348 
5349 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5350 			rc = emulate_nm(ctxt);
5351 			goto done;
5352 		}
5353 
5354 		if (ctxt->d & Mmx) {
5355 			rc = flush_pending_x87_faults(ctxt);
5356 			if (rc != X86EMUL_CONTINUE)
5357 				goto done;
5358 			/*
5359 			 * Now that we know the fpu is exception safe, we can fetch
5360 			 * operands from it.
5361 			 */
5362 			fetch_possible_mmx_operand(ctxt, &ctxt->src);
5363 			fetch_possible_mmx_operand(ctxt, &ctxt->src2);
5364 			if (!(ctxt->d & Mov))
5365 				fetch_possible_mmx_operand(ctxt, &ctxt->dst);
5366 		}
5367 
5368 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5369 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5370 						      X86_ICPT_PRE_EXCEPT);
5371 			if (rc != X86EMUL_CONTINUE)
5372 				goto done;
5373 		}
5374 
5375 		/* Instruction can only be executed in protected mode */
5376 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5377 			rc = emulate_ud(ctxt);
5378 			goto done;
5379 		}
5380 
5381 		/* Privileged instruction can be executed only in CPL=0 */
5382 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5383 			if (ctxt->d & PrivUD)
5384 				rc = emulate_ud(ctxt);
5385 			else
5386 				rc = emulate_gp(ctxt, 0);
5387 			goto done;
5388 		}
5389 
5390 		/* Do instruction specific permission checks */
5391 		if (ctxt->d & CheckPerm) {
5392 			rc = ctxt->check_perm(ctxt);
5393 			if (rc != X86EMUL_CONTINUE)
5394 				goto done;
5395 		}
5396 
5397 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5398 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5399 						      X86_ICPT_POST_EXCEPT);
5400 			if (rc != X86EMUL_CONTINUE)
5401 				goto done;
5402 		}
5403 
5404 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5405 			/* All REP prefixes have the same first termination condition */
5406 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5407 				string_registers_quirk(ctxt);
5408 				ctxt->eip = ctxt->_eip;
5409 				ctxt->eflags &= ~X86_EFLAGS_RF;
5410 				goto done;
5411 			}
5412 		}
5413 	}
5414 
5415 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5416 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5417 				    ctxt->src.valptr, ctxt->src.bytes);
5418 		if (rc != X86EMUL_CONTINUE)
5419 			goto done;
5420 		ctxt->src.orig_val64 = ctxt->src.val64;
5421 	}
5422 
5423 	if (ctxt->src2.type == OP_MEM) {
5424 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5425 				    &ctxt->src2.val, ctxt->src2.bytes);
5426 		if (rc != X86EMUL_CONTINUE)
5427 			goto done;
5428 	}
5429 
5430 	if ((ctxt->d & DstMask) == ImplicitOps)
5431 		goto special_insn;
5432 
5433 
5434 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5435 		/* optimisation - avoid slow emulated read if Mov */
5436 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5437 				   &ctxt->dst.val, ctxt->dst.bytes);
5438 		if (rc != X86EMUL_CONTINUE) {
5439 			if (!(ctxt->d & NoWrite) &&
5440 			    rc == X86EMUL_PROPAGATE_FAULT &&
5441 			    ctxt->exception.vector == PF_VECTOR)
5442 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5443 			goto done;
5444 		}
5445 	}
5446 	/* Copy full 64-bit value for CMPXCHG8B.  */
5447 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5448 
5449 special_insn:
5450 
5451 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5452 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5453 					      X86_ICPT_POST_MEMACCESS);
5454 		if (rc != X86EMUL_CONTINUE)
5455 			goto done;
5456 	}
5457 
5458 	if (ctxt->rep_prefix && (ctxt->d & String))
5459 		ctxt->eflags |= X86_EFLAGS_RF;
5460 	else
5461 		ctxt->eflags &= ~X86_EFLAGS_RF;
5462 
5463 	if (ctxt->execute) {
5464 		if (ctxt->d & Fastop) {
5465 			void (*fop)(struct fastop *) = (void *)ctxt->execute;
5466 			rc = fastop(ctxt, fop);
5467 			if (rc != X86EMUL_CONTINUE)
5468 				goto done;
5469 			goto writeback;
5470 		}
5471 		rc = ctxt->execute(ctxt);
5472 		if (rc != X86EMUL_CONTINUE)
5473 			goto done;
5474 		goto writeback;
5475 	}
5476 
5477 	if (ctxt->opcode_len == 2)
5478 		goto twobyte_insn;
5479 	else if (ctxt->opcode_len == 3)
5480 		goto threebyte_insn;
5481 
5482 	switch (ctxt->b) {
5483 	case 0x70 ... 0x7f: /* jcc (short) */
5484 		if (test_cc(ctxt->b, ctxt->eflags))
5485 			rc = jmp_rel(ctxt, ctxt->src.val);
5486 		break;
5487 	case 0x8d: /* lea r16/r32, m */
5488 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5489 		break;
5490 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5491 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5492 			ctxt->dst.type = OP_NONE;
5493 		else
5494 			rc = em_xchg(ctxt);
5495 		break;
5496 	case 0x98: /* cbw/cwde/cdqe */
5497 		switch (ctxt->op_bytes) {
5498 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5499 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5500 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5501 		}
5502 		break;
5503 	case 0xcc:		/* int3 */
5504 		rc = emulate_int(ctxt, 3);
5505 		break;
5506 	case 0xcd:		/* int n */
5507 		rc = emulate_int(ctxt, ctxt->src.val);
5508 		break;
5509 	case 0xce:		/* into */
5510 		if (ctxt->eflags & X86_EFLAGS_OF)
5511 			rc = emulate_int(ctxt, 4);
5512 		break;
5513 	case 0xe9: /* jmp rel */
5514 	case 0xeb: /* jmp rel short */
5515 		rc = jmp_rel(ctxt, ctxt->src.val);
5516 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5517 		break;
5518 	case 0xf4:              /* hlt */
5519 		ctxt->ops->halt(ctxt);
5520 		break;
5521 	case 0xf5:	/* cmc */
5522 		/* complement carry flag from eflags reg */
5523 		ctxt->eflags ^= X86_EFLAGS_CF;
5524 		break;
5525 	case 0xf8: /* clc */
5526 		ctxt->eflags &= ~X86_EFLAGS_CF;
5527 		break;
5528 	case 0xf9: /* stc */
5529 		ctxt->eflags |= X86_EFLAGS_CF;
5530 		break;
5531 	case 0xfc: /* cld */
5532 		ctxt->eflags &= ~X86_EFLAGS_DF;
5533 		break;
5534 	case 0xfd: /* std */
5535 		ctxt->eflags |= X86_EFLAGS_DF;
5536 		break;
5537 	default:
5538 		goto cannot_emulate;
5539 	}
5540 
5541 	if (rc != X86EMUL_CONTINUE)
5542 		goto done;
5543 
5544 writeback:
5545 	if (ctxt->d & SrcWrite) {
5546 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5547 		rc = writeback(ctxt, &ctxt->src);
5548 		if (rc != X86EMUL_CONTINUE)
5549 			goto done;
5550 	}
5551 	if (!(ctxt->d & NoWrite)) {
5552 		rc = writeback(ctxt, &ctxt->dst);
5553 		if (rc != X86EMUL_CONTINUE)
5554 			goto done;
5555 	}
5556 
5557 	/*
5558 	 * restore dst type in case the decoding will be reused
5559 	 * (happens for string instruction )
5560 	 */
5561 	ctxt->dst.type = saved_dst_type;
5562 
5563 	if ((ctxt->d & SrcMask) == SrcSI)
5564 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5565 
5566 	if ((ctxt->d & DstMask) == DstDI)
5567 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5568 
5569 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5570 		unsigned int count;
5571 		struct read_cache *r = &ctxt->io_read;
5572 		if ((ctxt->d & SrcMask) == SrcSI)
5573 			count = ctxt->src.count;
5574 		else
5575 			count = ctxt->dst.count;
5576 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5577 
5578 		if (!string_insn_completed(ctxt)) {
5579 			/*
5580 			 * Re-enter guest when pio read ahead buffer is empty
5581 			 * or, if it is not used, after each 1024 iteration.
5582 			 */
5583 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5584 			    (r->end == 0 || r->end != r->pos)) {
5585 				/*
5586 				 * Reset read cache. Usually happens before
5587 				 * decode, but since instruction is restarted
5588 				 * we have to do it here.
5589 				 */
5590 				ctxt->mem_read.end = 0;
5591 				writeback_registers(ctxt);
5592 				return EMULATION_RESTART;
5593 			}
5594 			goto done; /* skip rip writeback */
5595 		}
5596 		ctxt->eflags &= ~X86_EFLAGS_RF;
5597 	}
5598 
5599 	ctxt->eip = ctxt->_eip;
5600 
5601 done:
5602 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5603 		WARN_ON(ctxt->exception.vector > 0x1f);
5604 		ctxt->have_exception = true;
5605 	}
5606 	if (rc == X86EMUL_INTERCEPTED)
5607 		return EMULATION_INTERCEPTED;
5608 
5609 	if (rc == X86EMUL_CONTINUE)
5610 		writeback_registers(ctxt);
5611 
5612 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5613 
5614 twobyte_insn:
5615 	switch (ctxt->b) {
5616 	case 0x09:		/* wbinvd */
5617 		(ctxt->ops->wbinvd)(ctxt);
5618 		break;
5619 	case 0x08:		/* invd */
5620 	case 0x0d:		/* GrpP (prefetch) */
5621 	case 0x18:		/* Grp16 (prefetch/nop) */
5622 	case 0x1f:		/* nop */
5623 		break;
5624 	case 0x20: /* mov cr, reg */
5625 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5626 		break;
5627 	case 0x21: /* mov from dr to reg */
5628 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5629 		break;
5630 	case 0x40 ... 0x4f:	/* cmov */
5631 		if (test_cc(ctxt->b, ctxt->eflags))
5632 			ctxt->dst.val = ctxt->src.val;
5633 		else if (ctxt->op_bytes != 4)
5634 			ctxt->dst.type = OP_NONE; /* no writeback */
5635 		break;
5636 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5637 		if (test_cc(ctxt->b, ctxt->eflags))
5638 			rc = jmp_rel(ctxt, ctxt->src.val);
5639 		break;
5640 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5641 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5642 		break;
5643 	case 0xb6 ... 0xb7:	/* movzx */
5644 		ctxt->dst.bytes = ctxt->op_bytes;
5645 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5646 						       : (u16) ctxt->src.val;
5647 		break;
5648 	case 0xbe ... 0xbf:	/* movsx */
5649 		ctxt->dst.bytes = ctxt->op_bytes;
5650 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5651 							(s16) ctxt->src.val;
5652 		break;
5653 	default:
5654 		goto cannot_emulate;
5655 	}
5656 
5657 threebyte_insn:
5658 
5659 	if (rc != X86EMUL_CONTINUE)
5660 		goto done;
5661 
5662 	goto writeback;
5663 
5664 cannot_emulate:
5665 	return EMULATION_FAILED;
5666 }
5667 
5668 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5669 {
5670 	invalidate_registers(ctxt);
5671 }
5672 
5673 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5674 {
5675 	writeback_registers(ctxt);
5676 }
5677 
5678 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5679 {
5680 	if (ctxt->rep_prefix && (ctxt->d & String))
5681 		return false;
5682 
5683 	if (ctxt->d & TwoMemOp)
5684 		return false;
5685 
5686 	return true;
5687 }
5688