xref: /linux/arch/x86/kvm/emulate.c (revision 68550cbc6129159b7a6434796b721e8b66ee12f6)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3  * emulate.c
4  *
5  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
6  *
7  * Copyright (c) 2005 Keir Fraser
8  *
9  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
10  * privileged instructions:
11  *
12  * Copyright (C) 2006 Qumranet
13  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
14  *
15  *   Avi Kivity <avi@qumranet.com>
16  *   Yaniv Kamay <yaniv@qumranet.com>
17  *
18  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
19  */
20 
21 #include <linux/kvm_host.h>
22 #include "kvm_cache_regs.h"
23 #include "kvm_emulate.h"
24 #include <linux/stringify.h>
25 #include <asm/debugreg.h>
26 #include <asm/nospec-branch.h>
27 
28 #include "x86.h"
29 #include "tss.h"
30 #include "mmu.h"
31 #include "pmu.h"
32 
33 /*
34  * Operand types
35  */
36 #define OpNone             0ull
37 #define OpImplicit         1ull  /* No generic decode */
38 #define OpReg              2ull  /* Register */
39 #define OpMem              3ull  /* Memory */
40 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
41 #define OpDI               5ull  /* ES:DI/EDI/RDI */
42 #define OpMem64            6ull  /* Memory, 64-bit */
43 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
44 #define OpDX               8ull  /* DX register */
45 #define OpCL               9ull  /* CL register (for shifts) */
46 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
47 #define OpOne             11ull  /* Implied 1 */
48 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
49 #define OpMem16           13ull  /* Memory operand (16-bit). */
50 #define OpMem32           14ull  /* Memory operand (32-bit). */
51 #define OpImmU            15ull  /* Immediate operand, zero extended */
52 #define OpSI              16ull  /* SI/ESI/RSI */
53 #define OpImmFAddr        17ull  /* Immediate far address */
54 #define OpMemFAddr        18ull  /* Far address in memory */
55 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
56 #define OpES              20ull  /* ES */
57 #define OpCS              21ull  /* CS */
58 #define OpSS              22ull  /* SS */
59 #define OpDS              23ull  /* DS */
60 #define OpFS              24ull  /* FS */
61 #define OpGS              25ull  /* GS */
62 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
63 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
64 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
65 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
66 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
67 
68 #define OpBits             5  /* Width of operand field */
69 #define OpMask             ((1ull << OpBits) - 1)
70 
71 /*
72  * Opcode effective-address decode tables.
73  * Note that we only emulate instructions that have at least one memory
74  * operand (excluding implicit stack references). We assume that stack
75  * references and instruction fetches will never occur in special memory
76  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
77  * not be handled.
78  */
79 
80 /* Operand sizes: 8-bit operands or specified/overridden size. */
81 #define ByteOp      (1<<0)	/* 8-bit operands. */
82 /* Destination operand type. */
83 #define DstShift    1
84 #define ImplicitOps (OpImplicit << DstShift)
85 #define DstReg      (OpReg << DstShift)
86 #define DstMem      (OpMem << DstShift)
87 #define DstAcc      (OpAcc << DstShift)
88 #define DstDI       (OpDI << DstShift)
89 #define DstMem64    (OpMem64 << DstShift)
90 #define DstMem16    (OpMem16 << DstShift)
91 #define DstImmUByte (OpImmUByte << DstShift)
92 #define DstDX       (OpDX << DstShift)
93 #define DstAccLo    (OpAccLo << DstShift)
94 #define DstMask     (OpMask << DstShift)
95 /* Source operand type. */
96 #define SrcShift    6
97 #define SrcNone     (OpNone << SrcShift)
98 #define SrcReg      (OpReg << SrcShift)
99 #define SrcMem      (OpMem << SrcShift)
100 #define SrcMem16    (OpMem16 << SrcShift)
101 #define SrcMem32    (OpMem32 << SrcShift)
102 #define SrcImm      (OpImm << SrcShift)
103 #define SrcImmByte  (OpImmByte << SrcShift)
104 #define SrcOne      (OpOne << SrcShift)
105 #define SrcImmUByte (OpImmUByte << SrcShift)
106 #define SrcImmU     (OpImmU << SrcShift)
107 #define SrcSI       (OpSI << SrcShift)
108 #define SrcXLat     (OpXLat << SrcShift)
109 #define SrcImmFAddr (OpImmFAddr << SrcShift)
110 #define SrcMemFAddr (OpMemFAddr << SrcShift)
111 #define SrcAcc      (OpAcc << SrcShift)
112 #define SrcImmU16   (OpImmU16 << SrcShift)
113 #define SrcImm64    (OpImm64 << SrcShift)
114 #define SrcDX       (OpDX << SrcShift)
115 #define SrcMem8     (OpMem8 << SrcShift)
116 #define SrcAccHi    (OpAccHi << SrcShift)
117 #define SrcMask     (OpMask << SrcShift)
118 #define BitOp       (1<<11)
119 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
120 #define String      (1<<13)     /* String instruction (rep capable) */
121 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
122 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
123 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
124 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
125 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
126 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
127 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
128 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
129 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
130 #define Sse         (1<<18)     /* SSE Vector instruction */
131 /* Generic ModRM decode. */
132 #define ModRM       (1<<19)
133 /* Destination is only written; never read. */
134 #define Mov         (1<<20)
135 /* Misc flags */
136 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
137 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
138 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
139 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
140 #define Undefined   (1<<25) /* No Such Instruction */
141 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
142 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
143 #define No64	    (1<<28)
144 #define PageTable   (1 << 29)   /* instruction used to write page table */
145 #define NotImpl     (1 << 30)   /* instruction is not implemented */
146 /* Source 2 operand type */
147 #define Src2Shift   (31)
148 #define Src2None    (OpNone << Src2Shift)
149 #define Src2Mem     (OpMem << Src2Shift)
150 #define Src2CL      (OpCL << Src2Shift)
151 #define Src2ImmByte (OpImmByte << Src2Shift)
152 #define Src2One     (OpOne << Src2Shift)
153 #define Src2Imm     (OpImm << Src2Shift)
154 #define Src2ES      (OpES << Src2Shift)
155 #define Src2CS      (OpCS << Src2Shift)
156 #define Src2SS      (OpSS << Src2Shift)
157 #define Src2DS      (OpDS << Src2Shift)
158 #define Src2FS      (OpFS << Src2Shift)
159 #define Src2GS      (OpGS << Src2Shift)
160 #define Src2Mask    (OpMask << Src2Shift)
161 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
162 #define AlignMask   ((u64)7 << 41)
163 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
164 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
165 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
166 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
167 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
168 #define NoWrite     ((u64)1 << 45)  /* No writeback */
169 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
170 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
171 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
172 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
173 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
174 #define NearBranch  ((u64)1 << 52)  /* Near branches */
175 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
176 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
177 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
178 
179 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
180 
181 #define X2(x...) x, x
182 #define X3(x...) X2(x), x
183 #define X4(x...) X2(x), X2(x)
184 #define X5(x...) X4(x), x
185 #define X6(x...) X4(x), X2(x)
186 #define X7(x...) X4(x), X3(x)
187 #define X8(x...) X4(x), X4(x)
188 #define X16(x...) X8(x), X8(x)
189 
190 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1)
191 #define FASTOP_SIZE 8
192 
193 struct opcode {
194 	u64 flags : 56;
195 	u64 intercept : 8;
196 	union {
197 		int (*execute)(struct x86_emulate_ctxt *ctxt);
198 		const struct opcode *group;
199 		const struct group_dual *gdual;
200 		const struct gprefix *gprefix;
201 		const struct escape *esc;
202 		const struct instr_dual *idual;
203 		const struct mode_dual *mdual;
204 		void (*fastop)(struct fastop *fake);
205 	} u;
206 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
207 };
208 
209 struct group_dual {
210 	struct opcode mod012[8];
211 	struct opcode mod3[8];
212 };
213 
214 struct gprefix {
215 	struct opcode pfx_no;
216 	struct opcode pfx_66;
217 	struct opcode pfx_f2;
218 	struct opcode pfx_f3;
219 };
220 
221 struct escape {
222 	struct opcode op[8];
223 	struct opcode high[64];
224 };
225 
226 struct instr_dual {
227 	struct opcode mod012;
228 	struct opcode mod3;
229 };
230 
231 struct mode_dual {
232 	struct opcode mode32;
233 	struct opcode mode64;
234 };
235 
236 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
237 
238 enum x86_transfer_type {
239 	X86_TRANSFER_NONE,
240 	X86_TRANSFER_CALL_JMP,
241 	X86_TRANSFER_RET,
242 	X86_TRANSFER_TASK_SWITCH,
243 };
244 
245 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
246 {
247 	if (!(ctxt->regs_valid & (1 << nr))) {
248 		ctxt->regs_valid |= 1 << nr;
249 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
250 	}
251 	return ctxt->_regs[nr];
252 }
253 
254 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
255 {
256 	ctxt->regs_valid |= 1 << nr;
257 	ctxt->regs_dirty |= 1 << nr;
258 	return &ctxt->_regs[nr];
259 }
260 
261 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
262 {
263 	reg_read(ctxt, nr);
264 	return reg_write(ctxt, nr);
265 }
266 
267 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
268 {
269 	unsigned reg;
270 
271 	for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16)
272 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
273 }
274 
275 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
276 {
277 	ctxt->regs_dirty = 0;
278 	ctxt->regs_valid = 0;
279 }
280 
281 /*
282  * These EFLAGS bits are restored from saved value during emulation, and
283  * any changes are written back to the saved value after emulation.
284  */
285 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
286 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
287 
288 #ifdef CONFIG_X86_64
289 #define ON64(x) x
290 #else
291 #define ON64(x)
292 #endif
293 
294 /*
295  * fastop functions have a special calling convention:
296  *
297  * dst:    rax        (in/out)
298  * src:    rdx        (in/out)
299  * src2:   rcx        (in)
300  * flags:  rflags     (in/out)
301  * ex:     rsi        (in:fastop pointer, out:zero if exception)
302  *
303  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
304  * different operand sizes can be reached by calculation, rather than a jump
305  * table (which would be bigger than the code).
306  */
307 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
308 
309 #define __FOP_FUNC(name) \
310 	".align " __stringify(FASTOP_SIZE) " \n\t" \
311 	".type " name ", @function \n\t" \
312 	name ":\n\t"
313 
314 #define FOP_FUNC(name) \
315 	__FOP_FUNC(#name)
316 
317 #define __FOP_RET(name) \
318 	"ret \n\t" \
319 	".size " name ", .-" name "\n\t"
320 
321 #define FOP_RET(name) \
322 	__FOP_RET(#name)
323 
324 #define FOP_START(op) \
325 	extern void em_##op(struct fastop *fake); \
326 	asm(".pushsection .text, \"ax\" \n\t" \
327 	    ".global em_" #op " \n\t" \
328 	    ".align " __stringify(FASTOP_SIZE) " \n\t" \
329 	    "em_" #op ":\n\t"
330 
331 #define FOP_END \
332 	    ".popsection")
333 
334 #define __FOPNOP(name) \
335 	__FOP_FUNC(name) \
336 	__FOP_RET(name)
337 
338 #define FOPNOP() \
339 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
340 
341 #define FOP1E(op,  dst) \
342 	__FOP_FUNC(#op "_" #dst) \
343 	"10: " #op " %" #dst " \n\t" \
344 	__FOP_RET(#op "_" #dst)
345 
346 #define FOP1EEX(op,  dst) \
347 	FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception)
348 
349 #define FASTOP1(op) \
350 	FOP_START(op) \
351 	FOP1E(op##b, al) \
352 	FOP1E(op##w, ax) \
353 	FOP1E(op##l, eax) \
354 	ON64(FOP1E(op##q, rax))	\
355 	FOP_END
356 
357 /* 1-operand, using src2 (for MUL/DIV r/m) */
358 #define FASTOP1SRC2(op, name) \
359 	FOP_START(name) \
360 	FOP1E(op, cl) \
361 	FOP1E(op, cx) \
362 	FOP1E(op, ecx) \
363 	ON64(FOP1E(op, rcx)) \
364 	FOP_END
365 
366 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
367 #define FASTOP1SRC2EX(op, name) \
368 	FOP_START(name) \
369 	FOP1EEX(op, cl) \
370 	FOP1EEX(op, cx) \
371 	FOP1EEX(op, ecx) \
372 	ON64(FOP1EEX(op, rcx)) \
373 	FOP_END
374 
375 #define FOP2E(op,  dst, src)	   \
376 	__FOP_FUNC(#op "_" #dst "_" #src) \
377 	#op " %" #src ", %" #dst " \n\t" \
378 	__FOP_RET(#op "_" #dst "_" #src)
379 
380 #define FASTOP2(op) \
381 	FOP_START(op) \
382 	FOP2E(op##b, al, dl) \
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, word only */
389 #define FASTOP2W(op) \
390 	FOP_START(op) \
391 	FOPNOP() \
392 	FOP2E(op##w, ax, dx) \
393 	FOP2E(op##l, eax, edx) \
394 	ON64(FOP2E(op##q, rax, rdx)) \
395 	FOP_END
396 
397 /* 2 operand, src is CL */
398 #define FASTOP2CL(op) \
399 	FOP_START(op) \
400 	FOP2E(op##b, al, cl) \
401 	FOP2E(op##w, ax, cl) \
402 	FOP2E(op##l, eax, cl) \
403 	ON64(FOP2E(op##q, rax, cl)) \
404 	FOP_END
405 
406 /* 2 operand, src and dest are reversed */
407 #define FASTOP2R(op, name) \
408 	FOP_START(name) \
409 	FOP2E(op##b, dl, al) \
410 	FOP2E(op##w, dx, ax) \
411 	FOP2E(op##l, edx, eax) \
412 	ON64(FOP2E(op##q, rdx, rax)) \
413 	FOP_END
414 
415 #define FOP3E(op,  dst, src, src2) \
416 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
417 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
418 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
419 
420 /* 3-operand, word-only, src2=cl */
421 #define FASTOP3WCL(op) \
422 	FOP_START(op) \
423 	FOPNOP() \
424 	FOP3E(op##w, ax, dx, cl) \
425 	FOP3E(op##l, eax, edx, cl) \
426 	ON64(FOP3E(op##q, rax, rdx, cl)) \
427 	FOP_END
428 
429 /* Special case for SETcc - 1 instruction per cc */
430 #define FOP_SETCC(op) \
431 	".align 4 \n\t" \
432 	".type " #op ", @function \n\t" \
433 	#op ": \n\t" \
434 	#op " %al \n\t" \
435 	__FOP_RET(#op)
436 
437 asm(".pushsection .fixup, \"ax\"\n"
438     "kvm_fastop_exception: xor %esi, %esi; ret\n"
439     ".popsection");
440 
441 FOP_START(setcc)
442 FOP_SETCC(seto)
443 FOP_SETCC(setno)
444 FOP_SETCC(setc)
445 FOP_SETCC(setnc)
446 FOP_SETCC(setz)
447 FOP_SETCC(setnz)
448 FOP_SETCC(setbe)
449 FOP_SETCC(setnbe)
450 FOP_SETCC(sets)
451 FOP_SETCC(setns)
452 FOP_SETCC(setp)
453 FOP_SETCC(setnp)
454 FOP_SETCC(setl)
455 FOP_SETCC(setnl)
456 FOP_SETCC(setle)
457 FOP_SETCC(setnle)
458 FOP_END;
459 
460 FOP_START(salc)
461 FOP_FUNC(salc)
462 "pushf; sbb %al, %al; popf \n\t"
463 FOP_RET(salc)
464 FOP_END;
465 
466 /*
467  * XXX: inoutclob user must know where the argument is being expanded.
468  *      Relying on CONFIG_CC_HAS_ASM_GOTO would allow us to remove _fault.
469  */
470 #define asm_safe(insn, inoutclob...) \
471 ({ \
472 	int _fault = 0; \
473  \
474 	asm volatile("1:" insn "\n" \
475 	             "2:\n" \
476 	             ".pushsection .fixup, \"ax\"\n" \
477 	             "3: movl $1, %[_fault]\n" \
478 	             "   jmp  2b\n" \
479 	             ".popsection\n" \
480 	             _ASM_EXTABLE(1b, 3b) \
481 	             : [_fault] "+qm"(_fault) inoutclob ); \
482  \
483 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
484 })
485 
486 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
487 				    enum x86_intercept intercept,
488 				    enum x86_intercept_stage stage)
489 {
490 	struct x86_instruction_info info = {
491 		.intercept  = intercept,
492 		.rep_prefix = ctxt->rep_prefix,
493 		.modrm_mod  = ctxt->modrm_mod,
494 		.modrm_reg  = ctxt->modrm_reg,
495 		.modrm_rm   = ctxt->modrm_rm,
496 		.src_val    = ctxt->src.val64,
497 		.dst_val    = ctxt->dst.val64,
498 		.src_bytes  = ctxt->src.bytes,
499 		.dst_bytes  = ctxt->dst.bytes,
500 		.ad_bytes   = ctxt->ad_bytes,
501 		.next_rip   = ctxt->eip,
502 	};
503 
504 	return ctxt->ops->intercept(ctxt, &info, stage);
505 }
506 
507 static void assign_masked(ulong *dest, ulong src, ulong mask)
508 {
509 	*dest = (*dest & ~mask) | (src & mask);
510 }
511 
512 static void assign_register(unsigned long *reg, u64 val, int bytes)
513 {
514 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
515 	switch (bytes) {
516 	case 1:
517 		*(u8 *)reg = (u8)val;
518 		break;
519 	case 2:
520 		*(u16 *)reg = (u16)val;
521 		break;
522 	case 4:
523 		*reg = (u32)val;
524 		break;	/* 64b: zero-extend */
525 	case 8:
526 		*reg = val;
527 		break;
528 	}
529 }
530 
531 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
532 {
533 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
534 }
535 
536 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
537 {
538 	u16 sel;
539 	struct desc_struct ss;
540 
541 	if (ctxt->mode == X86EMUL_MODE_PROT64)
542 		return ~0UL;
543 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
544 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
545 }
546 
547 static int stack_size(struct x86_emulate_ctxt *ctxt)
548 {
549 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
550 }
551 
552 /* Access/update address held in a register, based on addressing mode. */
553 static inline unsigned long
554 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
555 {
556 	if (ctxt->ad_bytes == sizeof(unsigned long))
557 		return reg;
558 	else
559 		return reg & ad_mask(ctxt);
560 }
561 
562 static inline unsigned long
563 register_address(struct x86_emulate_ctxt *ctxt, int reg)
564 {
565 	return address_mask(ctxt, reg_read(ctxt, reg));
566 }
567 
568 static void masked_increment(ulong *reg, ulong mask, int inc)
569 {
570 	assign_masked(reg, *reg + inc, mask);
571 }
572 
573 static inline void
574 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
575 {
576 	ulong *preg = reg_rmw(ctxt, reg);
577 
578 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
579 }
580 
581 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
582 {
583 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
584 }
585 
586 static u32 desc_limit_scaled(struct desc_struct *desc)
587 {
588 	u32 limit = get_desc_limit(desc);
589 
590 	return desc->g ? (limit << 12) | 0xfff : limit;
591 }
592 
593 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
594 {
595 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
596 		return 0;
597 
598 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
599 }
600 
601 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
602 			     u32 error, bool valid)
603 {
604 	WARN_ON(vec > 0x1f);
605 	ctxt->exception.vector = vec;
606 	ctxt->exception.error_code = error;
607 	ctxt->exception.error_code_valid = valid;
608 	return X86EMUL_PROPAGATE_FAULT;
609 }
610 
611 static int emulate_db(struct x86_emulate_ctxt *ctxt)
612 {
613 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
614 }
615 
616 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
617 {
618 	return emulate_exception(ctxt, GP_VECTOR, err, true);
619 }
620 
621 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
622 {
623 	return emulate_exception(ctxt, SS_VECTOR, err, true);
624 }
625 
626 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
627 {
628 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
629 }
630 
631 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
632 {
633 	return emulate_exception(ctxt, TS_VECTOR, err, true);
634 }
635 
636 static int emulate_de(struct x86_emulate_ctxt *ctxt)
637 {
638 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
639 }
640 
641 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
642 {
643 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
644 }
645 
646 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
647 {
648 	u16 selector;
649 	struct desc_struct desc;
650 
651 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
652 	return selector;
653 }
654 
655 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
656 				 unsigned seg)
657 {
658 	u16 dummy;
659 	u32 base3;
660 	struct desc_struct desc;
661 
662 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
663 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
664 }
665 
666 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
667 {
668 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
669 }
670 
671 static inline bool emul_is_noncanonical_address(u64 la,
672 						struct x86_emulate_ctxt *ctxt)
673 {
674 	return get_canonical(la, ctxt_virt_addr_bits(ctxt)) != la;
675 }
676 
677 /*
678  * x86 defines three classes of vector instructions: explicitly
679  * aligned, explicitly unaligned, and the rest, which change behaviour
680  * depending on whether they're AVX encoded or not.
681  *
682  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
683  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
684  * 512 bytes of data must be aligned to a 16 byte boundary.
685  */
686 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
687 {
688 	u64 alignment = ctxt->d & AlignMask;
689 
690 	if (likely(size < 16))
691 		return 1;
692 
693 	switch (alignment) {
694 	case Unaligned:
695 	case Avx:
696 		return 1;
697 	case Aligned16:
698 		return 16;
699 	case Aligned:
700 	default:
701 		return size;
702 	}
703 }
704 
705 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
706 				       struct segmented_address addr,
707 				       unsigned *max_size, unsigned size,
708 				       bool write, bool fetch,
709 				       enum x86emul_mode mode, ulong *linear)
710 {
711 	struct desc_struct desc;
712 	bool usable;
713 	ulong la;
714 	u32 lim;
715 	u16 sel;
716 	u8  va_bits;
717 
718 	la = seg_base(ctxt, addr.seg) + addr.ea;
719 	*max_size = 0;
720 	switch (mode) {
721 	case X86EMUL_MODE_PROT64:
722 		*linear = la;
723 		va_bits = ctxt_virt_addr_bits(ctxt);
724 		if (get_canonical(la, va_bits) != la)
725 			goto bad;
726 
727 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
728 		if (size > *max_size)
729 			goto bad;
730 		break;
731 	default:
732 		*linear = la = (u32)la;
733 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
734 						addr.seg);
735 		if (!usable)
736 			goto bad;
737 		/* code segment in protected mode or read-only data segment */
738 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
739 					|| !(desc.type & 2)) && write)
740 			goto bad;
741 		/* unreadable code segment */
742 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
743 			goto bad;
744 		lim = desc_limit_scaled(&desc);
745 		if (!(desc.type & 8) && (desc.type & 4)) {
746 			/* expand-down segment */
747 			if (addr.ea <= lim)
748 				goto bad;
749 			lim = desc.d ? 0xffffffff : 0xffff;
750 		}
751 		if (addr.ea > lim)
752 			goto bad;
753 		if (lim == 0xffffffff)
754 			*max_size = ~0u;
755 		else {
756 			*max_size = (u64)lim + 1 - addr.ea;
757 			if (size > *max_size)
758 				goto bad;
759 		}
760 		break;
761 	}
762 	if (la & (insn_alignment(ctxt, size) - 1))
763 		return emulate_gp(ctxt, 0);
764 	return X86EMUL_CONTINUE;
765 bad:
766 	if (addr.seg == VCPU_SREG_SS)
767 		return emulate_ss(ctxt, 0);
768 	else
769 		return emulate_gp(ctxt, 0);
770 }
771 
772 static int linearize(struct x86_emulate_ctxt *ctxt,
773 		     struct segmented_address addr,
774 		     unsigned size, bool write,
775 		     ulong *linear)
776 {
777 	unsigned max_size;
778 	return __linearize(ctxt, addr, &max_size, size, write, false,
779 			   ctxt->mode, linear);
780 }
781 
782 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
783 			     enum x86emul_mode mode)
784 {
785 	ulong linear;
786 	int rc;
787 	unsigned max_size;
788 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
789 					   .ea = dst };
790 
791 	if (ctxt->op_bytes != sizeof(unsigned long))
792 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
793 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
794 	if (rc == X86EMUL_CONTINUE)
795 		ctxt->_eip = addr.ea;
796 	return rc;
797 }
798 
799 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
800 {
801 	return assign_eip(ctxt, dst, ctxt->mode);
802 }
803 
804 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
805 			  const struct desc_struct *cs_desc)
806 {
807 	enum x86emul_mode mode = ctxt->mode;
808 	int rc;
809 
810 #ifdef CONFIG_X86_64
811 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
812 		if (cs_desc->l) {
813 			u64 efer = 0;
814 
815 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
816 			if (efer & EFER_LMA)
817 				mode = X86EMUL_MODE_PROT64;
818 		} else
819 			mode = X86EMUL_MODE_PROT32; /* temporary value */
820 	}
821 #endif
822 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
823 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
824 	rc = assign_eip(ctxt, dst, mode);
825 	if (rc == X86EMUL_CONTINUE)
826 		ctxt->mode = mode;
827 	return rc;
828 }
829 
830 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
831 {
832 	return assign_eip_near(ctxt, ctxt->_eip + rel);
833 }
834 
835 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear,
836 			      void *data, unsigned size)
837 {
838 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true);
839 }
840 
841 static int linear_write_system(struct x86_emulate_ctxt *ctxt,
842 			       ulong linear, void *data,
843 			       unsigned int size)
844 {
845 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true);
846 }
847 
848 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
849 			      struct segmented_address addr,
850 			      void *data,
851 			      unsigned size)
852 {
853 	int rc;
854 	ulong linear;
855 
856 	rc = linearize(ctxt, addr, size, false, &linear);
857 	if (rc != X86EMUL_CONTINUE)
858 		return rc;
859 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false);
860 }
861 
862 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
863 			       struct segmented_address addr,
864 			       void *data,
865 			       unsigned int size)
866 {
867 	int rc;
868 	ulong linear;
869 
870 	rc = linearize(ctxt, addr, size, true, &linear);
871 	if (rc != X86EMUL_CONTINUE)
872 		return rc;
873 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false);
874 }
875 
876 /*
877  * Prefetch the remaining bytes of the instruction without crossing page
878  * boundary if they are not in fetch_cache yet.
879  */
880 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
881 {
882 	int rc;
883 	unsigned size, max_size;
884 	unsigned long linear;
885 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
886 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
887 					   .ea = ctxt->eip + cur_size };
888 
889 	/*
890 	 * We do not know exactly how many bytes will be needed, and
891 	 * __linearize is expensive, so fetch as much as possible.  We
892 	 * just have to avoid going beyond the 15 byte limit, the end
893 	 * of the segment, or the end of the page.
894 	 *
895 	 * __linearize is called with size 0 so that it does not do any
896 	 * boundary check itself.  Instead, we use max_size to check
897 	 * against op_size.
898 	 */
899 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
900 			 &linear);
901 	if (unlikely(rc != X86EMUL_CONTINUE))
902 		return rc;
903 
904 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
905 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
906 
907 	/*
908 	 * One instruction can only straddle two pages,
909 	 * and one has been loaded at the beginning of
910 	 * x86_decode_insn.  So, if not enough bytes
911 	 * still, we must have hit the 15-byte boundary.
912 	 */
913 	if (unlikely(size < op_size))
914 		return emulate_gp(ctxt, 0);
915 
916 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
917 			      size, &ctxt->exception);
918 	if (unlikely(rc != X86EMUL_CONTINUE))
919 		return rc;
920 	ctxt->fetch.end += size;
921 	return X86EMUL_CONTINUE;
922 }
923 
924 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
925 					       unsigned size)
926 {
927 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
928 
929 	if (unlikely(done_size < size))
930 		return __do_insn_fetch_bytes(ctxt, size - done_size);
931 	else
932 		return X86EMUL_CONTINUE;
933 }
934 
935 /* Fetch next part of the instruction being emulated. */
936 #define insn_fetch(_type, _ctxt)					\
937 ({	_type _x;							\
938 									\
939 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
940 	if (rc != X86EMUL_CONTINUE)					\
941 		goto done;						\
942 	ctxt->_eip += sizeof(_type);					\
943 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
944 	ctxt->fetch.ptr += sizeof(_type);				\
945 	_x;								\
946 })
947 
948 #define insn_fetch_arr(_arr, _size, _ctxt)				\
949 ({									\
950 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
951 	if (rc != X86EMUL_CONTINUE)					\
952 		goto done;						\
953 	ctxt->_eip += (_size);						\
954 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
955 	ctxt->fetch.ptr += (_size);					\
956 })
957 
958 /*
959  * Given the 'reg' portion of a ModRM byte, and a register block, return a
960  * pointer into the block that addresses the relevant register.
961  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
962  */
963 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
964 			     int byteop)
965 {
966 	void *p;
967 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
968 
969 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
970 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
971 	else
972 		p = reg_rmw(ctxt, modrm_reg);
973 	return p;
974 }
975 
976 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
977 			   struct segmented_address addr,
978 			   u16 *size, unsigned long *address, int op_bytes)
979 {
980 	int rc;
981 
982 	if (op_bytes == 2)
983 		op_bytes = 3;
984 	*address = 0;
985 	rc = segmented_read_std(ctxt, addr, size, 2);
986 	if (rc != X86EMUL_CONTINUE)
987 		return rc;
988 	addr.ea += 2;
989 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
990 	return rc;
991 }
992 
993 FASTOP2(add);
994 FASTOP2(or);
995 FASTOP2(adc);
996 FASTOP2(sbb);
997 FASTOP2(and);
998 FASTOP2(sub);
999 FASTOP2(xor);
1000 FASTOP2(cmp);
1001 FASTOP2(test);
1002 
1003 FASTOP1SRC2(mul, mul_ex);
1004 FASTOP1SRC2(imul, imul_ex);
1005 FASTOP1SRC2EX(div, div_ex);
1006 FASTOP1SRC2EX(idiv, idiv_ex);
1007 
1008 FASTOP3WCL(shld);
1009 FASTOP3WCL(shrd);
1010 
1011 FASTOP2W(imul);
1012 
1013 FASTOP1(not);
1014 FASTOP1(neg);
1015 FASTOP1(inc);
1016 FASTOP1(dec);
1017 
1018 FASTOP2CL(rol);
1019 FASTOP2CL(ror);
1020 FASTOP2CL(rcl);
1021 FASTOP2CL(rcr);
1022 FASTOP2CL(shl);
1023 FASTOP2CL(shr);
1024 FASTOP2CL(sar);
1025 
1026 FASTOP2W(bsf);
1027 FASTOP2W(bsr);
1028 FASTOP2W(bt);
1029 FASTOP2W(bts);
1030 FASTOP2W(btr);
1031 FASTOP2W(btc);
1032 
1033 FASTOP2(xadd);
1034 
1035 FASTOP2R(cmp, cmp_r);
1036 
1037 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1038 {
1039 	/* If src is zero, do not writeback, but update flags */
1040 	if (ctxt->src.val == 0)
1041 		ctxt->dst.type = OP_NONE;
1042 	return fastop(ctxt, em_bsf);
1043 }
1044 
1045 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1046 {
1047 	/* If src is zero, do not writeback, but update flags */
1048 	if (ctxt->src.val == 0)
1049 		ctxt->dst.type = OP_NONE;
1050 	return fastop(ctxt, em_bsr);
1051 }
1052 
1053 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1054 {
1055 	u8 rc;
1056 	void (*fop)(void) = (void *)em_setcc + 4 * (condition & 0xf);
1057 
1058 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1059 	asm("push %[flags]; popf; " CALL_NOSPEC
1060 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1061 	return rc;
1062 }
1063 
1064 static void fetch_register_operand(struct operand *op)
1065 {
1066 	switch (op->bytes) {
1067 	case 1:
1068 		op->val = *(u8 *)op->addr.reg;
1069 		break;
1070 	case 2:
1071 		op->val = *(u16 *)op->addr.reg;
1072 		break;
1073 	case 4:
1074 		op->val = *(u32 *)op->addr.reg;
1075 		break;
1076 	case 8:
1077 		op->val = *(u64 *)op->addr.reg;
1078 		break;
1079 	}
1080 }
1081 
1082 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1083 {
1084 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1085 		return emulate_nm(ctxt);
1086 
1087 	kvm_fpu_get();
1088 	asm volatile("fninit");
1089 	kvm_fpu_put();
1090 	return X86EMUL_CONTINUE;
1091 }
1092 
1093 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1094 {
1095 	u16 fcw;
1096 
1097 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1098 		return emulate_nm(ctxt);
1099 
1100 	kvm_fpu_get();
1101 	asm volatile("fnstcw %0": "+m"(fcw));
1102 	kvm_fpu_put();
1103 
1104 	ctxt->dst.val = fcw;
1105 
1106 	return X86EMUL_CONTINUE;
1107 }
1108 
1109 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1110 {
1111 	u16 fsw;
1112 
1113 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1114 		return emulate_nm(ctxt);
1115 
1116 	kvm_fpu_get();
1117 	asm volatile("fnstsw %0": "+m"(fsw));
1118 	kvm_fpu_put();
1119 
1120 	ctxt->dst.val = fsw;
1121 
1122 	return X86EMUL_CONTINUE;
1123 }
1124 
1125 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1126 				    struct operand *op)
1127 {
1128 	unsigned reg = ctxt->modrm_reg;
1129 
1130 	if (!(ctxt->d & ModRM))
1131 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1132 
1133 	if (ctxt->d & Sse) {
1134 		op->type = OP_XMM;
1135 		op->bytes = 16;
1136 		op->addr.xmm = reg;
1137 		kvm_read_sse_reg(reg, &op->vec_val);
1138 		return;
1139 	}
1140 	if (ctxt->d & Mmx) {
1141 		reg &= 7;
1142 		op->type = OP_MM;
1143 		op->bytes = 8;
1144 		op->addr.mm = reg;
1145 		return;
1146 	}
1147 
1148 	op->type = OP_REG;
1149 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1150 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1151 
1152 	fetch_register_operand(op);
1153 	op->orig_val = op->val;
1154 }
1155 
1156 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1157 {
1158 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1159 		ctxt->modrm_seg = VCPU_SREG_SS;
1160 }
1161 
1162 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1163 			struct operand *op)
1164 {
1165 	u8 sib;
1166 	int index_reg, base_reg, scale;
1167 	int rc = X86EMUL_CONTINUE;
1168 	ulong modrm_ea = 0;
1169 
1170 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1171 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1172 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1173 
1174 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1175 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1176 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1177 	ctxt->modrm_seg = VCPU_SREG_DS;
1178 
1179 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1180 		op->type = OP_REG;
1181 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1182 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1183 				ctxt->d & ByteOp);
1184 		if (ctxt->d & Sse) {
1185 			op->type = OP_XMM;
1186 			op->bytes = 16;
1187 			op->addr.xmm = ctxt->modrm_rm;
1188 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1189 			return rc;
1190 		}
1191 		if (ctxt->d & Mmx) {
1192 			op->type = OP_MM;
1193 			op->bytes = 8;
1194 			op->addr.mm = ctxt->modrm_rm & 7;
1195 			return rc;
1196 		}
1197 		fetch_register_operand(op);
1198 		return rc;
1199 	}
1200 
1201 	op->type = OP_MEM;
1202 
1203 	if (ctxt->ad_bytes == 2) {
1204 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1205 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1206 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1207 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1208 
1209 		/* 16-bit ModR/M decode. */
1210 		switch (ctxt->modrm_mod) {
1211 		case 0:
1212 			if (ctxt->modrm_rm == 6)
1213 				modrm_ea += insn_fetch(u16, ctxt);
1214 			break;
1215 		case 1:
1216 			modrm_ea += insn_fetch(s8, ctxt);
1217 			break;
1218 		case 2:
1219 			modrm_ea += insn_fetch(u16, ctxt);
1220 			break;
1221 		}
1222 		switch (ctxt->modrm_rm) {
1223 		case 0:
1224 			modrm_ea += bx + si;
1225 			break;
1226 		case 1:
1227 			modrm_ea += bx + di;
1228 			break;
1229 		case 2:
1230 			modrm_ea += bp + si;
1231 			break;
1232 		case 3:
1233 			modrm_ea += bp + di;
1234 			break;
1235 		case 4:
1236 			modrm_ea += si;
1237 			break;
1238 		case 5:
1239 			modrm_ea += di;
1240 			break;
1241 		case 6:
1242 			if (ctxt->modrm_mod != 0)
1243 				modrm_ea += bp;
1244 			break;
1245 		case 7:
1246 			modrm_ea += bx;
1247 			break;
1248 		}
1249 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1250 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1251 			ctxt->modrm_seg = VCPU_SREG_SS;
1252 		modrm_ea = (u16)modrm_ea;
1253 	} else {
1254 		/* 32/64-bit ModR/M decode. */
1255 		if ((ctxt->modrm_rm & 7) == 4) {
1256 			sib = insn_fetch(u8, ctxt);
1257 			index_reg |= (sib >> 3) & 7;
1258 			base_reg |= sib & 7;
1259 			scale = sib >> 6;
1260 
1261 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1262 				modrm_ea += insn_fetch(s32, ctxt);
1263 			else {
1264 				modrm_ea += reg_read(ctxt, base_reg);
1265 				adjust_modrm_seg(ctxt, base_reg);
1266 				/* Increment ESP on POP [ESP] */
1267 				if ((ctxt->d & IncSP) &&
1268 				    base_reg == VCPU_REGS_RSP)
1269 					modrm_ea += ctxt->op_bytes;
1270 			}
1271 			if (index_reg != 4)
1272 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1273 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1274 			modrm_ea += insn_fetch(s32, ctxt);
1275 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1276 				ctxt->rip_relative = 1;
1277 		} else {
1278 			base_reg = ctxt->modrm_rm;
1279 			modrm_ea += reg_read(ctxt, base_reg);
1280 			adjust_modrm_seg(ctxt, base_reg);
1281 		}
1282 		switch (ctxt->modrm_mod) {
1283 		case 1:
1284 			modrm_ea += insn_fetch(s8, ctxt);
1285 			break;
1286 		case 2:
1287 			modrm_ea += insn_fetch(s32, ctxt);
1288 			break;
1289 		}
1290 	}
1291 	op->addr.mem.ea = modrm_ea;
1292 	if (ctxt->ad_bytes != 8)
1293 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1294 
1295 done:
1296 	return rc;
1297 }
1298 
1299 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1300 		      struct operand *op)
1301 {
1302 	int rc = X86EMUL_CONTINUE;
1303 
1304 	op->type = OP_MEM;
1305 	switch (ctxt->ad_bytes) {
1306 	case 2:
1307 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1308 		break;
1309 	case 4:
1310 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1311 		break;
1312 	case 8:
1313 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1314 		break;
1315 	}
1316 done:
1317 	return rc;
1318 }
1319 
1320 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1321 {
1322 	long sv = 0, mask;
1323 
1324 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1325 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1326 
1327 		if (ctxt->src.bytes == 2)
1328 			sv = (s16)ctxt->src.val & (s16)mask;
1329 		else if (ctxt->src.bytes == 4)
1330 			sv = (s32)ctxt->src.val & (s32)mask;
1331 		else
1332 			sv = (s64)ctxt->src.val & (s64)mask;
1333 
1334 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1335 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1336 	}
1337 
1338 	/* only subword offset */
1339 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1340 }
1341 
1342 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1343 			 unsigned long addr, void *dest, unsigned size)
1344 {
1345 	int rc;
1346 	struct read_cache *mc = &ctxt->mem_read;
1347 
1348 	if (mc->pos < mc->end)
1349 		goto read_cached;
1350 
1351 	WARN_ON((mc->end + size) >= sizeof(mc->data));
1352 
1353 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1354 				      &ctxt->exception);
1355 	if (rc != X86EMUL_CONTINUE)
1356 		return rc;
1357 
1358 	mc->end += size;
1359 
1360 read_cached:
1361 	memcpy(dest, mc->data + mc->pos, size);
1362 	mc->pos += size;
1363 	return X86EMUL_CONTINUE;
1364 }
1365 
1366 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1367 			  struct segmented_address addr,
1368 			  void *data,
1369 			  unsigned size)
1370 {
1371 	int rc;
1372 	ulong linear;
1373 
1374 	rc = linearize(ctxt, addr, size, false, &linear);
1375 	if (rc != X86EMUL_CONTINUE)
1376 		return rc;
1377 	return read_emulated(ctxt, linear, data, size);
1378 }
1379 
1380 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1381 			   struct segmented_address addr,
1382 			   const void *data,
1383 			   unsigned size)
1384 {
1385 	int rc;
1386 	ulong linear;
1387 
1388 	rc = linearize(ctxt, addr, size, true, &linear);
1389 	if (rc != X86EMUL_CONTINUE)
1390 		return rc;
1391 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1392 					 &ctxt->exception);
1393 }
1394 
1395 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1396 			     struct segmented_address addr,
1397 			     const void *orig_data, const void *data,
1398 			     unsigned size)
1399 {
1400 	int rc;
1401 	ulong linear;
1402 
1403 	rc = linearize(ctxt, addr, size, true, &linear);
1404 	if (rc != X86EMUL_CONTINUE)
1405 		return rc;
1406 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1407 					   size, &ctxt->exception);
1408 }
1409 
1410 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1411 			   unsigned int size, unsigned short port,
1412 			   void *dest)
1413 {
1414 	struct read_cache *rc = &ctxt->io_read;
1415 
1416 	if (rc->pos == rc->end) { /* refill pio read ahead */
1417 		unsigned int in_page, n;
1418 		unsigned int count = ctxt->rep_prefix ?
1419 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1420 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1421 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1422 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1423 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1424 		if (n == 0)
1425 			n = 1;
1426 		rc->pos = rc->end = 0;
1427 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1428 			return 0;
1429 		rc->end = n * size;
1430 	}
1431 
1432 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1433 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1434 		ctxt->dst.data = rc->data + rc->pos;
1435 		ctxt->dst.type = OP_MEM_STR;
1436 		ctxt->dst.count = (rc->end - rc->pos) / size;
1437 		rc->pos = rc->end;
1438 	} else {
1439 		memcpy(dest, rc->data + rc->pos, size);
1440 		rc->pos += size;
1441 	}
1442 	return 1;
1443 }
1444 
1445 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1446 				     u16 index, struct desc_struct *desc)
1447 {
1448 	struct desc_ptr dt;
1449 	ulong addr;
1450 
1451 	ctxt->ops->get_idt(ctxt, &dt);
1452 
1453 	if (dt.size < index * 8 + 7)
1454 		return emulate_gp(ctxt, index << 3 | 0x2);
1455 
1456 	addr = dt.address + index * 8;
1457 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1458 }
1459 
1460 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1461 				     u16 selector, struct desc_ptr *dt)
1462 {
1463 	const struct x86_emulate_ops *ops = ctxt->ops;
1464 	u32 base3 = 0;
1465 
1466 	if (selector & 1 << 2) {
1467 		struct desc_struct desc;
1468 		u16 sel;
1469 
1470 		memset(dt, 0, sizeof(*dt));
1471 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1472 				      VCPU_SREG_LDTR))
1473 			return;
1474 
1475 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1476 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1477 	} else
1478 		ops->get_gdt(ctxt, dt);
1479 }
1480 
1481 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1482 			      u16 selector, ulong *desc_addr_p)
1483 {
1484 	struct desc_ptr dt;
1485 	u16 index = selector >> 3;
1486 	ulong addr;
1487 
1488 	get_descriptor_table_ptr(ctxt, selector, &dt);
1489 
1490 	if (dt.size < index * 8 + 7)
1491 		return emulate_gp(ctxt, selector & 0xfffc);
1492 
1493 	addr = dt.address + index * 8;
1494 
1495 #ifdef CONFIG_X86_64
1496 	if (addr >> 32 != 0) {
1497 		u64 efer = 0;
1498 
1499 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1500 		if (!(efer & EFER_LMA))
1501 			addr &= (u32)-1;
1502 	}
1503 #endif
1504 
1505 	*desc_addr_p = addr;
1506 	return X86EMUL_CONTINUE;
1507 }
1508 
1509 /* allowed just for 8 bytes segments */
1510 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1511 				   u16 selector, struct desc_struct *desc,
1512 				   ulong *desc_addr_p)
1513 {
1514 	int rc;
1515 
1516 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1517 	if (rc != X86EMUL_CONTINUE)
1518 		return rc;
1519 
1520 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1521 }
1522 
1523 /* allowed just for 8 bytes segments */
1524 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1525 				    u16 selector, struct desc_struct *desc)
1526 {
1527 	int rc;
1528 	ulong addr;
1529 
1530 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1531 	if (rc != X86EMUL_CONTINUE)
1532 		return rc;
1533 
1534 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1535 }
1536 
1537 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1538 				     u16 selector, int seg, u8 cpl,
1539 				     enum x86_transfer_type transfer,
1540 				     struct desc_struct *desc)
1541 {
1542 	struct desc_struct seg_desc, old_desc;
1543 	u8 dpl, rpl;
1544 	unsigned err_vec = GP_VECTOR;
1545 	u32 err_code = 0;
1546 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1547 	ulong desc_addr;
1548 	int ret;
1549 	u16 dummy;
1550 	u32 base3 = 0;
1551 
1552 	memset(&seg_desc, 0, sizeof(seg_desc));
1553 
1554 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1555 		/* set real mode segment descriptor (keep limit etc. for
1556 		 * unreal mode) */
1557 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1558 		set_desc_base(&seg_desc, selector << 4);
1559 		goto load;
1560 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1561 		/* VM86 needs a clean new segment descriptor */
1562 		set_desc_base(&seg_desc, selector << 4);
1563 		set_desc_limit(&seg_desc, 0xffff);
1564 		seg_desc.type = 3;
1565 		seg_desc.p = 1;
1566 		seg_desc.s = 1;
1567 		seg_desc.dpl = 3;
1568 		goto load;
1569 	}
1570 
1571 	rpl = selector & 3;
1572 
1573 	/* TR should be in GDT only */
1574 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1575 		goto exception;
1576 
1577 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1578 	if (null_selector) {
1579 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1580 			goto exception;
1581 
1582 		if (seg == VCPU_SREG_SS) {
1583 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1584 				goto exception;
1585 
1586 			/*
1587 			 * ctxt->ops->set_segment expects the CPL to be in
1588 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1589 			 */
1590 			seg_desc.type = 3;
1591 			seg_desc.p = 1;
1592 			seg_desc.s = 1;
1593 			seg_desc.dpl = cpl;
1594 			seg_desc.d = 1;
1595 			seg_desc.g = 1;
1596 		}
1597 
1598 		/* Skip all following checks */
1599 		goto load;
1600 	}
1601 
1602 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1603 	if (ret != X86EMUL_CONTINUE)
1604 		return ret;
1605 
1606 	err_code = selector & 0xfffc;
1607 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1608 							   GP_VECTOR;
1609 
1610 	/* can't load system descriptor into segment selector */
1611 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1612 		if (transfer == X86_TRANSFER_CALL_JMP)
1613 			return X86EMUL_UNHANDLEABLE;
1614 		goto exception;
1615 	}
1616 
1617 	if (!seg_desc.p) {
1618 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1619 		goto exception;
1620 	}
1621 
1622 	dpl = seg_desc.dpl;
1623 
1624 	switch (seg) {
1625 	case VCPU_SREG_SS:
1626 		/*
1627 		 * segment is not a writable data segment or segment
1628 		 * selector's RPL != CPL or segment selector's RPL != CPL
1629 		 */
1630 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1631 			goto exception;
1632 		break;
1633 	case VCPU_SREG_CS:
1634 		if (!(seg_desc.type & 8))
1635 			goto exception;
1636 
1637 		if (seg_desc.type & 4) {
1638 			/* conforming */
1639 			if (dpl > cpl)
1640 				goto exception;
1641 		} else {
1642 			/* nonconforming */
1643 			if (rpl > cpl || dpl != cpl)
1644 				goto exception;
1645 		}
1646 		/* in long-mode d/b must be clear if l is set */
1647 		if (seg_desc.d && seg_desc.l) {
1648 			u64 efer = 0;
1649 
1650 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1651 			if (efer & EFER_LMA)
1652 				goto exception;
1653 		}
1654 
1655 		/* CS(RPL) <- CPL */
1656 		selector = (selector & 0xfffc) | cpl;
1657 		break;
1658 	case VCPU_SREG_TR:
1659 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1660 			goto exception;
1661 		old_desc = seg_desc;
1662 		seg_desc.type |= 2; /* busy */
1663 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1664 						  sizeof(seg_desc), &ctxt->exception);
1665 		if (ret != X86EMUL_CONTINUE)
1666 			return ret;
1667 		break;
1668 	case VCPU_SREG_LDTR:
1669 		if (seg_desc.s || seg_desc.type != 2)
1670 			goto exception;
1671 		break;
1672 	default: /*  DS, ES, FS, or GS */
1673 		/*
1674 		 * segment is not a data or readable code segment or
1675 		 * ((segment is a data or nonconforming code segment)
1676 		 * and (both RPL and CPL > DPL))
1677 		 */
1678 		if ((seg_desc.type & 0xa) == 0x8 ||
1679 		    (((seg_desc.type & 0xc) != 0xc) &&
1680 		     (rpl > dpl && cpl > dpl)))
1681 			goto exception;
1682 		break;
1683 	}
1684 
1685 	if (seg_desc.s) {
1686 		/* mark segment as accessed */
1687 		if (!(seg_desc.type & 1)) {
1688 			seg_desc.type |= 1;
1689 			ret = write_segment_descriptor(ctxt, selector,
1690 						       &seg_desc);
1691 			if (ret != X86EMUL_CONTINUE)
1692 				return ret;
1693 		}
1694 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1695 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1696 		if (ret != X86EMUL_CONTINUE)
1697 			return ret;
1698 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1699 				((u64)base3 << 32), ctxt))
1700 			return emulate_gp(ctxt, 0);
1701 	}
1702 load:
1703 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1704 	if (desc)
1705 		*desc = seg_desc;
1706 	return X86EMUL_CONTINUE;
1707 exception:
1708 	return emulate_exception(ctxt, err_vec, err_code, true);
1709 }
1710 
1711 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1712 				   u16 selector, int seg)
1713 {
1714 	u8 cpl = ctxt->ops->cpl(ctxt);
1715 
1716 	/*
1717 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1718 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1719 	 * but it's wrong).
1720 	 *
1721 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1722 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1723 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1724 	 * and only forbid it here.
1725 	 */
1726 	if (seg == VCPU_SREG_SS && selector == 3 &&
1727 	    ctxt->mode == X86EMUL_MODE_PROT64)
1728 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1729 
1730 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1731 					 X86_TRANSFER_NONE, NULL);
1732 }
1733 
1734 static void write_register_operand(struct operand *op)
1735 {
1736 	return assign_register(op->addr.reg, op->val, op->bytes);
1737 }
1738 
1739 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1740 {
1741 	switch (op->type) {
1742 	case OP_REG:
1743 		write_register_operand(op);
1744 		break;
1745 	case OP_MEM:
1746 		if (ctxt->lock_prefix)
1747 			return segmented_cmpxchg(ctxt,
1748 						 op->addr.mem,
1749 						 &op->orig_val,
1750 						 &op->val,
1751 						 op->bytes);
1752 		else
1753 			return segmented_write(ctxt,
1754 					       op->addr.mem,
1755 					       &op->val,
1756 					       op->bytes);
1757 		break;
1758 	case OP_MEM_STR:
1759 		return segmented_write(ctxt,
1760 				       op->addr.mem,
1761 				       op->data,
1762 				       op->bytes * op->count);
1763 		break;
1764 	case OP_XMM:
1765 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1766 		break;
1767 	case OP_MM:
1768 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1769 		break;
1770 	case OP_NONE:
1771 		/* no writeback */
1772 		break;
1773 	default:
1774 		break;
1775 	}
1776 	return X86EMUL_CONTINUE;
1777 }
1778 
1779 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1780 {
1781 	struct segmented_address addr;
1782 
1783 	rsp_increment(ctxt, -bytes);
1784 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1785 	addr.seg = VCPU_SREG_SS;
1786 
1787 	return segmented_write(ctxt, addr, data, bytes);
1788 }
1789 
1790 static int em_push(struct x86_emulate_ctxt *ctxt)
1791 {
1792 	/* Disable writeback. */
1793 	ctxt->dst.type = OP_NONE;
1794 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1795 }
1796 
1797 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1798 		       void *dest, int len)
1799 {
1800 	int rc;
1801 	struct segmented_address addr;
1802 
1803 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1804 	addr.seg = VCPU_SREG_SS;
1805 	rc = segmented_read(ctxt, addr, dest, len);
1806 	if (rc != X86EMUL_CONTINUE)
1807 		return rc;
1808 
1809 	rsp_increment(ctxt, len);
1810 	return rc;
1811 }
1812 
1813 static int em_pop(struct x86_emulate_ctxt *ctxt)
1814 {
1815 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1816 }
1817 
1818 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1819 			void *dest, int len)
1820 {
1821 	int rc;
1822 	unsigned long val, change_mask;
1823 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1824 	int cpl = ctxt->ops->cpl(ctxt);
1825 
1826 	rc = emulate_pop(ctxt, &val, len);
1827 	if (rc != X86EMUL_CONTINUE)
1828 		return rc;
1829 
1830 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1831 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1832 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1833 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1834 
1835 	switch(ctxt->mode) {
1836 	case X86EMUL_MODE_PROT64:
1837 	case X86EMUL_MODE_PROT32:
1838 	case X86EMUL_MODE_PROT16:
1839 		if (cpl == 0)
1840 			change_mask |= X86_EFLAGS_IOPL;
1841 		if (cpl <= iopl)
1842 			change_mask |= X86_EFLAGS_IF;
1843 		break;
1844 	case X86EMUL_MODE_VM86:
1845 		if (iopl < 3)
1846 			return emulate_gp(ctxt, 0);
1847 		change_mask |= X86_EFLAGS_IF;
1848 		break;
1849 	default: /* real mode */
1850 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1851 		break;
1852 	}
1853 
1854 	*(unsigned long *)dest =
1855 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1856 
1857 	return rc;
1858 }
1859 
1860 static int em_popf(struct x86_emulate_ctxt *ctxt)
1861 {
1862 	ctxt->dst.type = OP_REG;
1863 	ctxt->dst.addr.reg = &ctxt->eflags;
1864 	ctxt->dst.bytes = ctxt->op_bytes;
1865 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1866 }
1867 
1868 static int em_enter(struct x86_emulate_ctxt *ctxt)
1869 {
1870 	int rc;
1871 	unsigned frame_size = ctxt->src.val;
1872 	unsigned nesting_level = ctxt->src2.val & 31;
1873 	ulong rbp;
1874 
1875 	if (nesting_level)
1876 		return X86EMUL_UNHANDLEABLE;
1877 
1878 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1879 	rc = push(ctxt, &rbp, stack_size(ctxt));
1880 	if (rc != X86EMUL_CONTINUE)
1881 		return rc;
1882 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1883 		      stack_mask(ctxt));
1884 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1885 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1886 		      stack_mask(ctxt));
1887 	return X86EMUL_CONTINUE;
1888 }
1889 
1890 static int em_leave(struct x86_emulate_ctxt *ctxt)
1891 {
1892 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1893 		      stack_mask(ctxt));
1894 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1895 }
1896 
1897 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1898 {
1899 	int seg = ctxt->src2.val;
1900 
1901 	ctxt->src.val = get_segment_selector(ctxt, seg);
1902 	if (ctxt->op_bytes == 4) {
1903 		rsp_increment(ctxt, -2);
1904 		ctxt->op_bytes = 2;
1905 	}
1906 
1907 	return em_push(ctxt);
1908 }
1909 
1910 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1911 {
1912 	int seg = ctxt->src2.val;
1913 	unsigned long selector;
1914 	int rc;
1915 
1916 	rc = emulate_pop(ctxt, &selector, 2);
1917 	if (rc != X86EMUL_CONTINUE)
1918 		return rc;
1919 
1920 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1921 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1922 	if (ctxt->op_bytes > 2)
1923 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1924 
1925 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1926 	return rc;
1927 }
1928 
1929 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1930 {
1931 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1932 	int rc = X86EMUL_CONTINUE;
1933 	int reg = VCPU_REGS_RAX;
1934 
1935 	while (reg <= VCPU_REGS_RDI) {
1936 		(reg == VCPU_REGS_RSP) ?
1937 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1938 
1939 		rc = em_push(ctxt);
1940 		if (rc != X86EMUL_CONTINUE)
1941 			return rc;
1942 
1943 		++reg;
1944 	}
1945 
1946 	return rc;
1947 }
1948 
1949 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1950 {
1951 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1952 	return em_push(ctxt);
1953 }
1954 
1955 static int em_popa(struct x86_emulate_ctxt *ctxt)
1956 {
1957 	int rc = X86EMUL_CONTINUE;
1958 	int reg = VCPU_REGS_RDI;
1959 	u32 val;
1960 
1961 	while (reg >= VCPU_REGS_RAX) {
1962 		if (reg == VCPU_REGS_RSP) {
1963 			rsp_increment(ctxt, ctxt->op_bytes);
1964 			--reg;
1965 		}
1966 
1967 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
1968 		if (rc != X86EMUL_CONTINUE)
1969 			break;
1970 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
1971 		--reg;
1972 	}
1973 	return rc;
1974 }
1975 
1976 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
1977 {
1978 	const struct x86_emulate_ops *ops = ctxt->ops;
1979 	int rc;
1980 	struct desc_ptr dt;
1981 	gva_t cs_addr;
1982 	gva_t eip_addr;
1983 	u16 cs, eip;
1984 
1985 	/* TODO: Add limit checks */
1986 	ctxt->src.val = ctxt->eflags;
1987 	rc = em_push(ctxt);
1988 	if (rc != X86EMUL_CONTINUE)
1989 		return rc;
1990 
1991 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
1992 
1993 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
1994 	rc = em_push(ctxt);
1995 	if (rc != X86EMUL_CONTINUE)
1996 		return rc;
1997 
1998 	ctxt->src.val = ctxt->_eip;
1999 	rc = em_push(ctxt);
2000 	if (rc != X86EMUL_CONTINUE)
2001 		return rc;
2002 
2003 	ops->get_idt(ctxt, &dt);
2004 
2005 	eip_addr = dt.address + (irq << 2);
2006 	cs_addr = dt.address + (irq << 2) + 2;
2007 
2008 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2009 	if (rc != X86EMUL_CONTINUE)
2010 		return rc;
2011 
2012 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2013 	if (rc != X86EMUL_CONTINUE)
2014 		return rc;
2015 
2016 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2017 	if (rc != X86EMUL_CONTINUE)
2018 		return rc;
2019 
2020 	ctxt->_eip = eip;
2021 
2022 	return rc;
2023 }
2024 
2025 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2026 {
2027 	int rc;
2028 
2029 	invalidate_registers(ctxt);
2030 	rc = __emulate_int_real(ctxt, irq);
2031 	if (rc == X86EMUL_CONTINUE)
2032 		writeback_registers(ctxt);
2033 	return rc;
2034 }
2035 
2036 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2037 {
2038 	switch(ctxt->mode) {
2039 	case X86EMUL_MODE_REAL:
2040 		return __emulate_int_real(ctxt, irq);
2041 	case X86EMUL_MODE_VM86:
2042 	case X86EMUL_MODE_PROT16:
2043 	case X86EMUL_MODE_PROT32:
2044 	case X86EMUL_MODE_PROT64:
2045 	default:
2046 		/* Protected mode interrupts unimplemented yet */
2047 		return X86EMUL_UNHANDLEABLE;
2048 	}
2049 }
2050 
2051 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2052 {
2053 	int rc = X86EMUL_CONTINUE;
2054 	unsigned long temp_eip = 0;
2055 	unsigned long temp_eflags = 0;
2056 	unsigned long cs = 0;
2057 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2058 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2059 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2060 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2061 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2062 			     X86_EFLAGS_FIXED;
2063 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2064 				  X86_EFLAGS_VIP;
2065 
2066 	/* TODO: Add stack limit check */
2067 
2068 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2069 
2070 	if (rc != X86EMUL_CONTINUE)
2071 		return rc;
2072 
2073 	if (temp_eip & ~0xffff)
2074 		return emulate_gp(ctxt, 0);
2075 
2076 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2077 
2078 	if (rc != X86EMUL_CONTINUE)
2079 		return rc;
2080 
2081 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2082 
2083 	if (rc != X86EMUL_CONTINUE)
2084 		return rc;
2085 
2086 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2087 
2088 	if (rc != X86EMUL_CONTINUE)
2089 		return rc;
2090 
2091 	ctxt->_eip = temp_eip;
2092 
2093 	if (ctxt->op_bytes == 4)
2094 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2095 	else if (ctxt->op_bytes == 2) {
2096 		ctxt->eflags &= ~0xffff;
2097 		ctxt->eflags |= temp_eflags;
2098 	}
2099 
2100 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2101 	ctxt->eflags |= X86_EFLAGS_FIXED;
2102 	ctxt->ops->set_nmi_mask(ctxt, false);
2103 
2104 	return rc;
2105 }
2106 
2107 static int em_iret(struct x86_emulate_ctxt *ctxt)
2108 {
2109 	switch(ctxt->mode) {
2110 	case X86EMUL_MODE_REAL:
2111 		return emulate_iret_real(ctxt);
2112 	case X86EMUL_MODE_VM86:
2113 	case X86EMUL_MODE_PROT16:
2114 	case X86EMUL_MODE_PROT32:
2115 	case X86EMUL_MODE_PROT64:
2116 	default:
2117 		/* iret from protected mode unimplemented yet */
2118 		return X86EMUL_UNHANDLEABLE;
2119 	}
2120 }
2121 
2122 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2123 {
2124 	int rc;
2125 	unsigned short sel;
2126 	struct desc_struct new_desc;
2127 	u8 cpl = ctxt->ops->cpl(ctxt);
2128 
2129 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2130 
2131 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2132 				       X86_TRANSFER_CALL_JMP,
2133 				       &new_desc);
2134 	if (rc != X86EMUL_CONTINUE)
2135 		return rc;
2136 
2137 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2138 	/* Error handling is not implemented. */
2139 	if (rc != X86EMUL_CONTINUE)
2140 		return X86EMUL_UNHANDLEABLE;
2141 
2142 	return rc;
2143 }
2144 
2145 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2146 {
2147 	return assign_eip_near(ctxt, ctxt->src.val);
2148 }
2149 
2150 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2151 {
2152 	int rc;
2153 	long int old_eip;
2154 
2155 	old_eip = ctxt->_eip;
2156 	rc = assign_eip_near(ctxt, ctxt->src.val);
2157 	if (rc != X86EMUL_CONTINUE)
2158 		return rc;
2159 	ctxt->src.val = old_eip;
2160 	rc = em_push(ctxt);
2161 	return rc;
2162 }
2163 
2164 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2165 {
2166 	u64 old = ctxt->dst.orig_val64;
2167 
2168 	if (ctxt->dst.bytes == 16)
2169 		return X86EMUL_UNHANDLEABLE;
2170 
2171 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2172 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2173 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2174 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2175 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2176 	} else {
2177 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2178 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2179 
2180 		ctxt->eflags |= X86_EFLAGS_ZF;
2181 	}
2182 	return X86EMUL_CONTINUE;
2183 }
2184 
2185 static int em_ret(struct x86_emulate_ctxt *ctxt)
2186 {
2187 	int rc;
2188 	unsigned long eip;
2189 
2190 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2191 	if (rc != X86EMUL_CONTINUE)
2192 		return rc;
2193 
2194 	return assign_eip_near(ctxt, eip);
2195 }
2196 
2197 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2198 {
2199 	int rc;
2200 	unsigned long eip, cs;
2201 	int cpl = ctxt->ops->cpl(ctxt);
2202 	struct desc_struct new_desc;
2203 
2204 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2205 	if (rc != X86EMUL_CONTINUE)
2206 		return rc;
2207 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2208 	if (rc != X86EMUL_CONTINUE)
2209 		return rc;
2210 	/* Outer-privilege level return is not implemented */
2211 	if (ctxt->mode >= X86EMUL_MODE_PROT16 && (cs & 3) > cpl)
2212 		return X86EMUL_UNHANDLEABLE;
2213 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2214 				       X86_TRANSFER_RET,
2215 				       &new_desc);
2216 	if (rc != X86EMUL_CONTINUE)
2217 		return rc;
2218 	rc = assign_eip_far(ctxt, eip, &new_desc);
2219 	/* Error handling is not implemented. */
2220 	if (rc != X86EMUL_CONTINUE)
2221 		return X86EMUL_UNHANDLEABLE;
2222 
2223 	return rc;
2224 }
2225 
2226 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2227 {
2228         int rc;
2229 
2230         rc = em_ret_far(ctxt);
2231         if (rc != X86EMUL_CONTINUE)
2232                 return rc;
2233         rsp_increment(ctxt, ctxt->src.val);
2234         return X86EMUL_CONTINUE;
2235 }
2236 
2237 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2238 {
2239 	/* Save real source value, then compare EAX against destination. */
2240 	ctxt->dst.orig_val = ctxt->dst.val;
2241 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2242 	ctxt->src.orig_val = ctxt->src.val;
2243 	ctxt->src.val = ctxt->dst.orig_val;
2244 	fastop(ctxt, em_cmp);
2245 
2246 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2247 		/* Success: write back to memory; no update of EAX */
2248 		ctxt->src.type = OP_NONE;
2249 		ctxt->dst.val = ctxt->src.orig_val;
2250 	} else {
2251 		/* Failure: write the value we saw to EAX. */
2252 		ctxt->src.type = OP_REG;
2253 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2254 		ctxt->src.val = ctxt->dst.orig_val;
2255 		/* Create write-cycle to dest by writing the same value */
2256 		ctxt->dst.val = ctxt->dst.orig_val;
2257 	}
2258 	return X86EMUL_CONTINUE;
2259 }
2260 
2261 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2262 {
2263 	int seg = ctxt->src2.val;
2264 	unsigned short sel;
2265 	int rc;
2266 
2267 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2268 
2269 	rc = load_segment_descriptor(ctxt, sel, seg);
2270 	if (rc != X86EMUL_CONTINUE)
2271 		return rc;
2272 
2273 	ctxt->dst.val = ctxt->src.val;
2274 	return rc;
2275 }
2276 
2277 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2278 {
2279 #ifdef CONFIG_X86_64
2280 	return ctxt->ops->guest_has_long_mode(ctxt);
2281 #else
2282 	return false;
2283 #endif
2284 }
2285 
2286 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2287 {
2288 	desc->g    = (flags >> 23) & 1;
2289 	desc->d    = (flags >> 22) & 1;
2290 	desc->l    = (flags >> 21) & 1;
2291 	desc->avl  = (flags >> 20) & 1;
2292 	desc->p    = (flags >> 15) & 1;
2293 	desc->dpl  = (flags >> 13) & 3;
2294 	desc->s    = (flags >> 12) & 1;
2295 	desc->type = (flags >>  8) & 15;
2296 }
2297 
2298 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, const char *smstate,
2299 			   int n)
2300 {
2301 	struct desc_struct desc;
2302 	int offset;
2303 	u16 selector;
2304 
2305 	selector = GET_SMSTATE(u32, smstate, 0x7fa8 + n * 4);
2306 
2307 	if (n < 3)
2308 		offset = 0x7f84 + n * 12;
2309 	else
2310 		offset = 0x7f2c + (n - 3) * 12;
2311 
2312 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2313 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2314 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, offset));
2315 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2316 	return X86EMUL_CONTINUE;
2317 }
2318 
2319 #ifdef CONFIG_X86_64
2320 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, const char *smstate,
2321 			   int n)
2322 {
2323 	struct desc_struct desc;
2324 	int offset;
2325 	u16 selector;
2326 	u32 base3;
2327 
2328 	offset = 0x7e00 + n * 16;
2329 
2330 	selector =                GET_SMSTATE(u16, smstate, offset);
2331 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smstate, offset + 2) << 8);
2332 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2333 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2334 	base3 =                   GET_SMSTATE(u32, smstate, offset + 12);
2335 
2336 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2337 	return X86EMUL_CONTINUE;
2338 }
2339 #endif
2340 
2341 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2342 				    u64 cr0, u64 cr3, u64 cr4)
2343 {
2344 	int bad;
2345 	u64 pcid;
2346 
2347 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
2348 	pcid = 0;
2349 	if (cr4 & X86_CR4_PCIDE) {
2350 		pcid = cr3 & 0xfff;
2351 		cr3 &= ~0xfff;
2352 	}
2353 
2354 	bad = ctxt->ops->set_cr(ctxt, 3, cr3);
2355 	if (bad)
2356 		return X86EMUL_UNHANDLEABLE;
2357 
2358 	/*
2359 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2360 	 * Then enable protected mode.	However, PCID cannot be enabled
2361 	 * if EFER.LMA=0, so set it separately.
2362 	 */
2363 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2364 	if (bad)
2365 		return X86EMUL_UNHANDLEABLE;
2366 
2367 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2368 	if (bad)
2369 		return X86EMUL_UNHANDLEABLE;
2370 
2371 	if (cr4 & X86_CR4_PCIDE) {
2372 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2373 		if (bad)
2374 			return X86EMUL_UNHANDLEABLE;
2375 		if (pcid) {
2376 			bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid);
2377 			if (bad)
2378 				return X86EMUL_UNHANDLEABLE;
2379 		}
2380 
2381 	}
2382 
2383 	return X86EMUL_CONTINUE;
2384 }
2385 
2386 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
2387 			     const char *smstate)
2388 {
2389 	struct desc_struct desc;
2390 	struct desc_ptr dt;
2391 	u16 selector;
2392 	u32 val, cr0, cr3, cr4;
2393 	int i;
2394 
2395 	cr0 =                      GET_SMSTATE(u32, smstate, 0x7ffc);
2396 	cr3 =                      GET_SMSTATE(u32, smstate, 0x7ff8);
2397 	ctxt->eflags =             GET_SMSTATE(u32, smstate, 0x7ff4) | X86_EFLAGS_FIXED;
2398 	ctxt->_eip =               GET_SMSTATE(u32, smstate, 0x7ff0);
2399 
2400 	for (i = 0; i < 8; i++)
2401 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smstate, 0x7fd0 + i * 4);
2402 
2403 	val = GET_SMSTATE(u32, smstate, 0x7fcc);
2404 
2405 	if (ctxt->ops->set_dr(ctxt, 6, val))
2406 		return X86EMUL_UNHANDLEABLE;
2407 
2408 	val = GET_SMSTATE(u32, smstate, 0x7fc8);
2409 
2410 	if (ctxt->ops->set_dr(ctxt, 7, val))
2411 		return X86EMUL_UNHANDLEABLE;
2412 
2413 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc4);
2414 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f64));
2415 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f60));
2416 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f5c));
2417 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2418 
2419 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc0);
2420 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f80));
2421 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f7c));
2422 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f78));
2423 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2424 
2425 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f74);
2426 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f70);
2427 	ctxt->ops->set_gdt(ctxt, &dt);
2428 
2429 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f58);
2430 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f54);
2431 	ctxt->ops->set_idt(ctxt, &dt);
2432 
2433 	for (i = 0; i < 6; i++) {
2434 		int r = rsm_load_seg_32(ctxt, smstate, i);
2435 		if (r != X86EMUL_CONTINUE)
2436 			return r;
2437 	}
2438 
2439 	cr4 = GET_SMSTATE(u32, smstate, 0x7f14);
2440 
2441 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7ef8));
2442 
2443 	return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2444 }
2445 
2446 #ifdef CONFIG_X86_64
2447 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
2448 			     const char *smstate)
2449 {
2450 	struct desc_struct desc;
2451 	struct desc_ptr dt;
2452 	u64 val, cr0, cr3, cr4;
2453 	u32 base3;
2454 	u16 selector;
2455 	int i, r;
2456 
2457 	for (i = 0; i < 16; i++)
2458 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smstate, 0x7ff8 - i * 8);
2459 
2460 	ctxt->_eip   = GET_SMSTATE(u64, smstate, 0x7f78);
2461 	ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7f70) | X86_EFLAGS_FIXED;
2462 
2463 	val = GET_SMSTATE(u64, smstate, 0x7f68);
2464 
2465 	if (ctxt->ops->set_dr(ctxt, 6, val))
2466 		return X86EMUL_UNHANDLEABLE;
2467 
2468 	val = GET_SMSTATE(u64, smstate, 0x7f60);
2469 
2470 	if (ctxt->ops->set_dr(ctxt, 7, val))
2471 		return X86EMUL_UNHANDLEABLE;
2472 
2473 	cr0 =                       GET_SMSTATE(u64, smstate, 0x7f58);
2474 	cr3 =                       GET_SMSTATE(u64, smstate, 0x7f50);
2475 	cr4 =                       GET_SMSTATE(u64, smstate, 0x7f48);
2476 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7f00));
2477 	val =                       GET_SMSTATE(u64, smstate, 0x7ed0);
2478 
2479 	if (ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA))
2480 		return X86EMUL_UNHANDLEABLE;
2481 
2482 	selector =                  GET_SMSTATE(u32, smstate, 0x7e90);
2483 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e92) << 8);
2484 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e94));
2485 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e98));
2486 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e9c);
2487 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2488 
2489 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e84);
2490 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e88);
2491 	ctxt->ops->set_idt(ctxt, &dt);
2492 
2493 	selector =                  GET_SMSTATE(u32, smstate, 0x7e70);
2494 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e72) << 8);
2495 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e74));
2496 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e78));
2497 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e7c);
2498 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2499 
2500 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e64);
2501 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e68);
2502 	ctxt->ops->set_gdt(ctxt, &dt);
2503 
2504 	r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2505 	if (r != X86EMUL_CONTINUE)
2506 		return r;
2507 
2508 	for (i = 0; i < 6; i++) {
2509 		r = rsm_load_seg_64(ctxt, smstate, i);
2510 		if (r != X86EMUL_CONTINUE)
2511 			return r;
2512 	}
2513 
2514 	return X86EMUL_CONTINUE;
2515 }
2516 #endif
2517 
2518 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2519 {
2520 	unsigned long cr0, cr4, efer;
2521 	char buf[512];
2522 	u64 smbase;
2523 	int ret;
2524 
2525 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2526 		return emulate_ud(ctxt);
2527 
2528 	smbase = ctxt->ops->get_smbase(ctxt);
2529 
2530 	ret = ctxt->ops->read_phys(ctxt, smbase + 0xfe00, buf, sizeof(buf));
2531 	if (ret != X86EMUL_CONTINUE)
2532 		return X86EMUL_UNHANDLEABLE;
2533 
2534 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2535 		ctxt->ops->set_nmi_mask(ctxt, false);
2536 
2537 	ctxt->ops->exiting_smm(ctxt);
2538 
2539 	/*
2540 	 * Get back to real mode, to prepare a safe state in which to load
2541 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2542 	 * supports long mode.
2543 	 */
2544 	if (emulator_has_longmode(ctxt)) {
2545 		struct desc_struct cs_desc;
2546 
2547 		/* Zero CR4.PCIDE before CR0.PG.  */
2548 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2549 		if (cr4 & X86_CR4_PCIDE)
2550 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2551 
2552 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2553 		memset(&cs_desc, 0, sizeof(cs_desc));
2554 		cs_desc.type = 0xb;
2555 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2556 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2557 	}
2558 
2559 	/* For the 64-bit case, this will clear EFER.LMA.  */
2560 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2561 	if (cr0 & X86_CR0_PE)
2562 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2563 
2564 	if (emulator_has_longmode(ctxt)) {
2565 		/* Clear CR4.PAE before clearing EFER.LME. */
2566 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2567 		if (cr4 & X86_CR4_PAE)
2568 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2569 
2570 		/* And finally go back to 32-bit mode.  */
2571 		efer = 0;
2572 		ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2573 	}
2574 
2575 	/*
2576 	 * Give leave_smm() a chance to make ISA-specific changes to the vCPU
2577 	 * state (e.g. enter guest mode) before loading state from the SMM
2578 	 * state-save area.
2579 	 */
2580 	if (ctxt->ops->leave_smm(ctxt, buf))
2581 		goto emulate_shutdown;
2582 
2583 #ifdef CONFIG_X86_64
2584 	if (emulator_has_longmode(ctxt))
2585 		ret = rsm_load_state_64(ctxt, buf);
2586 	else
2587 #endif
2588 		ret = rsm_load_state_32(ctxt, buf);
2589 
2590 	if (ret != X86EMUL_CONTINUE)
2591 		goto emulate_shutdown;
2592 
2593 	/*
2594 	 * Note, the ctxt->ops callbacks are responsible for handling side
2595 	 * effects when writing MSRs and CRs, e.g. MMU context resets, CPUID
2596 	 * runtime updates, etc...  If that changes, e.g. this flow is moved
2597 	 * out of the emulator to make it look more like enter_smm(), then
2598 	 * those side effects need to be explicitly handled for both success
2599 	 * and shutdown.
2600 	 */
2601 	return X86EMUL_CONTINUE;
2602 
2603 emulate_shutdown:
2604 	ctxt->ops->triple_fault(ctxt);
2605 	return X86EMUL_CONTINUE;
2606 }
2607 
2608 static void
2609 setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
2610 			struct desc_struct *cs, struct desc_struct *ss)
2611 {
2612 	cs->l = 0;		/* will be adjusted later */
2613 	set_desc_base(cs, 0);	/* flat segment */
2614 	cs->g = 1;		/* 4kb granularity */
2615 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2616 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2617 	cs->s = 1;
2618 	cs->dpl = 0;		/* will be adjusted later */
2619 	cs->p = 1;
2620 	cs->d = 1;
2621 	cs->avl = 0;
2622 
2623 	set_desc_base(ss, 0);	/* flat segment */
2624 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2625 	ss->g = 1;		/* 4kb granularity */
2626 	ss->s = 1;
2627 	ss->type = 0x03;	/* Read/Write, Accessed */
2628 	ss->d = 1;		/* 32bit stack segment */
2629 	ss->dpl = 0;
2630 	ss->p = 1;
2631 	ss->l = 0;
2632 	ss->avl = 0;
2633 }
2634 
2635 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2636 {
2637 	u32 eax, ebx, ecx, edx;
2638 
2639 	eax = ecx = 0;
2640 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2641 	return is_guest_vendor_intel(ebx, ecx, edx);
2642 }
2643 
2644 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2645 {
2646 	const struct x86_emulate_ops *ops = ctxt->ops;
2647 	u32 eax, ebx, ecx, edx;
2648 
2649 	/*
2650 	 * syscall should always be enabled in longmode - so only become
2651 	 * vendor specific (cpuid) if other modes are active...
2652 	 */
2653 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2654 		return true;
2655 
2656 	eax = 0x00000000;
2657 	ecx = 0x00000000;
2658 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2659 	/*
2660 	 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a
2661 	 * 64bit guest with a 32bit compat-app running will #UD !! While this
2662 	 * behaviour can be fixed (by emulating) into AMD response - CPUs of
2663 	 * AMD can't behave like Intel.
2664 	 */
2665 	if (is_guest_vendor_intel(ebx, ecx, edx))
2666 		return false;
2667 
2668 	if (is_guest_vendor_amd(ebx, ecx, edx) ||
2669 	    is_guest_vendor_hygon(ebx, ecx, edx))
2670 		return true;
2671 
2672 	/*
2673 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2674 	 * stricter rules...
2675 	 */
2676 	return false;
2677 }
2678 
2679 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2680 {
2681 	const struct x86_emulate_ops *ops = ctxt->ops;
2682 	struct desc_struct cs, ss;
2683 	u64 msr_data;
2684 	u16 cs_sel, ss_sel;
2685 	u64 efer = 0;
2686 
2687 	/* syscall is not available in real mode */
2688 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2689 	    ctxt->mode == X86EMUL_MODE_VM86)
2690 		return emulate_ud(ctxt);
2691 
2692 	if (!(em_syscall_is_enabled(ctxt)))
2693 		return emulate_ud(ctxt);
2694 
2695 	ops->get_msr(ctxt, MSR_EFER, &efer);
2696 	if (!(efer & EFER_SCE))
2697 		return emulate_ud(ctxt);
2698 
2699 	setup_syscalls_segments(ctxt, &cs, &ss);
2700 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2701 	msr_data >>= 32;
2702 	cs_sel = (u16)(msr_data & 0xfffc);
2703 	ss_sel = (u16)(msr_data + 8);
2704 
2705 	if (efer & EFER_LMA) {
2706 		cs.d = 0;
2707 		cs.l = 1;
2708 	}
2709 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2710 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2711 
2712 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2713 	if (efer & EFER_LMA) {
2714 #ifdef CONFIG_X86_64
2715 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2716 
2717 		ops->get_msr(ctxt,
2718 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2719 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2720 		ctxt->_eip = msr_data;
2721 
2722 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2723 		ctxt->eflags &= ~msr_data;
2724 		ctxt->eflags |= X86_EFLAGS_FIXED;
2725 #endif
2726 	} else {
2727 		/* legacy mode */
2728 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2729 		ctxt->_eip = (u32)msr_data;
2730 
2731 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2732 	}
2733 
2734 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2735 	return X86EMUL_CONTINUE;
2736 }
2737 
2738 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2739 {
2740 	const struct x86_emulate_ops *ops = ctxt->ops;
2741 	struct desc_struct cs, ss;
2742 	u64 msr_data;
2743 	u16 cs_sel, ss_sel;
2744 	u64 efer = 0;
2745 
2746 	ops->get_msr(ctxt, MSR_EFER, &efer);
2747 	/* inject #GP if in real mode */
2748 	if (ctxt->mode == X86EMUL_MODE_REAL)
2749 		return emulate_gp(ctxt, 0);
2750 
2751 	/*
2752 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2753 	 * mode).
2754 	 */
2755 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2756 	    && !vendor_intel(ctxt))
2757 		return emulate_ud(ctxt);
2758 
2759 	/* sysenter/sysexit have not been tested in 64bit mode. */
2760 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2761 		return X86EMUL_UNHANDLEABLE;
2762 
2763 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2764 	if ((msr_data & 0xfffc) == 0x0)
2765 		return emulate_gp(ctxt, 0);
2766 
2767 	setup_syscalls_segments(ctxt, &cs, &ss);
2768 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2769 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2770 	ss_sel = cs_sel + 8;
2771 	if (efer & EFER_LMA) {
2772 		cs.d = 0;
2773 		cs.l = 1;
2774 	}
2775 
2776 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2777 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2778 
2779 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2780 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2781 
2782 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2783 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2784 							      (u32)msr_data;
2785 	if (efer & EFER_LMA)
2786 		ctxt->mode = X86EMUL_MODE_PROT64;
2787 
2788 	return X86EMUL_CONTINUE;
2789 }
2790 
2791 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2792 {
2793 	const struct x86_emulate_ops *ops = ctxt->ops;
2794 	struct desc_struct cs, ss;
2795 	u64 msr_data, rcx, rdx;
2796 	int usermode;
2797 	u16 cs_sel = 0, ss_sel = 0;
2798 
2799 	/* inject #GP if in real mode or Virtual 8086 mode */
2800 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2801 	    ctxt->mode == X86EMUL_MODE_VM86)
2802 		return emulate_gp(ctxt, 0);
2803 
2804 	setup_syscalls_segments(ctxt, &cs, &ss);
2805 
2806 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2807 		usermode = X86EMUL_MODE_PROT64;
2808 	else
2809 		usermode = X86EMUL_MODE_PROT32;
2810 
2811 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2812 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2813 
2814 	cs.dpl = 3;
2815 	ss.dpl = 3;
2816 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2817 	switch (usermode) {
2818 	case X86EMUL_MODE_PROT32:
2819 		cs_sel = (u16)(msr_data + 16);
2820 		if ((msr_data & 0xfffc) == 0x0)
2821 			return emulate_gp(ctxt, 0);
2822 		ss_sel = (u16)(msr_data + 24);
2823 		rcx = (u32)rcx;
2824 		rdx = (u32)rdx;
2825 		break;
2826 	case X86EMUL_MODE_PROT64:
2827 		cs_sel = (u16)(msr_data + 32);
2828 		if (msr_data == 0x0)
2829 			return emulate_gp(ctxt, 0);
2830 		ss_sel = cs_sel + 8;
2831 		cs.d = 0;
2832 		cs.l = 1;
2833 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2834 		    emul_is_noncanonical_address(rdx, ctxt))
2835 			return emulate_gp(ctxt, 0);
2836 		break;
2837 	}
2838 	cs_sel |= SEGMENT_RPL_MASK;
2839 	ss_sel |= SEGMENT_RPL_MASK;
2840 
2841 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2842 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2843 
2844 	ctxt->_eip = rdx;
2845 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2846 
2847 	return X86EMUL_CONTINUE;
2848 }
2849 
2850 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2851 {
2852 	int iopl;
2853 	if (ctxt->mode == X86EMUL_MODE_REAL)
2854 		return false;
2855 	if (ctxt->mode == X86EMUL_MODE_VM86)
2856 		return true;
2857 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2858 	return ctxt->ops->cpl(ctxt) > iopl;
2859 }
2860 
2861 #define VMWARE_PORT_VMPORT	(0x5658)
2862 #define VMWARE_PORT_VMRPC	(0x5659)
2863 
2864 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2865 					    u16 port, u16 len)
2866 {
2867 	const struct x86_emulate_ops *ops = ctxt->ops;
2868 	struct desc_struct tr_seg;
2869 	u32 base3;
2870 	int r;
2871 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2872 	unsigned mask = (1 << len) - 1;
2873 	unsigned long base;
2874 
2875 	/*
2876 	 * VMware allows access to these ports even if denied
2877 	 * by TSS I/O permission bitmap. Mimic behavior.
2878 	 */
2879 	if (enable_vmware_backdoor &&
2880 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2881 		return true;
2882 
2883 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2884 	if (!tr_seg.p)
2885 		return false;
2886 	if (desc_limit_scaled(&tr_seg) < 103)
2887 		return false;
2888 	base = get_desc_base(&tr_seg);
2889 #ifdef CONFIG_X86_64
2890 	base |= ((u64)base3) << 32;
2891 #endif
2892 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2893 	if (r != X86EMUL_CONTINUE)
2894 		return false;
2895 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2896 		return false;
2897 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2898 	if (r != X86EMUL_CONTINUE)
2899 		return false;
2900 	if ((perm >> bit_idx) & mask)
2901 		return false;
2902 	return true;
2903 }
2904 
2905 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2906 				 u16 port, u16 len)
2907 {
2908 	if (ctxt->perm_ok)
2909 		return true;
2910 
2911 	if (emulator_bad_iopl(ctxt))
2912 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2913 			return false;
2914 
2915 	ctxt->perm_ok = true;
2916 
2917 	return true;
2918 }
2919 
2920 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2921 {
2922 	/*
2923 	 * Intel CPUs mask the counter and pointers in quite strange
2924 	 * manner when ECX is zero due to REP-string optimizations.
2925 	 */
2926 #ifdef CONFIG_X86_64
2927 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2928 		return;
2929 
2930 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2931 
2932 	switch (ctxt->b) {
2933 	case 0xa4:	/* movsb */
2934 	case 0xa5:	/* movsd/w */
2935 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2936 		fallthrough;
2937 	case 0xaa:	/* stosb */
2938 	case 0xab:	/* stosd/w */
2939 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2940 	}
2941 #endif
2942 }
2943 
2944 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2945 				struct tss_segment_16 *tss)
2946 {
2947 	tss->ip = ctxt->_eip;
2948 	tss->flag = ctxt->eflags;
2949 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2950 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2951 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2952 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2953 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2954 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2955 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2956 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2957 
2958 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2959 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2960 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2961 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2962 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2963 }
2964 
2965 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2966 				 struct tss_segment_16 *tss)
2967 {
2968 	int ret;
2969 	u8 cpl;
2970 
2971 	ctxt->_eip = tss->ip;
2972 	ctxt->eflags = tss->flag | 2;
2973 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2974 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2975 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2976 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2977 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2978 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2979 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2980 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2981 
2982 	/*
2983 	 * SDM says that segment selectors are loaded before segment
2984 	 * descriptors
2985 	 */
2986 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
2987 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2988 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2989 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2990 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2991 
2992 	cpl = tss->cs & 3;
2993 
2994 	/*
2995 	 * Now load segment descriptors. If fault happens at this stage
2996 	 * it is handled in a context of new task
2997 	 */
2998 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
2999 					X86_TRANSFER_TASK_SWITCH, NULL);
3000 	if (ret != X86EMUL_CONTINUE)
3001 		return ret;
3002 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3003 					X86_TRANSFER_TASK_SWITCH, NULL);
3004 	if (ret != X86EMUL_CONTINUE)
3005 		return ret;
3006 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3007 					X86_TRANSFER_TASK_SWITCH, NULL);
3008 	if (ret != X86EMUL_CONTINUE)
3009 		return ret;
3010 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3011 					X86_TRANSFER_TASK_SWITCH, NULL);
3012 	if (ret != X86EMUL_CONTINUE)
3013 		return ret;
3014 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3015 					X86_TRANSFER_TASK_SWITCH, NULL);
3016 	if (ret != X86EMUL_CONTINUE)
3017 		return ret;
3018 
3019 	return X86EMUL_CONTINUE;
3020 }
3021 
3022 static int task_switch_16(struct x86_emulate_ctxt *ctxt,
3023 			  u16 tss_selector, u16 old_tss_sel,
3024 			  ulong old_tss_base, struct desc_struct *new_desc)
3025 {
3026 	struct tss_segment_16 tss_seg;
3027 	int ret;
3028 	u32 new_tss_base = get_desc_base(new_desc);
3029 
3030 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3031 	if (ret != X86EMUL_CONTINUE)
3032 		return ret;
3033 
3034 	save_state_to_tss16(ctxt, &tss_seg);
3035 
3036 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3037 	if (ret != X86EMUL_CONTINUE)
3038 		return ret;
3039 
3040 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3041 	if (ret != X86EMUL_CONTINUE)
3042 		return ret;
3043 
3044 	if (old_tss_sel != 0xffff) {
3045 		tss_seg.prev_task_link = old_tss_sel;
3046 
3047 		ret = linear_write_system(ctxt, new_tss_base,
3048 					  &tss_seg.prev_task_link,
3049 					  sizeof(tss_seg.prev_task_link));
3050 		if (ret != X86EMUL_CONTINUE)
3051 			return ret;
3052 	}
3053 
3054 	return load_state_from_tss16(ctxt, &tss_seg);
3055 }
3056 
3057 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3058 				struct tss_segment_32 *tss)
3059 {
3060 	/* CR3 and ldt selector are not saved intentionally */
3061 	tss->eip = ctxt->_eip;
3062 	tss->eflags = ctxt->eflags;
3063 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3064 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3065 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3066 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3067 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3068 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3069 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3070 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3071 
3072 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3073 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3074 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3075 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3076 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3077 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3078 }
3079 
3080 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3081 				 struct tss_segment_32 *tss)
3082 {
3083 	int ret;
3084 	u8 cpl;
3085 
3086 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3087 		return emulate_gp(ctxt, 0);
3088 	ctxt->_eip = tss->eip;
3089 	ctxt->eflags = tss->eflags | 2;
3090 
3091 	/* General purpose registers */
3092 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3093 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3094 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3095 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3096 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3097 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3098 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3099 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3100 
3101 	/*
3102 	 * SDM says that segment selectors are loaded before segment
3103 	 * descriptors.  This is important because CPL checks will
3104 	 * use CS.RPL.
3105 	 */
3106 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3107 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3108 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3109 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3110 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3111 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3112 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3113 
3114 	/*
3115 	 * If we're switching between Protected Mode and VM86, we need to make
3116 	 * sure to update the mode before loading the segment descriptors so
3117 	 * that the selectors are interpreted correctly.
3118 	 */
3119 	if (ctxt->eflags & X86_EFLAGS_VM) {
3120 		ctxt->mode = X86EMUL_MODE_VM86;
3121 		cpl = 3;
3122 	} else {
3123 		ctxt->mode = X86EMUL_MODE_PROT32;
3124 		cpl = tss->cs & 3;
3125 	}
3126 
3127 	/*
3128 	 * Now load segment descriptors. If fault happens at this stage
3129 	 * it is handled in a context of new task
3130 	 */
3131 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3132 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3133 	if (ret != X86EMUL_CONTINUE)
3134 		return ret;
3135 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3136 					X86_TRANSFER_TASK_SWITCH, NULL);
3137 	if (ret != X86EMUL_CONTINUE)
3138 		return ret;
3139 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3140 					X86_TRANSFER_TASK_SWITCH, NULL);
3141 	if (ret != X86EMUL_CONTINUE)
3142 		return ret;
3143 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3144 					X86_TRANSFER_TASK_SWITCH, NULL);
3145 	if (ret != X86EMUL_CONTINUE)
3146 		return ret;
3147 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3148 					X86_TRANSFER_TASK_SWITCH, NULL);
3149 	if (ret != X86EMUL_CONTINUE)
3150 		return ret;
3151 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3152 					X86_TRANSFER_TASK_SWITCH, NULL);
3153 	if (ret != X86EMUL_CONTINUE)
3154 		return ret;
3155 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3156 					X86_TRANSFER_TASK_SWITCH, NULL);
3157 
3158 	return ret;
3159 }
3160 
3161 static int task_switch_32(struct x86_emulate_ctxt *ctxt,
3162 			  u16 tss_selector, u16 old_tss_sel,
3163 			  ulong old_tss_base, struct desc_struct *new_desc)
3164 {
3165 	struct tss_segment_32 tss_seg;
3166 	int ret;
3167 	u32 new_tss_base = get_desc_base(new_desc);
3168 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3169 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3170 
3171 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3172 	if (ret != X86EMUL_CONTINUE)
3173 		return ret;
3174 
3175 	save_state_to_tss32(ctxt, &tss_seg);
3176 
3177 	/* Only GP registers and segment selectors are saved */
3178 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3179 				  ldt_sel_offset - eip_offset);
3180 	if (ret != X86EMUL_CONTINUE)
3181 		return ret;
3182 
3183 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3184 	if (ret != X86EMUL_CONTINUE)
3185 		return ret;
3186 
3187 	if (old_tss_sel != 0xffff) {
3188 		tss_seg.prev_task_link = old_tss_sel;
3189 
3190 		ret = linear_write_system(ctxt, new_tss_base,
3191 					  &tss_seg.prev_task_link,
3192 					  sizeof(tss_seg.prev_task_link));
3193 		if (ret != X86EMUL_CONTINUE)
3194 			return ret;
3195 	}
3196 
3197 	return load_state_from_tss32(ctxt, &tss_seg);
3198 }
3199 
3200 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3201 				   u16 tss_selector, int idt_index, int reason,
3202 				   bool has_error_code, u32 error_code)
3203 {
3204 	const struct x86_emulate_ops *ops = ctxt->ops;
3205 	struct desc_struct curr_tss_desc, next_tss_desc;
3206 	int ret;
3207 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3208 	ulong old_tss_base =
3209 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3210 	u32 desc_limit;
3211 	ulong desc_addr, dr7;
3212 
3213 	/* FIXME: old_tss_base == ~0 ? */
3214 
3215 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3216 	if (ret != X86EMUL_CONTINUE)
3217 		return ret;
3218 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3219 	if (ret != X86EMUL_CONTINUE)
3220 		return ret;
3221 
3222 	/* FIXME: check that next_tss_desc is tss */
3223 
3224 	/*
3225 	 * Check privileges. The three cases are task switch caused by...
3226 	 *
3227 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3228 	 * 2. Exception/IRQ/iret: No check is performed
3229 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3230 	 *    hardware checks it before exiting.
3231 	 */
3232 	if (reason == TASK_SWITCH_GATE) {
3233 		if (idt_index != -1) {
3234 			/* Software interrupts */
3235 			struct desc_struct task_gate_desc;
3236 			int dpl;
3237 
3238 			ret = read_interrupt_descriptor(ctxt, idt_index,
3239 							&task_gate_desc);
3240 			if (ret != X86EMUL_CONTINUE)
3241 				return ret;
3242 
3243 			dpl = task_gate_desc.dpl;
3244 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3245 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3246 		}
3247 	}
3248 
3249 	desc_limit = desc_limit_scaled(&next_tss_desc);
3250 	if (!next_tss_desc.p ||
3251 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3252 	     desc_limit < 0x2b)) {
3253 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3254 	}
3255 
3256 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3257 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3258 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3259 	}
3260 
3261 	if (reason == TASK_SWITCH_IRET)
3262 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3263 
3264 	/* set back link to prev task only if NT bit is set in eflags
3265 	   note that old_tss_sel is not used after this point */
3266 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3267 		old_tss_sel = 0xffff;
3268 
3269 	if (next_tss_desc.type & 8)
3270 		ret = task_switch_32(ctxt, tss_selector, old_tss_sel,
3271 				     old_tss_base, &next_tss_desc);
3272 	else
3273 		ret = task_switch_16(ctxt, tss_selector, old_tss_sel,
3274 				     old_tss_base, &next_tss_desc);
3275 	if (ret != X86EMUL_CONTINUE)
3276 		return ret;
3277 
3278 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3279 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3280 
3281 	if (reason != TASK_SWITCH_IRET) {
3282 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3283 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3284 	}
3285 
3286 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3287 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3288 
3289 	if (has_error_code) {
3290 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3291 		ctxt->lock_prefix = 0;
3292 		ctxt->src.val = (unsigned long) error_code;
3293 		ret = em_push(ctxt);
3294 	}
3295 
3296 	ops->get_dr(ctxt, 7, &dr7);
3297 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3298 
3299 	return ret;
3300 }
3301 
3302 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3303 			 u16 tss_selector, int idt_index, int reason,
3304 			 bool has_error_code, u32 error_code)
3305 {
3306 	int rc;
3307 
3308 	invalidate_registers(ctxt);
3309 	ctxt->_eip = ctxt->eip;
3310 	ctxt->dst.type = OP_NONE;
3311 
3312 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3313 				     has_error_code, error_code);
3314 
3315 	if (rc == X86EMUL_CONTINUE) {
3316 		ctxt->eip = ctxt->_eip;
3317 		writeback_registers(ctxt);
3318 	}
3319 
3320 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3321 }
3322 
3323 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3324 		struct operand *op)
3325 {
3326 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3327 
3328 	register_address_increment(ctxt, reg, df * op->bytes);
3329 	op->addr.mem.ea = register_address(ctxt, reg);
3330 }
3331 
3332 static int em_das(struct x86_emulate_ctxt *ctxt)
3333 {
3334 	u8 al, old_al;
3335 	bool af, cf, old_cf;
3336 
3337 	cf = ctxt->eflags & X86_EFLAGS_CF;
3338 	al = ctxt->dst.val;
3339 
3340 	old_al = al;
3341 	old_cf = cf;
3342 	cf = false;
3343 	af = ctxt->eflags & X86_EFLAGS_AF;
3344 	if ((al & 0x0f) > 9 || af) {
3345 		al -= 6;
3346 		cf = old_cf | (al >= 250);
3347 		af = true;
3348 	} else {
3349 		af = false;
3350 	}
3351 	if (old_al > 0x99 || old_cf) {
3352 		al -= 0x60;
3353 		cf = true;
3354 	}
3355 
3356 	ctxt->dst.val = al;
3357 	/* Set PF, ZF, SF */
3358 	ctxt->src.type = OP_IMM;
3359 	ctxt->src.val = 0;
3360 	ctxt->src.bytes = 1;
3361 	fastop(ctxt, em_or);
3362 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3363 	if (cf)
3364 		ctxt->eflags |= X86_EFLAGS_CF;
3365 	if (af)
3366 		ctxt->eflags |= X86_EFLAGS_AF;
3367 	return X86EMUL_CONTINUE;
3368 }
3369 
3370 static int em_aam(struct x86_emulate_ctxt *ctxt)
3371 {
3372 	u8 al, ah;
3373 
3374 	if (ctxt->src.val == 0)
3375 		return emulate_de(ctxt);
3376 
3377 	al = ctxt->dst.val & 0xff;
3378 	ah = al / ctxt->src.val;
3379 	al %= ctxt->src.val;
3380 
3381 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3382 
3383 	/* Set PF, ZF, SF */
3384 	ctxt->src.type = OP_IMM;
3385 	ctxt->src.val = 0;
3386 	ctxt->src.bytes = 1;
3387 	fastop(ctxt, em_or);
3388 
3389 	return X86EMUL_CONTINUE;
3390 }
3391 
3392 static int em_aad(struct x86_emulate_ctxt *ctxt)
3393 {
3394 	u8 al = ctxt->dst.val & 0xff;
3395 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3396 
3397 	al = (al + (ah * ctxt->src.val)) & 0xff;
3398 
3399 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3400 
3401 	/* Set PF, ZF, SF */
3402 	ctxt->src.type = OP_IMM;
3403 	ctxt->src.val = 0;
3404 	ctxt->src.bytes = 1;
3405 	fastop(ctxt, em_or);
3406 
3407 	return X86EMUL_CONTINUE;
3408 }
3409 
3410 static int em_call(struct x86_emulate_ctxt *ctxt)
3411 {
3412 	int rc;
3413 	long rel = ctxt->src.val;
3414 
3415 	ctxt->src.val = (unsigned long)ctxt->_eip;
3416 	rc = jmp_rel(ctxt, rel);
3417 	if (rc != X86EMUL_CONTINUE)
3418 		return rc;
3419 	return em_push(ctxt);
3420 }
3421 
3422 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3423 {
3424 	u16 sel, old_cs;
3425 	ulong old_eip;
3426 	int rc;
3427 	struct desc_struct old_desc, new_desc;
3428 	const struct x86_emulate_ops *ops = ctxt->ops;
3429 	int cpl = ctxt->ops->cpl(ctxt);
3430 	enum x86emul_mode prev_mode = ctxt->mode;
3431 
3432 	old_eip = ctxt->_eip;
3433 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3434 
3435 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3436 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3437 				       X86_TRANSFER_CALL_JMP, &new_desc);
3438 	if (rc != X86EMUL_CONTINUE)
3439 		return rc;
3440 
3441 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3442 	if (rc != X86EMUL_CONTINUE)
3443 		goto fail;
3444 
3445 	ctxt->src.val = old_cs;
3446 	rc = em_push(ctxt);
3447 	if (rc != X86EMUL_CONTINUE)
3448 		goto fail;
3449 
3450 	ctxt->src.val = old_eip;
3451 	rc = em_push(ctxt);
3452 	/* If we failed, we tainted the memory, but the very least we should
3453 	   restore cs */
3454 	if (rc != X86EMUL_CONTINUE) {
3455 		pr_warn_once("faulting far call emulation tainted memory\n");
3456 		goto fail;
3457 	}
3458 	return rc;
3459 fail:
3460 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3461 	ctxt->mode = prev_mode;
3462 	return rc;
3463 
3464 }
3465 
3466 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3467 {
3468 	int rc;
3469 	unsigned long eip;
3470 
3471 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3472 	if (rc != X86EMUL_CONTINUE)
3473 		return rc;
3474 	rc = assign_eip_near(ctxt, eip);
3475 	if (rc != X86EMUL_CONTINUE)
3476 		return rc;
3477 	rsp_increment(ctxt, ctxt->src.val);
3478 	return X86EMUL_CONTINUE;
3479 }
3480 
3481 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3482 {
3483 	/* Write back the register source. */
3484 	ctxt->src.val = ctxt->dst.val;
3485 	write_register_operand(&ctxt->src);
3486 
3487 	/* Write back the memory destination with implicit LOCK prefix. */
3488 	ctxt->dst.val = ctxt->src.orig_val;
3489 	ctxt->lock_prefix = 1;
3490 	return X86EMUL_CONTINUE;
3491 }
3492 
3493 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3494 {
3495 	ctxt->dst.val = ctxt->src2.val;
3496 	return fastop(ctxt, em_imul);
3497 }
3498 
3499 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3500 {
3501 	ctxt->dst.type = OP_REG;
3502 	ctxt->dst.bytes = ctxt->src.bytes;
3503 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3504 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3505 
3506 	return X86EMUL_CONTINUE;
3507 }
3508 
3509 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3510 {
3511 	u64 tsc_aux = 0;
3512 
3513 	if (ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux))
3514 		return emulate_ud(ctxt);
3515 	ctxt->dst.val = tsc_aux;
3516 	return X86EMUL_CONTINUE;
3517 }
3518 
3519 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3520 {
3521 	u64 tsc = 0;
3522 
3523 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3524 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3525 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3526 	return X86EMUL_CONTINUE;
3527 }
3528 
3529 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3530 {
3531 	u64 pmc;
3532 
3533 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3534 		return emulate_gp(ctxt, 0);
3535 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3536 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3537 	return X86EMUL_CONTINUE;
3538 }
3539 
3540 static int em_mov(struct x86_emulate_ctxt *ctxt)
3541 {
3542 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3543 	return X86EMUL_CONTINUE;
3544 }
3545 
3546 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3547 {
3548 	u16 tmp;
3549 
3550 	if (!ctxt->ops->guest_has_movbe(ctxt))
3551 		return emulate_ud(ctxt);
3552 
3553 	switch (ctxt->op_bytes) {
3554 	case 2:
3555 		/*
3556 		 * From MOVBE definition: "...When the operand size is 16 bits,
3557 		 * the upper word of the destination register remains unchanged
3558 		 * ..."
3559 		 *
3560 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3561 		 * rules so we have to do the operation almost per hand.
3562 		 */
3563 		tmp = (u16)ctxt->src.val;
3564 		ctxt->dst.val &= ~0xffffUL;
3565 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3566 		break;
3567 	case 4:
3568 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3569 		break;
3570 	case 8:
3571 		ctxt->dst.val = swab64(ctxt->src.val);
3572 		break;
3573 	default:
3574 		BUG();
3575 	}
3576 	return X86EMUL_CONTINUE;
3577 }
3578 
3579 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3580 {
3581 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3582 		return emulate_gp(ctxt, 0);
3583 
3584 	/* Disable writeback. */
3585 	ctxt->dst.type = OP_NONE;
3586 	return X86EMUL_CONTINUE;
3587 }
3588 
3589 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3590 {
3591 	unsigned long val;
3592 
3593 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3594 		val = ctxt->src.val & ~0ULL;
3595 	else
3596 		val = ctxt->src.val & ~0U;
3597 
3598 	/* #UD condition is already handled. */
3599 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3600 		return emulate_gp(ctxt, 0);
3601 
3602 	/* Disable writeback. */
3603 	ctxt->dst.type = OP_NONE;
3604 	return X86EMUL_CONTINUE;
3605 }
3606 
3607 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3608 {
3609 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3610 	u64 msr_data;
3611 	int r;
3612 
3613 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3614 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3615 	r = ctxt->ops->set_msr(ctxt, msr_index, msr_data);
3616 
3617 	if (r == X86EMUL_IO_NEEDED)
3618 		return r;
3619 
3620 	if (r > 0)
3621 		return emulate_gp(ctxt, 0);
3622 
3623 	return r < 0 ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
3624 }
3625 
3626 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3627 {
3628 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3629 	u64 msr_data;
3630 	int r;
3631 
3632 	r = ctxt->ops->get_msr(ctxt, msr_index, &msr_data);
3633 
3634 	if (r == X86EMUL_IO_NEEDED)
3635 		return r;
3636 
3637 	if (r)
3638 		return emulate_gp(ctxt, 0);
3639 
3640 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3641 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3642 	return X86EMUL_CONTINUE;
3643 }
3644 
3645 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3646 {
3647 	if (segment > VCPU_SREG_GS &&
3648 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3649 	    ctxt->ops->cpl(ctxt) > 0)
3650 		return emulate_gp(ctxt, 0);
3651 
3652 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3653 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3654 		ctxt->dst.bytes = 2;
3655 	return X86EMUL_CONTINUE;
3656 }
3657 
3658 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3659 {
3660 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3661 		return emulate_ud(ctxt);
3662 
3663 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3664 }
3665 
3666 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3667 {
3668 	u16 sel = ctxt->src.val;
3669 
3670 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3671 		return emulate_ud(ctxt);
3672 
3673 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3674 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3675 
3676 	/* Disable writeback. */
3677 	ctxt->dst.type = OP_NONE;
3678 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3679 }
3680 
3681 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3682 {
3683 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3684 }
3685 
3686 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3687 {
3688 	u16 sel = ctxt->src.val;
3689 
3690 	/* Disable writeback. */
3691 	ctxt->dst.type = OP_NONE;
3692 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3693 }
3694 
3695 static int em_str(struct x86_emulate_ctxt *ctxt)
3696 {
3697 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3698 }
3699 
3700 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3701 {
3702 	u16 sel = ctxt->src.val;
3703 
3704 	/* Disable writeback. */
3705 	ctxt->dst.type = OP_NONE;
3706 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3707 }
3708 
3709 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3710 {
3711 	int rc;
3712 	ulong linear;
3713 
3714 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3715 	if (rc == X86EMUL_CONTINUE)
3716 		ctxt->ops->invlpg(ctxt, linear);
3717 	/* Disable writeback. */
3718 	ctxt->dst.type = OP_NONE;
3719 	return X86EMUL_CONTINUE;
3720 }
3721 
3722 static int em_clts(struct x86_emulate_ctxt *ctxt)
3723 {
3724 	ulong cr0;
3725 
3726 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3727 	cr0 &= ~X86_CR0_TS;
3728 	ctxt->ops->set_cr(ctxt, 0, cr0);
3729 	return X86EMUL_CONTINUE;
3730 }
3731 
3732 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3733 {
3734 	int rc = ctxt->ops->fix_hypercall(ctxt);
3735 
3736 	if (rc != X86EMUL_CONTINUE)
3737 		return rc;
3738 
3739 	/* Let the processor re-execute the fixed hypercall */
3740 	ctxt->_eip = ctxt->eip;
3741 	/* Disable writeback. */
3742 	ctxt->dst.type = OP_NONE;
3743 	return X86EMUL_CONTINUE;
3744 }
3745 
3746 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3747 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3748 					      struct desc_ptr *ptr))
3749 {
3750 	struct desc_ptr desc_ptr;
3751 
3752 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3753 	    ctxt->ops->cpl(ctxt) > 0)
3754 		return emulate_gp(ctxt, 0);
3755 
3756 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3757 		ctxt->op_bytes = 8;
3758 	get(ctxt, &desc_ptr);
3759 	if (ctxt->op_bytes == 2) {
3760 		ctxt->op_bytes = 4;
3761 		desc_ptr.address &= 0x00ffffff;
3762 	}
3763 	/* Disable writeback. */
3764 	ctxt->dst.type = OP_NONE;
3765 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3766 				   &desc_ptr, 2 + ctxt->op_bytes);
3767 }
3768 
3769 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3770 {
3771 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3772 }
3773 
3774 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3775 {
3776 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3777 }
3778 
3779 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3780 {
3781 	struct desc_ptr desc_ptr;
3782 	int rc;
3783 
3784 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3785 		ctxt->op_bytes = 8;
3786 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3787 			     &desc_ptr.size, &desc_ptr.address,
3788 			     ctxt->op_bytes);
3789 	if (rc != X86EMUL_CONTINUE)
3790 		return rc;
3791 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3792 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3793 		return emulate_gp(ctxt, 0);
3794 	if (lgdt)
3795 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3796 	else
3797 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3798 	/* Disable writeback. */
3799 	ctxt->dst.type = OP_NONE;
3800 	return X86EMUL_CONTINUE;
3801 }
3802 
3803 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3804 {
3805 	return em_lgdt_lidt(ctxt, true);
3806 }
3807 
3808 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3809 {
3810 	return em_lgdt_lidt(ctxt, false);
3811 }
3812 
3813 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3814 {
3815 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3816 	    ctxt->ops->cpl(ctxt) > 0)
3817 		return emulate_gp(ctxt, 0);
3818 
3819 	if (ctxt->dst.type == OP_MEM)
3820 		ctxt->dst.bytes = 2;
3821 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3822 	return X86EMUL_CONTINUE;
3823 }
3824 
3825 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3826 {
3827 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3828 			  | (ctxt->src.val & 0x0f));
3829 	ctxt->dst.type = OP_NONE;
3830 	return X86EMUL_CONTINUE;
3831 }
3832 
3833 static int em_loop(struct x86_emulate_ctxt *ctxt)
3834 {
3835 	int rc = X86EMUL_CONTINUE;
3836 
3837 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3838 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3839 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3840 		rc = jmp_rel(ctxt, ctxt->src.val);
3841 
3842 	return rc;
3843 }
3844 
3845 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3846 {
3847 	int rc = X86EMUL_CONTINUE;
3848 
3849 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3850 		rc = jmp_rel(ctxt, ctxt->src.val);
3851 
3852 	return rc;
3853 }
3854 
3855 static int em_in(struct x86_emulate_ctxt *ctxt)
3856 {
3857 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3858 			     &ctxt->dst.val))
3859 		return X86EMUL_IO_NEEDED;
3860 
3861 	return X86EMUL_CONTINUE;
3862 }
3863 
3864 static int em_out(struct x86_emulate_ctxt *ctxt)
3865 {
3866 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3867 				    &ctxt->src.val, 1);
3868 	/* Disable writeback. */
3869 	ctxt->dst.type = OP_NONE;
3870 	return X86EMUL_CONTINUE;
3871 }
3872 
3873 static int em_cli(struct x86_emulate_ctxt *ctxt)
3874 {
3875 	if (emulator_bad_iopl(ctxt))
3876 		return emulate_gp(ctxt, 0);
3877 
3878 	ctxt->eflags &= ~X86_EFLAGS_IF;
3879 	return X86EMUL_CONTINUE;
3880 }
3881 
3882 static int em_sti(struct x86_emulate_ctxt *ctxt)
3883 {
3884 	if (emulator_bad_iopl(ctxt))
3885 		return emulate_gp(ctxt, 0);
3886 
3887 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3888 	ctxt->eflags |= X86_EFLAGS_IF;
3889 	return X86EMUL_CONTINUE;
3890 }
3891 
3892 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3893 {
3894 	u32 eax, ebx, ecx, edx;
3895 	u64 msr = 0;
3896 
3897 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3898 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3899 	    ctxt->ops->cpl(ctxt)) {
3900 		return emulate_gp(ctxt, 0);
3901 	}
3902 
3903 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3904 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3905 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3906 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3907 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3908 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3909 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3910 	return X86EMUL_CONTINUE;
3911 }
3912 
3913 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3914 {
3915 	u32 flags;
3916 
3917 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3918 		X86_EFLAGS_SF;
3919 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3920 
3921 	ctxt->eflags &= ~0xffUL;
3922 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3923 	return X86EMUL_CONTINUE;
3924 }
3925 
3926 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3927 {
3928 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3929 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3930 	return X86EMUL_CONTINUE;
3931 }
3932 
3933 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3934 {
3935 	switch (ctxt->op_bytes) {
3936 #ifdef CONFIG_X86_64
3937 	case 8:
3938 		asm("bswap %0" : "+r"(ctxt->dst.val));
3939 		break;
3940 #endif
3941 	default:
3942 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3943 		break;
3944 	}
3945 	return X86EMUL_CONTINUE;
3946 }
3947 
3948 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3949 {
3950 	/* emulating clflush regardless of cpuid */
3951 	return X86EMUL_CONTINUE;
3952 }
3953 
3954 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
3955 {
3956 	/* emulating clflushopt regardless of cpuid */
3957 	return X86EMUL_CONTINUE;
3958 }
3959 
3960 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3961 {
3962 	ctxt->dst.val = (s32) ctxt->src.val;
3963 	return X86EMUL_CONTINUE;
3964 }
3965 
3966 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3967 {
3968 	if (!ctxt->ops->guest_has_fxsr(ctxt))
3969 		return emulate_ud(ctxt);
3970 
3971 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3972 		return emulate_nm(ctxt);
3973 
3974 	/*
3975 	 * Don't emulate a case that should never be hit, instead of working
3976 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3977 	 */
3978 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3979 		return X86EMUL_UNHANDLEABLE;
3980 
3981 	return X86EMUL_CONTINUE;
3982 }
3983 
3984 /*
3985  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
3986  * and restore MXCSR.
3987  */
3988 static size_t __fxstate_size(int nregs)
3989 {
3990 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
3991 }
3992 
3993 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
3994 {
3995 	bool cr4_osfxsr;
3996 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3997 		return __fxstate_size(16);
3998 
3999 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
4000 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
4001 }
4002 
4003 /*
4004  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
4005  *  1) 16 bit mode
4006  *  2) 32 bit mode
4007  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
4008  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
4009  *       save and restore
4010  *  3) 64-bit mode with REX.W prefix
4011  *     - like (2), but XMM 8-15 are being saved and restored
4012  *  4) 64-bit mode without REX.W prefix
4013  *     - like (3), but FIP and FDP are 64 bit
4014  *
4015  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
4016  * desired result.  (4) is not emulated.
4017  *
4018  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
4019  * and FPU DS) should match.
4020  */
4021 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
4022 {
4023 	struct fxregs_state fx_state;
4024 	int rc;
4025 
4026 	rc = check_fxsr(ctxt);
4027 	if (rc != X86EMUL_CONTINUE)
4028 		return rc;
4029 
4030 	kvm_fpu_get();
4031 
4032 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4033 
4034 	kvm_fpu_put();
4035 
4036 	if (rc != X86EMUL_CONTINUE)
4037 		return rc;
4038 
4039 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4040 		                   fxstate_size(ctxt));
4041 }
4042 
4043 /*
4044  * FXRSTOR might restore XMM registers not provided by the guest. Fill
4045  * in the host registers (via FXSAVE) instead, so they won't be modified.
4046  * (preemption has to stay disabled until FXRSTOR).
4047  *
4048  * Use noinline to keep the stack for other functions called by callers small.
4049  */
4050 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
4051 				 const size_t used_size)
4052 {
4053 	struct fxregs_state fx_tmp;
4054 	int rc;
4055 
4056 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
4057 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
4058 	       __fxstate_size(16) - used_size);
4059 
4060 	return rc;
4061 }
4062 
4063 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4064 {
4065 	struct fxregs_state fx_state;
4066 	int rc;
4067 	size_t size;
4068 
4069 	rc = check_fxsr(ctxt);
4070 	if (rc != X86EMUL_CONTINUE)
4071 		return rc;
4072 
4073 	size = fxstate_size(ctxt);
4074 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4075 	if (rc != X86EMUL_CONTINUE)
4076 		return rc;
4077 
4078 	kvm_fpu_get();
4079 
4080 	if (size < __fxstate_size(16)) {
4081 		rc = fxregs_fixup(&fx_state, size);
4082 		if (rc != X86EMUL_CONTINUE)
4083 			goto out;
4084 	}
4085 
4086 	if (fx_state.mxcsr >> 16) {
4087 		rc = emulate_gp(ctxt, 0);
4088 		goto out;
4089 	}
4090 
4091 	if (rc == X86EMUL_CONTINUE)
4092 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4093 
4094 out:
4095 	kvm_fpu_put();
4096 
4097 	return rc;
4098 }
4099 
4100 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
4101 {
4102 	u32 eax, ecx, edx;
4103 
4104 	eax = reg_read(ctxt, VCPU_REGS_RAX);
4105 	edx = reg_read(ctxt, VCPU_REGS_RDX);
4106 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
4107 
4108 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
4109 		return emulate_gp(ctxt, 0);
4110 
4111 	return X86EMUL_CONTINUE;
4112 }
4113 
4114 static bool valid_cr(int nr)
4115 {
4116 	switch (nr) {
4117 	case 0:
4118 	case 2 ... 4:
4119 	case 8:
4120 		return true;
4121 	default:
4122 		return false;
4123 	}
4124 }
4125 
4126 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
4127 {
4128 	if (!valid_cr(ctxt->modrm_reg))
4129 		return emulate_ud(ctxt);
4130 
4131 	return X86EMUL_CONTINUE;
4132 }
4133 
4134 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4135 {
4136 	unsigned long dr7;
4137 
4138 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4139 
4140 	/* Check if DR7.Global_Enable is set */
4141 	return dr7 & (1 << 13);
4142 }
4143 
4144 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4145 {
4146 	int dr = ctxt->modrm_reg;
4147 	u64 cr4;
4148 
4149 	if (dr > 7)
4150 		return emulate_ud(ctxt);
4151 
4152 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4153 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4154 		return emulate_ud(ctxt);
4155 
4156 	if (check_dr7_gd(ctxt)) {
4157 		ulong dr6;
4158 
4159 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4160 		dr6 &= ~DR_TRAP_BITS;
4161 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
4162 		ctxt->ops->set_dr(ctxt, 6, dr6);
4163 		return emulate_db(ctxt);
4164 	}
4165 
4166 	return X86EMUL_CONTINUE;
4167 }
4168 
4169 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4170 {
4171 	u64 new_val = ctxt->src.val64;
4172 	int dr = ctxt->modrm_reg;
4173 
4174 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4175 		return emulate_gp(ctxt, 0);
4176 
4177 	return check_dr_read(ctxt);
4178 }
4179 
4180 static int check_svme(struct x86_emulate_ctxt *ctxt)
4181 {
4182 	u64 efer = 0;
4183 
4184 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4185 
4186 	if (!(efer & EFER_SVME))
4187 		return emulate_ud(ctxt);
4188 
4189 	return X86EMUL_CONTINUE;
4190 }
4191 
4192 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4193 {
4194 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4195 
4196 	/* Valid physical address? */
4197 	if (rax & 0xffff000000000000ULL)
4198 		return emulate_gp(ctxt, 0);
4199 
4200 	return check_svme(ctxt);
4201 }
4202 
4203 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4204 {
4205 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4206 
4207 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4208 		return emulate_gp(ctxt, 0);
4209 
4210 	return X86EMUL_CONTINUE;
4211 }
4212 
4213 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4214 {
4215 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4216 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4217 
4218 	/*
4219 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
4220 	 * in Ring3 when CR4.PCE=0.
4221 	 */
4222 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
4223 		return X86EMUL_CONTINUE;
4224 
4225 	/*
4226 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
4227 	 * check however is unnecessary because CPL is always 0 outside
4228 	 * protected mode.
4229 	 */
4230 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4231 	    ctxt->ops->check_pmc(ctxt, rcx))
4232 		return emulate_gp(ctxt, 0);
4233 
4234 	return X86EMUL_CONTINUE;
4235 }
4236 
4237 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4238 {
4239 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4240 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4241 		return emulate_gp(ctxt, 0);
4242 
4243 	return X86EMUL_CONTINUE;
4244 }
4245 
4246 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4247 {
4248 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4249 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4250 		return emulate_gp(ctxt, 0);
4251 
4252 	return X86EMUL_CONTINUE;
4253 }
4254 
4255 #define D(_y) { .flags = (_y) }
4256 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4257 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4258 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4259 #define N    D(NotImpl)
4260 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4261 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4262 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4263 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4264 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4265 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4266 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4267 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4268 #define II(_f, _e, _i) \
4269 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4270 #define IIP(_f, _e, _i, _p) \
4271 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4272 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4273 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4274 
4275 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4276 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4277 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4278 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4279 #define I2bvIP(_f, _e, _i, _p) \
4280 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4281 
4282 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4283 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4284 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4285 
4286 static const struct opcode group7_rm0[] = {
4287 	N,
4288 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4289 	N, N, N, N, N, N,
4290 };
4291 
4292 static const struct opcode group7_rm1[] = {
4293 	DI(SrcNone | Priv, monitor),
4294 	DI(SrcNone | Priv, mwait),
4295 	N, N, N, N, N, N,
4296 };
4297 
4298 static const struct opcode group7_rm2[] = {
4299 	N,
4300 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4301 	N, N, N, N, N, N,
4302 };
4303 
4304 static const struct opcode group7_rm3[] = {
4305 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4306 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4307 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4308 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4309 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4310 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4311 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4312 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4313 };
4314 
4315 static const struct opcode group7_rm7[] = {
4316 	N,
4317 	DIP(SrcNone, rdtscp, check_rdtsc),
4318 	N, N, N, N, N, N,
4319 };
4320 
4321 static const struct opcode group1[] = {
4322 	F(Lock, em_add),
4323 	F(Lock | PageTable, em_or),
4324 	F(Lock, em_adc),
4325 	F(Lock, em_sbb),
4326 	F(Lock | PageTable, em_and),
4327 	F(Lock, em_sub),
4328 	F(Lock, em_xor),
4329 	F(NoWrite, em_cmp),
4330 };
4331 
4332 static const struct opcode group1A[] = {
4333 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4334 };
4335 
4336 static const struct opcode group2[] = {
4337 	F(DstMem | ModRM, em_rol),
4338 	F(DstMem | ModRM, em_ror),
4339 	F(DstMem | ModRM, em_rcl),
4340 	F(DstMem | ModRM, em_rcr),
4341 	F(DstMem | ModRM, em_shl),
4342 	F(DstMem | ModRM, em_shr),
4343 	F(DstMem | ModRM, em_shl),
4344 	F(DstMem | ModRM, em_sar),
4345 };
4346 
4347 static const struct opcode group3[] = {
4348 	F(DstMem | SrcImm | NoWrite, em_test),
4349 	F(DstMem | SrcImm | NoWrite, em_test),
4350 	F(DstMem | SrcNone | Lock, em_not),
4351 	F(DstMem | SrcNone | Lock, em_neg),
4352 	F(DstXacc | Src2Mem, em_mul_ex),
4353 	F(DstXacc | Src2Mem, em_imul_ex),
4354 	F(DstXacc | Src2Mem, em_div_ex),
4355 	F(DstXacc | Src2Mem, em_idiv_ex),
4356 };
4357 
4358 static const struct opcode group4[] = {
4359 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4360 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4361 	N, N, N, N, N, N,
4362 };
4363 
4364 static const struct opcode group5[] = {
4365 	F(DstMem | SrcNone | Lock,		em_inc),
4366 	F(DstMem | SrcNone | Lock,		em_dec),
4367 	I(SrcMem | NearBranch,			em_call_near_abs),
4368 	I(SrcMemFAddr | ImplicitOps,		em_call_far),
4369 	I(SrcMem | NearBranch,			em_jmp_abs),
4370 	I(SrcMemFAddr | ImplicitOps,		em_jmp_far),
4371 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4372 };
4373 
4374 static const struct opcode group6[] = {
4375 	II(Prot | DstMem,	   em_sldt, sldt),
4376 	II(Prot | DstMem,	   em_str, str),
4377 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4378 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4379 	N, N, N, N,
4380 };
4381 
4382 static const struct group_dual group7 = { {
4383 	II(Mov | DstMem,			em_sgdt, sgdt),
4384 	II(Mov | DstMem,			em_sidt, sidt),
4385 	II(SrcMem | Priv,			em_lgdt, lgdt),
4386 	II(SrcMem | Priv,			em_lidt, lidt),
4387 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4388 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4389 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4390 }, {
4391 	EXT(0, group7_rm0),
4392 	EXT(0, group7_rm1),
4393 	EXT(0, group7_rm2),
4394 	EXT(0, group7_rm3),
4395 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4396 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4397 	EXT(0, group7_rm7),
4398 } };
4399 
4400 static const struct opcode group8[] = {
4401 	N, N, N, N,
4402 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4403 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4404 	F(DstMem | SrcImmByte | Lock,			em_btr),
4405 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4406 };
4407 
4408 /*
4409  * The "memory" destination is actually always a register, since we come
4410  * from the register case of group9.
4411  */
4412 static const struct gprefix pfx_0f_c7_7 = {
4413 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4414 };
4415 
4416 
4417 static const struct group_dual group9 = { {
4418 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4419 }, {
4420 	N, N, N, N, N, N, N,
4421 	GP(0, &pfx_0f_c7_7),
4422 } };
4423 
4424 static const struct opcode group11[] = {
4425 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4426 	X7(D(Undefined)),
4427 };
4428 
4429 static const struct gprefix pfx_0f_ae_7 = {
4430 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4431 };
4432 
4433 static const struct group_dual group15 = { {
4434 	I(ModRM | Aligned16, em_fxsave),
4435 	I(ModRM | Aligned16, em_fxrstor),
4436 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4437 }, {
4438 	N, N, N, N, N, N, N, N,
4439 } };
4440 
4441 static const struct gprefix pfx_0f_6f_0f_7f = {
4442 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4443 };
4444 
4445 static const struct instr_dual instr_dual_0f_2b = {
4446 	I(0, em_mov), N
4447 };
4448 
4449 static const struct gprefix pfx_0f_2b = {
4450 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4451 };
4452 
4453 static const struct gprefix pfx_0f_10_0f_11 = {
4454 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4455 };
4456 
4457 static const struct gprefix pfx_0f_28_0f_29 = {
4458 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4459 };
4460 
4461 static const struct gprefix pfx_0f_e7 = {
4462 	N, I(Sse, em_mov), N, N,
4463 };
4464 
4465 static const struct escape escape_d9 = { {
4466 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4467 }, {
4468 	/* 0xC0 - 0xC7 */
4469 	N, N, N, N, N, N, N, N,
4470 	/* 0xC8 - 0xCF */
4471 	N, N, N, N, N, N, N, N,
4472 	/* 0xD0 - 0xC7 */
4473 	N, N, N, N, N, N, N, N,
4474 	/* 0xD8 - 0xDF */
4475 	N, N, N, N, N, N, N, N,
4476 	/* 0xE0 - 0xE7 */
4477 	N, N, N, N, N, N, N, N,
4478 	/* 0xE8 - 0xEF */
4479 	N, N, N, N, N, N, N, N,
4480 	/* 0xF0 - 0xF7 */
4481 	N, N, N, N, N, N, N, N,
4482 	/* 0xF8 - 0xFF */
4483 	N, N, N, N, N, N, N, N,
4484 } };
4485 
4486 static const struct escape escape_db = { {
4487 	N, N, N, N, N, N, N, N,
4488 }, {
4489 	/* 0xC0 - 0xC7 */
4490 	N, N, N, N, N, N, N, N,
4491 	/* 0xC8 - 0xCF */
4492 	N, N, N, N, N, N, N, N,
4493 	/* 0xD0 - 0xC7 */
4494 	N, N, N, N, N, N, N, N,
4495 	/* 0xD8 - 0xDF */
4496 	N, N, N, N, N, N, N, N,
4497 	/* 0xE0 - 0xE7 */
4498 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4499 	/* 0xE8 - 0xEF */
4500 	N, N, N, N, N, N, N, N,
4501 	/* 0xF0 - 0xF7 */
4502 	N, N, N, N, N, N, N, N,
4503 	/* 0xF8 - 0xFF */
4504 	N, N, N, N, N, N, N, N,
4505 } };
4506 
4507 static const struct escape escape_dd = { {
4508 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4509 }, {
4510 	/* 0xC0 - 0xC7 */
4511 	N, N, N, N, N, N, N, N,
4512 	/* 0xC8 - 0xCF */
4513 	N, N, N, N, N, N, N, N,
4514 	/* 0xD0 - 0xC7 */
4515 	N, N, N, N, N, N, N, N,
4516 	/* 0xD8 - 0xDF */
4517 	N, N, N, N, N, N, N, N,
4518 	/* 0xE0 - 0xE7 */
4519 	N, N, N, N, N, N, N, N,
4520 	/* 0xE8 - 0xEF */
4521 	N, N, N, N, N, N, N, N,
4522 	/* 0xF0 - 0xF7 */
4523 	N, N, N, N, N, N, N, N,
4524 	/* 0xF8 - 0xFF */
4525 	N, N, N, N, N, N, N, N,
4526 } };
4527 
4528 static const struct instr_dual instr_dual_0f_c3 = {
4529 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4530 };
4531 
4532 static const struct mode_dual mode_dual_63 = {
4533 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4534 };
4535 
4536 static const struct opcode opcode_table[256] = {
4537 	/* 0x00 - 0x07 */
4538 	F6ALU(Lock, em_add),
4539 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4540 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4541 	/* 0x08 - 0x0F */
4542 	F6ALU(Lock | PageTable, em_or),
4543 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4544 	N,
4545 	/* 0x10 - 0x17 */
4546 	F6ALU(Lock, em_adc),
4547 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4548 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4549 	/* 0x18 - 0x1F */
4550 	F6ALU(Lock, em_sbb),
4551 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4552 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4553 	/* 0x20 - 0x27 */
4554 	F6ALU(Lock | PageTable, em_and), N, N,
4555 	/* 0x28 - 0x2F */
4556 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4557 	/* 0x30 - 0x37 */
4558 	F6ALU(Lock, em_xor), N, N,
4559 	/* 0x38 - 0x3F */
4560 	F6ALU(NoWrite, em_cmp), N, N,
4561 	/* 0x40 - 0x4F */
4562 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4563 	/* 0x50 - 0x57 */
4564 	X8(I(SrcReg | Stack, em_push)),
4565 	/* 0x58 - 0x5F */
4566 	X8(I(DstReg | Stack, em_pop)),
4567 	/* 0x60 - 0x67 */
4568 	I(ImplicitOps | Stack | No64, em_pusha),
4569 	I(ImplicitOps | Stack | No64, em_popa),
4570 	N, MD(ModRM, &mode_dual_63),
4571 	N, N, N, N,
4572 	/* 0x68 - 0x6F */
4573 	I(SrcImm | Mov | Stack, em_push),
4574 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4575 	I(SrcImmByte | Mov | Stack, em_push),
4576 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4577 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4578 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4579 	/* 0x70 - 0x7F */
4580 	X16(D(SrcImmByte | NearBranch)),
4581 	/* 0x80 - 0x87 */
4582 	G(ByteOp | DstMem | SrcImm, group1),
4583 	G(DstMem | SrcImm, group1),
4584 	G(ByteOp | DstMem | SrcImm | No64, group1),
4585 	G(DstMem | SrcImmByte, group1),
4586 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4587 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4588 	/* 0x88 - 0x8F */
4589 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4590 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4591 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4592 	D(ModRM | SrcMem | NoAccess | DstReg),
4593 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4594 	G(0, group1A),
4595 	/* 0x90 - 0x97 */
4596 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4597 	/* 0x98 - 0x9F */
4598 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4599 	I(SrcImmFAddr | No64, em_call_far), N,
4600 	II(ImplicitOps | Stack, em_pushf, pushf),
4601 	II(ImplicitOps | Stack, em_popf, popf),
4602 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4603 	/* 0xA0 - 0xA7 */
4604 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4605 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4606 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4607 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4608 	/* 0xA8 - 0xAF */
4609 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4610 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4611 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4612 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4613 	/* 0xB0 - 0xB7 */
4614 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4615 	/* 0xB8 - 0xBF */
4616 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4617 	/* 0xC0 - 0xC7 */
4618 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4619 	I(ImplicitOps | NearBranch | SrcImmU16, em_ret_near_imm),
4620 	I(ImplicitOps | NearBranch, em_ret),
4621 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4622 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4623 	G(ByteOp, group11), G(0, group11),
4624 	/* 0xC8 - 0xCF */
4625 	I(Stack | SrcImmU16 | Src2ImmByte, em_enter), I(Stack, em_leave),
4626 	I(ImplicitOps | SrcImmU16, em_ret_far_imm),
4627 	I(ImplicitOps, em_ret_far),
4628 	D(ImplicitOps), DI(SrcImmByte, intn),
4629 	D(ImplicitOps | No64), II(ImplicitOps, em_iret, iret),
4630 	/* 0xD0 - 0xD7 */
4631 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4632 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4633 	I(DstAcc | SrcImmUByte | No64, em_aam),
4634 	I(DstAcc | SrcImmUByte | No64, em_aad),
4635 	F(DstAcc | ByteOp | No64, em_salc),
4636 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4637 	/* 0xD8 - 0xDF */
4638 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4639 	/* 0xE0 - 0xE7 */
4640 	X3(I(SrcImmByte | NearBranch, em_loop)),
4641 	I(SrcImmByte | NearBranch, em_jcxz),
4642 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4643 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4644 	/* 0xE8 - 0xEF */
4645 	I(SrcImm | NearBranch, em_call), D(SrcImm | ImplicitOps | NearBranch),
4646 	I(SrcImmFAddr | No64, em_jmp_far),
4647 	D(SrcImmByte | ImplicitOps | NearBranch),
4648 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4649 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4650 	/* 0xF0 - 0xF7 */
4651 	N, DI(ImplicitOps, icebp), N, N,
4652 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4653 	G(ByteOp, group3), G(0, group3),
4654 	/* 0xF8 - 0xFF */
4655 	D(ImplicitOps), D(ImplicitOps),
4656 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4657 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4658 };
4659 
4660 static const struct opcode twobyte_table[256] = {
4661 	/* 0x00 - 0x0F */
4662 	G(0, group6), GD(0, &group7), N, N,
4663 	N, I(ImplicitOps | EmulateOnUD, em_syscall),
4664 	II(ImplicitOps | Priv, em_clts, clts), N,
4665 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4666 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4667 	/* 0x10 - 0x1F */
4668 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4669 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4670 	N, N, N, N, N, N,
4671 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4672 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4673 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4674 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4675 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4676 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4677 	/* 0x20 - 0x2F */
4678 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4679 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4680 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4681 						check_cr_access),
4682 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4683 						check_dr_write),
4684 	N, N, N, N,
4685 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4686 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4687 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4688 	N, N, N, N,
4689 	/* 0x30 - 0x3F */
4690 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4691 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4692 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4693 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4694 	I(ImplicitOps | EmulateOnUD, em_sysenter),
4695 	I(ImplicitOps | Priv | EmulateOnUD, em_sysexit),
4696 	N, N,
4697 	N, N, N, N, N, N, N, N,
4698 	/* 0x40 - 0x4F */
4699 	X16(D(DstReg | SrcMem | ModRM)),
4700 	/* 0x50 - 0x5F */
4701 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4702 	/* 0x60 - 0x6F */
4703 	N, N, N, N,
4704 	N, N, N, N,
4705 	N, N, N, N,
4706 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4707 	/* 0x70 - 0x7F */
4708 	N, N, N, N,
4709 	N, N, N, N,
4710 	N, N, N, N,
4711 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4712 	/* 0x80 - 0x8F */
4713 	X16(D(SrcImm | NearBranch)),
4714 	/* 0x90 - 0x9F */
4715 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4716 	/* 0xA0 - 0xA7 */
4717 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4718 	II(ImplicitOps, em_cpuid, cpuid),
4719 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4720 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4721 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4722 	/* 0xA8 - 0xAF */
4723 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4724 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4725 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4726 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4727 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4728 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4729 	/* 0xB0 - 0xB7 */
4730 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4731 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4732 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4733 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4734 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4735 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4736 	/* 0xB8 - 0xBF */
4737 	N, N,
4738 	G(BitOp, group8),
4739 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4740 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4741 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4742 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4743 	/* 0xC0 - 0xC7 */
4744 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4745 	N, ID(0, &instr_dual_0f_c3),
4746 	N, N, N, GD(0, &group9),
4747 	/* 0xC8 - 0xCF */
4748 	X8(I(DstReg, em_bswap)),
4749 	/* 0xD0 - 0xDF */
4750 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4751 	/* 0xE0 - 0xEF */
4752 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4753 	N, N, N, N, N, N, N, N,
4754 	/* 0xF0 - 0xFF */
4755 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4756 };
4757 
4758 static const struct instr_dual instr_dual_0f_38_f0 = {
4759 	I(DstReg | SrcMem | Mov, em_movbe), N
4760 };
4761 
4762 static const struct instr_dual instr_dual_0f_38_f1 = {
4763 	I(DstMem | SrcReg | Mov, em_movbe), N
4764 };
4765 
4766 static const struct gprefix three_byte_0f_38_f0 = {
4767 	ID(0, &instr_dual_0f_38_f0), N, N, N
4768 };
4769 
4770 static const struct gprefix three_byte_0f_38_f1 = {
4771 	ID(0, &instr_dual_0f_38_f1), N, N, N
4772 };
4773 
4774 /*
4775  * Insns below are selected by the prefix which indexed by the third opcode
4776  * byte.
4777  */
4778 static const struct opcode opcode_map_0f_38[256] = {
4779 	/* 0x00 - 0x7f */
4780 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4781 	/* 0x80 - 0xef */
4782 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4783 	/* 0xf0 - 0xf1 */
4784 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4785 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4786 	/* 0xf2 - 0xff */
4787 	N, N, X4(N), X8(N)
4788 };
4789 
4790 #undef D
4791 #undef N
4792 #undef G
4793 #undef GD
4794 #undef I
4795 #undef GP
4796 #undef EXT
4797 #undef MD
4798 #undef ID
4799 
4800 #undef D2bv
4801 #undef D2bvIP
4802 #undef I2bv
4803 #undef I2bvIP
4804 #undef I6ALU
4805 
4806 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4807 {
4808 	unsigned size;
4809 
4810 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4811 	if (size == 8)
4812 		size = 4;
4813 	return size;
4814 }
4815 
4816 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4817 		      unsigned size, bool sign_extension)
4818 {
4819 	int rc = X86EMUL_CONTINUE;
4820 
4821 	op->type = OP_IMM;
4822 	op->bytes = size;
4823 	op->addr.mem.ea = ctxt->_eip;
4824 	/* NB. Immediates are sign-extended as necessary. */
4825 	switch (op->bytes) {
4826 	case 1:
4827 		op->val = insn_fetch(s8, ctxt);
4828 		break;
4829 	case 2:
4830 		op->val = insn_fetch(s16, ctxt);
4831 		break;
4832 	case 4:
4833 		op->val = insn_fetch(s32, ctxt);
4834 		break;
4835 	case 8:
4836 		op->val = insn_fetch(s64, ctxt);
4837 		break;
4838 	}
4839 	if (!sign_extension) {
4840 		switch (op->bytes) {
4841 		case 1:
4842 			op->val &= 0xff;
4843 			break;
4844 		case 2:
4845 			op->val &= 0xffff;
4846 			break;
4847 		case 4:
4848 			op->val &= 0xffffffff;
4849 			break;
4850 		}
4851 	}
4852 done:
4853 	return rc;
4854 }
4855 
4856 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4857 			  unsigned d)
4858 {
4859 	int rc = X86EMUL_CONTINUE;
4860 
4861 	switch (d) {
4862 	case OpReg:
4863 		decode_register_operand(ctxt, op);
4864 		break;
4865 	case OpImmUByte:
4866 		rc = decode_imm(ctxt, op, 1, false);
4867 		break;
4868 	case OpMem:
4869 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4870 	mem_common:
4871 		*op = ctxt->memop;
4872 		ctxt->memopp = op;
4873 		if (ctxt->d & BitOp)
4874 			fetch_bit_operand(ctxt);
4875 		op->orig_val = op->val;
4876 		break;
4877 	case OpMem64:
4878 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4879 		goto mem_common;
4880 	case OpAcc:
4881 		op->type = OP_REG;
4882 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4883 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4884 		fetch_register_operand(op);
4885 		op->orig_val = op->val;
4886 		break;
4887 	case OpAccLo:
4888 		op->type = OP_REG;
4889 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4890 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4891 		fetch_register_operand(op);
4892 		op->orig_val = op->val;
4893 		break;
4894 	case OpAccHi:
4895 		if (ctxt->d & ByteOp) {
4896 			op->type = OP_NONE;
4897 			break;
4898 		}
4899 		op->type = OP_REG;
4900 		op->bytes = ctxt->op_bytes;
4901 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4902 		fetch_register_operand(op);
4903 		op->orig_val = op->val;
4904 		break;
4905 	case OpDI:
4906 		op->type = OP_MEM;
4907 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4908 		op->addr.mem.ea =
4909 			register_address(ctxt, VCPU_REGS_RDI);
4910 		op->addr.mem.seg = VCPU_SREG_ES;
4911 		op->val = 0;
4912 		op->count = 1;
4913 		break;
4914 	case OpDX:
4915 		op->type = OP_REG;
4916 		op->bytes = 2;
4917 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4918 		fetch_register_operand(op);
4919 		break;
4920 	case OpCL:
4921 		op->type = OP_IMM;
4922 		op->bytes = 1;
4923 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4924 		break;
4925 	case OpImmByte:
4926 		rc = decode_imm(ctxt, op, 1, true);
4927 		break;
4928 	case OpOne:
4929 		op->type = OP_IMM;
4930 		op->bytes = 1;
4931 		op->val = 1;
4932 		break;
4933 	case OpImm:
4934 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4935 		break;
4936 	case OpImm64:
4937 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4938 		break;
4939 	case OpMem8:
4940 		ctxt->memop.bytes = 1;
4941 		if (ctxt->memop.type == OP_REG) {
4942 			ctxt->memop.addr.reg = decode_register(ctxt,
4943 					ctxt->modrm_rm, true);
4944 			fetch_register_operand(&ctxt->memop);
4945 		}
4946 		goto mem_common;
4947 	case OpMem16:
4948 		ctxt->memop.bytes = 2;
4949 		goto mem_common;
4950 	case OpMem32:
4951 		ctxt->memop.bytes = 4;
4952 		goto mem_common;
4953 	case OpImmU16:
4954 		rc = decode_imm(ctxt, op, 2, false);
4955 		break;
4956 	case OpImmU:
4957 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4958 		break;
4959 	case OpSI:
4960 		op->type = OP_MEM;
4961 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4962 		op->addr.mem.ea =
4963 			register_address(ctxt, VCPU_REGS_RSI);
4964 		op->addr.mem.seg = ctxt->seg_override;
4965 		op->val = 0;
4966 		op->count = 1;
4967 		break;
4968 	case OpXLat:
4969 		op->type = OP_MEM;
4970 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4971 		op->addr.mem.ea =
4972 			address_mask(ctxt,
4973 				reg_read(ctxt, VCPU_REGS_RBX) +
4974 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
4975 		op->addr.mem.seg = ctxt->seg_override;
4976 		op->val = 0;
4977 		break;
4978 	case OpImmFAddr:
4979 		op->type = OP_IMM;
4980 		op->addr.mem.ea = ctxt->_eip;
4981 		op->bytes = ctxt->op_bytes + 2;
4982 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
4983 		break;
4984 	case OpMemFAddr:
4985 		ctxt->memop.bytes = ctxt->op_bytes + 2;
4986 		goto mem_common;
4987 	case OpES:
4988 		op->type = OP_IMM;
4989 		op->val = VCPU_SREG_ES;
4990 		break;
4991 	case OpCS:
4992 		op->type = OP_IMM;
4993 		op->val = VCPU_SREG_CS;
4994 		break;
4995 	case OpSS:
4996 		op->type = OP_IMM;
4997 		op->val = VCPU_SREG_SS;
4998 		break;
4999 	case OpDS:
5000 		op->type = OP_IMM;
5001 		op->val = VCPU_SREG_DS;
5002 		break;
5003 	case OpFS:
5004 		op->type = OP_IMM;
5005 		op->val = VCPU_SREG_FS;
5006 		break;
5007 	case OpGS:
5008 		op->type = OP_IMM;
5009 		op->val = VCPU_SREG_GS;
5010 		break;
5011 	case OpImplicit:
5012 		/* Special instructions do their own operand decoding. */
5013 	default:
5014 		op->type = OP_NONE; /* Disable writeback. */
5015 		break;
5016 	}
5017 
5018 done:
5019 	return rc;
5020 }
5021 
5022 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
5023 {
5024 	int rc = X86EMUL_CONTINUE;
5025 	int mode = ctxt->mode;
5026 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
5027 	bool op_prefix = false;
5028 	bool has_seg_override = false;
5029 	struct opcode opcode;
5030 	u16 dummy;
5031 	struct desc_struct desc;
5032 
5033 	ctxt->memop.type = OP_NONE;
5034 	ctxt->memopp = NULL;
5035 	ctxt->_eip = ctxt->eip;
5036 	ctxt->fetch.ptr = ctxt->fetch.data;
5037 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
5038 	ctxt->opcode_len = 1;
5039 	ctxt->intercept = x86_intercept_none;
5040 	if (insn_len > 0)
5041 		memcpy(ctxt->fetch.data, insn, insn_len);
5042 	else {
5043 		rc = __do_insn_fetch_bytes(ctxt, 1);
5044 		if (rc != X86EMUL_CONTINUE)
5045 			goto done;
5046 	}
5047 
5048 	switch (mode) {
5049 	case X86EMUL_MODE_REAL:
5050 	case X86EMUL_MODE_VM86:
5051 		def_op_bytes = def_ad_bytes = 2;
5052 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
5053 		if (desc.d)
5054 			def_op_bytes = def_ad_bytes = 4;
5055 		break;
5056 	case X86EMUL_MODE_PROT16:
5057 		def_op_bytes = def_ad_bytes = 2;
5058 		break;
5059 	case X86EMUL_MODE_PROT32:
5060 		def_op_bytes = def_ad_bytes = 4;
5061 		break;
5062 #ifdef CONFIG_X86_64
5063 	case X86EMUL_MODE_PROT64:
5064 		def_op_bytes = 4;
5065 		def_ad_bytes = 8;
5066 		break;
5067 #endif
5068 	default:
5069 		return EMULATION_FAILED;
5070 	}
5071 
5072 	ctxt->op_bytes = def_op_bytes;
5073 	ctxt->ad_bytes = def_ad_bytes;
5074 
5075 	/* Legacy prefixes. */
5076 	for (;;) {
5077 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5078 		case 0x66:	/* operand-size override */
5079 			op_prefix = true;
5080 			/* switch between 2/4 bytes */
5081 			ctxt->op_bytes = def_op_bytes ^ 6;
5082 			break;
5083 		case 0x67:	/* address-size override */
5084 			if (mode == X86EMUL_MODE_PROT64)
5085 				/* switch between 4/8 bytes */
5086 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5087 			else
5088 				/* switch between 2/4 bytes */
5089 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5090 			break;
5091 		case 0x26:	/* ES override */
5092 			has_seg_override = true;
5093 			ctxt->seg_override = VCPU_SREG_ES;
5094 			break;
5095 		case 0x2e:	/* CS override */
5096 			has_seg_override = true;
5097 			ctxt->seg_override = VCPU_SREG_CS;
5098 			break;
5099 		case 0x36:	/* SS override */
5100 			has_seg_override = true;
5101 			ctxt->seg_override = VCPU_SREG_SS;
5102 			break;
5103 		case 0x3e:	/* DS override */
5104 			has_seg_override = true;
5105 			ctxt->seg_override = VCPU_SREG_DS;
5106 			break;
5107 		case 0x64:	/* FS override */
5108 			has_seg_override = true;
5109 			ctxt->seg_override = VCPU_SREG_FS;
5110 			break;
5111 		case 0x65:	/* GS override */
5112 			has_seg_override = true;
5113 			ctxt->seg_override = VCPU_SREG_GS;
5114 			break;
5115 		case 0x40 ... 0x4f: /* REX */
5116 			if (mode != X86EMUL_MODE_PROT64)
5117 				goto done_prefixes;
5118 			ctxt->rex_prefix = ctxt->b;
5119 			continue;
5120 		case 0xf0:	/* LOCK */
5121 			ctxt->lock_prefix = 1;
5122 			break;
5123 		case 0xf2:	/* REPNE/REPNZ */
5124 		case 0xf3:	/* REP/REPE/REPZ */
5125 			ctxt->rep_prefix = ctxt->b;
5126 			break;
5127 		default:
5128 			goto done_prefixes;
5129 		}
5130 
5131 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5132 
5133 		ctxt->rex_prefix = 0;
5134 	}
5135 
5136 done_prefixes:
5137 
5138 	/* REX prefix. */
5139 	if (ctxt->rex_prefix & 8)
5140 		ctxt->op_bytes = 8;	/* REX.W */
5141 
5142 	/* Opcode byte(s). */
5143 	opcode = opcode_table[ctxt->b];
5144 	/* Two-byte opcode? */
5145 	if (ctxt->b == 0x0f) {
5146 		ctxt->opcode_len = 2;
5147 		ctxt->b = insn_fetch(u8, ctxt);
5148 		opcode = twobyte_table[ctxt->b];
5149 
5150 		/* 0F_38 opcode map */
5151 		if (ctxt->b == 0x38) {
5152 			ctxt->opcode_len = 3;
5153 			ctxt->b = insn_fetch(u8, ctxt);
5154 			opcode = opcode_map_0f_38[ctxt->b];
5155 		}
5156 	}
5157 	ctxt->d = opcode.flags;
5158 
5159 	if (ctxt->d & ModRM)
5160 		ctxt->modrm = insn_fetch(u8, ctxt);
5161 
5162 	/* vex-prefix instructions are not implemented */
5163 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5164 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5165 		ctxt->d = NotImpl;
5166 	}
5167 
5168 	while (ctxt->d & GroupMask) {
5169 		switch (ctxt->d & GroupMask) {
5170 		case Group:
5171 			goffset = (ctxt->modrm >> 3) & 7;
5172 			opcode = opcode.u.group[goffset];
5173 			break;
5174 		case GroupDual:
5175 			goffset = (ctxt->modrm >> 3) & 7;
5176 			if ((ctxt->modrm >> 6) == 3)
5177 				opcode = opcode.u.gdual->mod3[goffset];
5178 			else
5179 				opcode = opcode.u.gdual->mod012[goffset];
5180 			break;
5181 		case RMExt:
5182 			goffset = ctxt->modrm & 7;
5183 			opcode = opcode.u.group[goffset];
5184 			break;
5185 		case Prefix:
5186 			if (ctxt->rep_prefix && op_prefix)
5187 				return EMULATION_FAILED;
5188 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5189 			switch (simd_prefix) {
5190 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5191 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5192 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5193 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5194 			}
5195 			break;
5196 		case Escape:
5197 			if (ctxt->modrm > 0xbf) {
5198 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
5199 				u32 index = array_index_nospec(
5200 					ctxt->modrm - 0xc0, size);
5201 
5202 				opcode = opcode.u.esc->high[index];
5203 			} else {
5204 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5205 			}
5206 			break;
5207 		case InstrDual:
5208 			if ((ctxt->modrm >> 6) == 3)
5209 				opcode = opcode.u.idual->mod3;
5210 			else
5211 				opcode = opcode.u.idual->mod012;
5212 			break;
5213 		case ModeDual:
5214 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5215 				opcode = opcode.u.mdual->mode64;
5216 			else
5217 				opcode = opcode.u.mdual->mode32;
5218 			break;
5219 		default:
5220 			return EMULATION_FAILED;
5221 		}
5222 
5223 		ctxt->d &= ~(u64)GroupMask;
5224 		ctxt->d |= opcode.flags;
5225 	}
5226 
5227 	/* Unrecognised? */
5228 	if (ctxt->d == 0)
5229 		return EMULATION_FAILED;
5230 
5231 	ctxt->execute = opcode.u.execute;
5232 
5233 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
5234 	    likely(!(ctxt->d & EmulateOnUD)))
5235 		return EMULATION_FAILED;
5236 
5237 	if (unlikely(ctxt->d &
5238 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5239 	     No16))) {
5240 		/*
5241 		 * These are copied unconditionally here, and checked unconditionally
5242 		 * in x86_emulate_insn.
5243 		 */
5244 		ctxt->check_perm = opcode.check_perm;
5245 		ctxt->intercept = opcode.intercept;
5246 
5247 		if (ctxt->d & NotImpl)
5248 			return EMULATION_FAILED;
5249 
5250 		if (mode == X86EMUL_MODE_PROT64) {
5251 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5252 				ctxt->op_bytes = 8;
5253 			else if (ctxt->d & NearBranch)
5254 				ctxt->op_bytes = 8;
5255 		}
5256 
5257 		if (ctxt->d & Op3264) {
5258 			if (mode == X86EMUL_MODE_PROT64)
5259 				ctxt->op_bytes = 8;
5260 			else
5261 				ctxt->op_bytes = 4;
5262 		}
5263 
5264 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5265 			ctxt->op_bytes = 4;
5266 
5267 		if (ctxt->d & Sse)
5268 			ctxt->op_bytes = 16;
5269 		else if (ctxt->d & Mmx)
5270 			ctxt->op_bytes = 8;
5271 	}
5272 
5273 	/* ModRM and SIB bytes. */
5274 	if (ctxt->d & ModRM) {
5275 		rc = decode_modrm(ctxt, &ctxt->memop);
5276 		if (!has_seg_override) {
5277 			has_seg_override = true;
5278 			ctxt->seg_override = ctxt->modrm_seg;
5279 		}
5280 	} else if (ctxt->d & MemAbs)
5281 		rc = decode_abs(ctxt, &ctxt->memop);
5282 	if (rc != X86EMUL_CONTINUE)
5283 		goto done;
5284 
5285 	if (!has_seg_override)
5286 		ctxt->seg_override = VCPU_SREG_DS;
5287 
5288 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5289 
5290 	/*
5291 	 * Decode and fetch the source operand: register, memory
5292 	 * or immediate.
5293 	 */
5294 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5295 	if (rc != X86EMUL_CONTINUE)
5296 		goto done;
5297 
5298 	/*
5299 	 * Decode and fetch the second source operand: register, memory
5300 	 * or immediate.
5301 	 */
5302 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5303 	if (rc != X86EMUL_CONTINUE)
5304 		goto done;
5305 
5306 	/* Decode and fetch the destination operand: register or memory. */
5307 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5308 
5309 	if (ctxt->rip_relative && likely(ctxt->memopp))
5310 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5311 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5312 
5313 done:
5314 	if (rc == X86EMUL_PROPAGATE_FAULT)
5315 		ctxt->have_exception = true;
5316 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5317 }
5318 
5319 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5320 {
5321 	return ctxt->d & PageTable;
5322 }
5323 
5324 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5325 {
5326 	/* The second termination condition only applies for REPE
5327 	 * and REPNE. Test if the repeat string operation prefix is
5328 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5329 	 * corresponding termination condition according to:
5330 	 * 	- if REPE/REPZ and ZF = 0 then done
5331 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5332 	 */
5333 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5334 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5335 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5336 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5337 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5338 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5339 		return true;
5340 
5341 	return false;
5342 }
5343 
5344 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5345 {
5346 	int rc;
5347 
5348 	kvm_fpu_get();
5349 	rc = asm_safe("fwait");
5350 	kvm_fpu_put();
5351 
5352 	if (unlikely(rc != X86EMUL_CONTINUE))
5353 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5354 
5355 	return X86EMUL_CONTINUE;
5356 }
5357 
5358 static void fetch_possible_mmx_operand(struct operand *op)
5359 {
5360 	if (op->type == OP_MM)
5361 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5362 }
5363 
5364 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5365 {
5366 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5367 
5368 	if (!(ctxt->d & ByteOp))
5369 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5370 
5371 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5372 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5373 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5374 	    : "c"(ctxt->src2.val));
5375 
5376 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5377 	if (!fop) /* exception is returned in fop variable */
5378 		return emulate_de(ctxt);
5379 	return X86EMUL_CONTINUE;
5380 }
5381 
5382 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5383 {
5384 	memset(&ctxt->rip_relative, 0,
5385 	       (void *)&ctxt->modrm - (void *)&ctxt->rip_relative);
5386 
5387 	ctxt->io_read.pos = 0;
5388 	ctxt->io_read.end = 0;
5389 	ctxt->mem_read.end = 0;
5390 }
5391 
5392 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5393 {
5394 	const struct x86_emulate_ops *ops = ctxt->ops;
5395 	int rc = X86EMUL_CONTINUE;
5396 	int saved_dst_type = ctxt->dst.type;
5397 	unsigned emul_flags;
5398 
5399 	ctxt->mem_read.pos = 0;
5400 
5401 	/* LOCK prefix is allowed only with some instructions */
5402 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5403 		rc = emulate_ud(ctxt);
5404 		goto done;
5405 	}
5406 
5407 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5408 		rc = emulate_ud(ctxt);
5409 		goto done;
5410 	}
5411 
5412 	emul_flags = ctxt->ops->get_hflags(ctxt);
5413 	if (unlikely(ctxt->d &
5414 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5415 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5416 				(ctxt->d & Undefined)) {
5417 			rc = emulate_ud(ctxt);
5418 			goto done;
5419 		}
5420 
5421 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5422 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5423 			rc = emulate_ud(ctxt);
5424 			goto done;
5425 		}
5426 
5427 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5428 			rc = emulate_nm(ctxt);
5429 			goto done;
5430 		}
5431 
5432 		if (ctxt->d & Mmx) {
5433 			rc = flush_pending_x87_faults(ctxt);
5434 			if (rc != X86EMUL_CONTINUE)
5435 				goto done;
5436 			/*
5437 			 * Now that we know the fpu is exception safe, we can fetch
5438 			 * operands from it.
5439 			 */
5440 			fetch_possible_mmx_operand(&ctxt->src);
5441 			fetch_possible_mmx_operand(&ctxt->src2);
5442 			if (!(ctxt->d & Mov))
5443 				fetch_possible_mmx_operand(&ctxt->dst);
5444 		}
5445 
5446 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5447 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5448 						      X86_ICPT_PRE_EXCEPT);
5449 			if (rc != X86EMUL_CONTINUE)
5450 				goto done;
5451 		}
5452 
5453 		/* Instruction can only be executed in protected mode */
5454 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5455 			rc = emulate_ud(ctxt);
5456 			goto done;
5457 		}
5458 
5459 		/* Privileged instruction can be executed only in CPL=0 */
5460 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5461 			if (ctxt->d & PrivUD)
5462 				rc = emulate_ud(ctxt);
5463 			else
5464 				rc = emulate_gp(ctxt, 0);
5465 			goto done;
5466 		}
5467 
5468 		/* Do instruction specific permission checks */
5469 		if (ctxt->d & CheckPerm) {
5470 			rc = ctxt->check_perm(ctxt);
5471 			if (rc != X86EMUL_CONTINUE)
5472 				goto done;
5473 		}
5474 
5475 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5476 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5477 						      X86_ICPT_POST_EXCEPT);
5478 			if (rc != X86EMUL_CONTINUE)
5479 				goto done;
5480 		}
5481 
5482 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5483 			/* All REP prefixes have the same first termination condition */
5484 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5485 				string_registers_quirk(ctxt);
5486 				ctxt->eip = ctxt->_eip;
5487 				ctxt->eflags &= ~X86_EFLAGS_RF;
5488 				goto done;
5489 			}
5490 		}
5491 	}
5492 
5493 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5494 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5495 				    ctxt->src.valptr, ctxt->src.bytes);
5496 		if (rc != X86EMUL_CONTINUE)
5497 			goto done;
5498 		ctxt->src.orig_val64 = ctxt->src.val64;
5499 	}
5500 
5501 	if (ctxt->src2.type == OP_MEM) {
5502 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5503 				    &ctxt->src2.val, ctxt->src2.bytes);
5504 		if (rc != X86EMUL_CONTINUE)
5505 			goto done;
5506 	}
5507 
5508 	if ((ctxt->d & DstMask) == ImplicitOps)
5509 		goto special_insn;
5510 
5511 
5512 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5513 		/* optimisation - avoid slow emulated read if Mov */
5514 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5515 				   &ctxt->dst.val, ctxt->dst.bytes);
5516 		if (rc != X86EMUL_CONTINUE) {
5517 			if (!(ctxt->d & NoWrite) &&
5518 			    rc == X86EMUL_PROPAGATE_FAULT &&
5519 			    ctxt->exception.vector == PF_VECTOR)
5520 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5521 			goto done;
5522 		}
5523 	}
5524 	/* Copy full 64-bit value for CMPXCHG8B.  */
5525 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5526 
5527 special_insn:
5528 
5529 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5530 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5531 					      X86_ICPT_POST_MEMACCESS);
5532 		if (rc != X86EMUL_CONTINUE)
5533 			goto done;
5534 	}
5535 
5536 	if (ctxt->rep_prefix && (ctxt->d & String))
5537 		ctxt->eflags |= X86_EFLAGS_RF;
5538 	else
5539 		ctxt->eflags &= ~X86_EFLAGS_RF;
5540 
5541 	if (ctxt->execute) {
5542 		if (ctxt->d & Fastop)
5543 			rc = fastop(ctxt, ctxt->fop);
5544 		else
5545 			rc = ctxt->execute(ctxt);
5546 		if (rc != X86EMUL_CONTINUE)
5547 			goto done;
5548 		goto writeback;
5549 	}
5550 
5551 	if (ctxt->opcode_len == 2)
5552 		goto twobyte_insn;
5553 	else if (ctxt->opcode_len == 3)
5554 		goto threebyte_insn;
5555 
5556 	switch (ctxt->b) {
5557 	case 0x70 ... 0x7f: /* jcc (short) */
5558 		if (test_cc(ctxt->b, ctxt->eflags))
5559 			rc = jmp_rel(ctxt, ctxt->src.val);
5560 		break;
5561 	case 0x8d: /* lea r16/r32, m */
5562 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5563 		break;
5564 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5565 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5566 			ctxt->dst.type = OP_NONE;
5567 		else
5568 			rc = em_xchg(ctxt);
5569 		break;
5570 	case 0x98: /* cbw/cwde/cdqe */
5571 		switch (ctxt->op_bytes) {
5572 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5573 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5574 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5575 		}
5576 		break;
5577 	case 0xcc:		/* int3 */
5578 		rc = emulate_int(ctxt, 3);
5579 		break;
5580 	case 0xcd:		/* int n */
5581 		rc = emulate_int(ctxt, ctxt->src.val);
5582 		break;
5583 	case 0xce:		/* into */
5584 		if (ctxt->eflags & X86_EFLAGS_OF)
5585 			rc = emulate_int(ctxt, 4);
5586 		break;
5587 	case 0xe9: /* jmp rel */
5588 	case 0xeb: /* jmp rel short */
5589 		rc = jmp_rel(ctxt, ctxt->src.val);
5590 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5591 		break;
5592 	case 0xf4:              /* hlt */
5593 		ctxt->ops->halt(ctxt);
5594 		break;
5595 	case 0xf5:	/* cmc */
5596 		/* complement carry flag from eflags reg */
5597 		ctxt->eflags ^= X86_EFLAGS_CF;
5598 		break;
5599 	case 0xf8: /* clc */
5600 		ctxt->eflags &= ~X86_EFLAGS_CF;
5601 		break;
5602 	case 0xf9: /* stc */
5603 		ctxt->eflags |= X86_EFLAGS_CF;
5604 		break;
5605 	case 0xfc: /* cld */
5606 		ctxt->eflags &= ~X86_EFLAGS_DF;
5607 		break;
5608 	case 0xfd: /* std */
5609 		ctxt->eflags |= X86_EFLAGS_DF;
5610 		break;
5611 	default:
5612 		goto cannot_emulate;
5613 	}
5614 
5615 	if (rc != X86EMUL_CONTINUE)
5616 		goto done;
5617 
5618 writeback:
5619 	if (ctxt->d & SrcWrite) {
5620 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5621 		rc = writeback(ctxt, &ctxt->src);
5622 		if (rc != X86EMUL_CONTINUE)
5623 			goto done;
5624 	}
5625 	if (!(ctxt->d & NoWrite)) {
5626 		rc = writeback(ctxt, &ctxt->dst);
5627 		if (rc != X86EMUL_CONTINUE)
5628 			goto done;
5629 	}
5630 
5631 	/*
5632 	 * restore dst type in case the decoding will be reused
5633 	 * (happens for string instruction )
5634 	 */
5635 	ctxt->dst.type = saved_dst_type;
5636 
5637 	if ((ctxt->d & SrcMask) == SrcSI)
5638 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5639 
5640 	if ((ctxt->d & DstMask) == DstDI)
5641 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5642 
5643 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5644 		unsigned int count;
5645 		struct read_cache *r = &ctxt->io_read;
5646 		if ((ctxt->d & SrcMask) == SrcSI)
5647 			count = ctxt->src.count;
5648 		else
5649 			count = ctxt->dst.count;
5650 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5651 
5652 		if (!string_insn_completed(ctxt)) {
5653 			/*
5654 			 * Re-enter guest when pio read ahead buffer is empty
5655 			 * or, if it is not used, after each 1024 iteration.
5656 			 */
5657 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5658 			    (r->end == 0 || r->end != r->pos)) {
5659 				/*
5660 				 * Reset read cache. Usually happens before
5661 				 * decode, but since instruction is restarted
5662 				 * we have to do it here.
5663 				 */
5664 				ctxt->mem_read.end = 0;
5665 				writeback_registers(ctxt);
5666 				return EMULATION_RESTART;
5667 			}
5668 			goto done; /* skip rip writeback */
5669 		}
5670 		ctxt->eflags &= ~X86_EFLAGS_RF;
5671 	}
5672 
5673 	ctxt->eip = ctxt->_eip;
5674 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5675 		ctxt->eip = (u32)ctxt->_eip;
5676 
5677 done:
5678 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5679 		WARN_ON(ctxt->exception.vector > 0x1f);
5680 		ctxt->have_exception = true;
5681 	}
5682 	if (rc == X86EMUL_INTERCEPTED)
5683 		return EMULATION_INTERCEPTED;
5684 
5685 	if (rc == X86EMUL_CONTINUE)
5686 		writeback_registers(ctxt);
5687 
5688 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5689 
5690 twobyte_insn:
5691 	switch (ctxt->b) {
5692 	case 0x09:		/* wbinvd */
5693 		(ctxt->ops->wbinvd)(ctxt);
5694 		break;
5695 	case 0x08:		/* invd */
5696 	case 0x0d:		/* GrpP (prefetch) */
5697 	case 0x18:		/* Grp16 (prefetch/nop) */
5698 	case 0x1f:		/* nop */
5699 		break;
5700 	case 0x20: /* mov cr, reg */
5701 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5702 		break;
5703 	case 0x21: /* mov from dr to reg */
5704 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5705 		break;
5706 	case 0x40 ... 0x4f:	/* cmov */
5707 		if (test_cc(ctxt->b, ctxt->eflags))
5708 			ctxt->dst.val = ctxt->src.val;
5709 		else if (ctxt->op_bytes != 4)
5710 			ctxt->dst.type = OP_NONE; /* no writeback */
5711 		break;
5712 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5713 		if (test_cc(ctxt->b, ctxt->eflags))
5714 			rc = jmp_rel(ctxt, ctxt->src.val);
5715 		break;
5716 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5717 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5718 		break;
5719 	case 0xb6 ... 0xb7:	/* movzx */
5720 		ctxt->dst.bytes = ctxt->op_bytes;
5721 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5722 						       : (u16) ctxt->src.val;
5723 		break;
5724 	case 0xbe ... 0xbf:	/* movsx */
5725 		ctxt->dst.bytes = ctxt->op_bytes;
5726 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5727 							(s16) ctxt->src.val;
5728 		break;
5729 	default:
5730 		goto cannot_emulate;
5731 	}
5732 
5733 threebyte_insn:
5734 
5735 	if (rc != X86EMUL_CONTINUE)
5736 		goto done;
5737 
5738 	goto writeback;
5739 
5740 cannot_emulate:
5741 	return EMULATION_FAILED;
5742 }
5743 
5744 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5745 {
5746 	invalidate_registers(ctxt);
5747 }
5748 
5749 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5750 {
5751 	writeback_registers(ctxt);
5752 }
5753 
5754 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5755 {
5756 	if (ctxt->rep_prefix && (ctxt->d & String))
5757 		return false;
5758 
5759 	if (ctxt->d & TwoMemOp)
5760 		return false;
5761 
5762 	return true;
5763 }
5764