xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision fbb6b31a)
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 #include <asm/ibt.h>
28 
29 #include "x86.h"
30 #include "tss.h"
31 #include "mmu.h"
32 #include "pmu.h"
33 
34 /*
35  * Operand types
36  */
37 #define OpNone             0ull
38 #define OpImplicit         1ull  /* No generic decode */
39 #define OpReg              2ull  /* Register */
40 #define OpMem              3ull  /* Memory */
41 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
42 #define OpDI               5ull  /* ES:DI/EDI/RDI */
43 #define OpMem64            6ull  /* Memory, 64-bit */
44 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
45 #define OpDX               8ull  /* DX register */
46 #define OpCL               9ull  /* CL register (for shifts) */
47 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
48 #define OpOne             11ull  /* Implied 1 */
49 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
50 #define OpMem16           13ull  /* Memory operand (16-bit). */
51 #define OpMem32           14ull  /* Memory operand (32-bit). */
52 #define OpImmU            15ull  /* Immediate operand, zero extended */
53 #define OpSI              16ull  /* SI/ESI/RSI */
54 #define OpImmFAddr        17ull  /* Immediate far address */
55 #define OpMemFAddr        18ull  /* Far address in memory */
56 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
57 #define OpES              20ull  /* ES */
58 #define OpCS              21ull  /* CS */
59 #define OpSS              22ull  /* SS */
60 #define OpDS              23ull  /* DS */
61 #define OpFS              24ull  /* FS */
62 #define OpGS              25ull  /* GS */
63 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
64 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
65 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
66 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
67 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
68 
69 #define OpBits             5  /* Width of operand field */
70 #define OpMask             ((1ull << OpBits) - 1)
71 
72 /*
73  * Opcode effective-address decode tables.
74  * Note that we only emulate instructions that have at least one memory
75  * operand (excluding implicit stack references). We assume that stack
76  * references and instruction fetches will never occur in special memory
77  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
78  * not be handled.
79  */
80 
81 /* Operand sizes: 8-bit operands or specified/overridden size. */
82 #define ByteOp      (1<<0)	/* 8-bit operands. */
83 /* Destination operand type. */
84 #define DstShift    1
85 #define ImplicitOps (OpImplicit << DstShift)
86 #define DstReg      (OpReg << DstShift)
87 #define DstMem      (OpMem << DstShift)
88 #define DstAcc      (OpAcc << DstShift)
89 #define DstDI       (OpDI << DstShift)
90 #define DstMem64    (OpMem64 << DstShift)
91 #define DstMem16    (OpMem16 << DstShift)
92 #define DstImmUByte (OpImmUByte << DstShift)
93 #define DstDX       (OpDX << DstShift)
94 #define DstAccLo    (OpAccLo << DstShift)
95 #define DstMask     (OpMask << DstShift)
96 /* Source operand type. */
97 #define SrcShift    6
98 #define SrcNone     (OpNone << SrcShift)
99 #define SrcReg      (OpReg << SrcShift)
100 #define SrcMem      (OpMem << SrcShift)
101 #define SrcMem16    (OpMem16 << SrcShift)
102 #define SrcMem32    (OpMem32 << SrcShift)
103 #define SrcImm      (OpImm << SrcShift)
104 #define SrcImmByte  (OpImmByte << SrcShift)
105 #define SrcOne      (OpOne << SrcShift)
106 #define SrcImmUByte (OpImmUByte << SrcShift)
107 #define SrcImmU     (OpImmU << SrcShift)
108 #define SrcSI       (OpSI << SrcShift)
109 #define SrcXLat     (OpXLat << SrcShift)
110 #define SrcImmFAddr (OpImmFAddr << SrcShift)
111 #define SrcMemFAddr (OpMemFAddr << SrcShift)
112 #define SrcAcc      (OpAcc << SrcShift)
113 #define SrcImmU16   (OpImmU16 << SrcShift)
114 #define SrcImm64    (OpImm64 << SrcShift)
115 #define SrcDX       (OpDX << SrcShift)
116 #define SrcMem8     (OpMem8 << SrcShift)
117 #define SrcAccHi    (OpAccHi << SrcShift)
118 #define SrcMask     (OpMask << SrcShift)
119 #define BitOp       (1<<11)
120 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
121 #define String      (1<<13)     /* String instruction (rep capable) */
122 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
123 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
124 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
125 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
126 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
127 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
128 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
129 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
130 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
131 #define Sse         (1<<18)     /* SSE Vector instruction */
132 /* Generic ModRM decode. */
133 #define ModRM       (1<<19)
134 /* Destination is only written; never read. */
135 #define Mov         (1<<20)
136 /* Misc flags */
137 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
138 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
139 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
140 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
141 #define Undefined   (1<<25) /* No Such Instruction */
142 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
143 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
144 #define No64	    (1<<28)
145 #define PageTable   (1 << 29)   /* instruction used to write page table */
146 #define NotImpl     (1 << 30)   /* instruction is not implemented */
147 /* Source 2 operand type */
148 #define Src2Shift   (31)
149 #define Src2None    (OpNone << Src2Shift)
150 #define Src2Mem     (OpMem << Src2Shift)
151 #define Src2CL      (OpCL << Src2Shift)
152 #define Src2ImmByte (OpImmByte << Src2Shift)
153 #define Src2One     (OpOne << Src2Shift)
154 #define Src2Imm     (OpImm << Src2Shift)
155 #define Src2ES      (OpES << Src2Shift)
156 #define Src2CS      (OpCS << Src2Shift)
157 #define Src2SS      (OpSS << Src2Shift)
158 #define Src2DS      (OpDS << Src2Shift)
159 #define Src2FS      (OpFS << Src2Shift)
160 #define Src2GS      (OpGS << Src2Shift)
161 #define Src2Mask    (OpMask << Src2Shift)
162 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
163 #define AlignMask   ((u64)7 << 41)
164 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
165 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
166 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
167 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
168 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
169 #define NoWrite     ((u64)1 << 45)  /* No writeback */
170 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
171 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
172 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
173 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
174 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
175 #define NearBranch  ((u64)1 << 52)  /* Near branches */
176 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
177 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
178 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
179 #define IsBranch    ((u64)1 << 56)  /* Instruction is considered a branch. */
180 
181 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
182 
183 #define X2(x...) x, x
184 #define X3(x...) X2(x), x
185 #define X4(x...) X2(x), X2(x)
186 #define X5(x...) X4(x), x
187 #define X6(x...) X4(x), X2(x)
188 #define X7(x...) X4(x), X3(x)
189 #define X8(x...) X4(x), X4(x)
190 #define X16(x...) X8(x), X8(x)
191 
192 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1)
193 #define FASTOP_SIZE (8 * (1 + HAS_KERNEL_IBT))
194 
195 struct opcode {
196 	u64 flags;
197 	u8 intercept;
198 	u8 pad[7];
199 	union {
200 		int (*execute)(struct x86_emulate_ctxt *ctxt);
201 		const struct opcode *group;
202 		const struct group_dual *gdual;
203 		const struct gprefix *gprefix;
204 		const struct escape *esc;
205 		const struct instr_dual *idual;
206 		const struct mode_dual *mdual;
207 		void (*fastop)(struct fastop *fake);
208 	} u;
209 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
210 };
211 
212 struct group_dual {
213 	struct opcode mod012[8];
214 	struct opcode mod3[8];
215 };
216 
217 struct gprefix {
218 	struct opcode pfx_no;
219 	struct opcode pfx_66;
220 	struct opcode pfx_f2;
221 	struct opcode pfx_f3;
222 };
223 
224 struct escape {
225 	struct opcode op[8];
226 	struct opcode high[64];
227 };
228 
229 struct instr_dual {
230 	struct opcode mod012;
231 	struct opcode mod3;
232 };
233 
234 struct mode_dual {
235 	struct opcode mode32;
236 	struct opcode mode64;
237 };
238 
239 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
240 
241 enum x86_transfer_type {
242 	X86_TRANSFER_NONE,
243 	X86_TRANSFER_CALL_JMP,
244 	X86_TRANSFER_RET,
245 	X86_TRANSFER_TASK_SWITCH,
246 };
247 
248 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
249 {
250 	if (!(ctxt->regs_valid & (1 << nr))) {
251 		ctxt->regs_valid |= 1 << nr;
252 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
253 	}
254 	return ctxt->_regs[nr];
255 }
256 
257 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
258 {
259 	ctxt->regs_valid |= 1 << nr;
260 	ctxt->regs_dirty |= 1 << nr;
261 	return &ctxt->_regs[nr];
262 }
263 
264 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
265 {
266 	reg_read(ctxt, nr);
267 	return reg_write(ctxt, nr);
268 }
269 
270 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
271 {
272 	unsigned reg;
273 
274 	for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16)
275 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
276 }
277 
278 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
279 {
280 	ctxt->regs_dirty = 0;
281 	ctxt->regs_valid = 0;
282 }
283 
284 /*
285  * These EFLAGS bits are restored from saved value during emulation, and
286  * any changes are written back to the saved value after emulation.
287  */
288 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
289 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
290 
291 #ifdef CONFIG_X86_64
292 #define ON64(x) x
293 #else
294 #define ON64(x)
295 #endif
296 
297 /*
298  * fastop functions have a special calling convention:
299  *
300  * dst:    rax        (in/out)
301  * src:    rdx        (in/out)
302  * src2:   rcx        (in)
303  * flags:  rflags     (in/out)
304  * ex:     rsi        (in:fastop pointer, out:zero if exception)
305  *
306  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
307  * different operand sizes can be reached by calculation, rather than a jump
308  * table (which would be bigger than the code).
309  */
310 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
311 
312 #define __FOP_FUNC(name) \
313 	".align " __stringify(FASTOP_SIZE) " \n\t" \
314 	".type " name ", @function \n\t" \
315 	name ":\n\t" \
316 	ASM_ENDBR
317 
318 #define FOP_FUNC(name) \
319 	__FOP_FUNC(#name)
320 
321 #define __FOP_RET(name) \
322 	"11: " ASM_RET \
323 	".size " name ", .-" name "\n\t"
324 
325 #define FOP_RET(name) \
326 	__FOP_RET(#name)
327 
328 #define FOP_START(op) \
329 	extern void em_##op(struct fastop *fake); \
330 	asm(".pushsection .text, \"ax\" \n\t" \
331 	    ".global em_" #op " \n\t" \
332 	    ".align " __stringify(FASTOP_SIZE) " \n\t" \
333 	    "em_" #op ":\n\t"
334 
335 #define FOP_END \
336 	    ".popsection")
337 
338 #define __FOPNOP(name) \
339 	__FOP_FUNC(name) \
340 	__FOP_RET(name)
341 
342 #define FOPNOP() \
343 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
344 
345 #define FOP1E(op,  dst) \
346 	__FOP_FUNC(#op "_" #dst) \
347 	"10: " #op " %" #dst " \n\t" \
348 	__FOP_RET(#op "_" #dst)
349 
350 #define FOP1EEX(op,  dst) \
351 	FOP1E(op, dst) _ASM_EXTABLE_TYPE_REG(10b, 11b, EX_TYPE_ZERO_REG, %%esi)
352 
353 #define FASTOP1(op) \
354 	FOP_START(op) \
355 	FOP1E(op##b, al) \
356 	FOP1E(op##w, ax) \
357 	FOP1E(op##l, eax) \
358 	ON64(FOP1E(op##q, rax))	\
359 	FOP_END
360 
361 /* 1-operand, using src2 (for MUL/DIV r/m) */
362 #define FASTOP1SRC2(op, name) \
363 	FOP_START(name) \
364 	FOP1E(op, cl) \
365 	FOP1E(op, cx) \
366 	FOP1E(op, ecx) \
367 	ON64(FOP1E(op, rcx)) \
368 	FOP_END
369 
370 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
371 #define FASTOP1SRC2EX(op, name) \
372 	FOP_START(name) \
373 	FOP1EEX(op, cl) \
374 	FOP1EEX(op, cx) \
375 	FOP1EEX(op, ecx) \
376 	ON64(FOP1EEX(op, rcx)) \
377 	FOP_END
378 
379 #define FOP2E(op,  dst, src)	   \
380 	__FOP_FUNC(#op "_" #dst "_" #src) \
381 	#op " %" #src ", %" #dst " \n\t" \
382 	__FOP_RET(#op "_" #dst "_" #src)
383 
384 #define FASTOP2(op) \
385 	FOP_START(op) \
386 	FOP2E(op##b, al, dl) \
387 	FOP2E(op##w, ax, dx) \
388 	FOP2E(op##l, eax, edx) \
389 	ON64(FOP2E(op##q, rax, rdx)) \
390 	FOP_END
391 
392 /* 2 operand, word only */
393 #define FASTOP2W(op) \
394 	FOP_START(op) \
395 	FOPNOP() \
396 	FOP2E(op##w, ax, dx) \
397 	FOP2E(op##l, eax, edx) \
398 	ON64(FOP2E(op##q, rax, rdx)) \
399 	FOP_END
400 
401 /* 2 operand, src is CL */
402 #define FASTOP2CL(op) \
403 	FOP_START(op) \
404 	FOP2E(op##b, al, cl) \
405 	FOP2E(op##w, ax, cl) \
406 	FOP2E(op##l, eax, cl) \
407 	ON64(FOP2E(op##q, rax, cl)) \
408 	FOP_END
409 
410 /* 2 operand, src and dest are reversed */
411 #define FASTOP2R(op, name) \
412 	FOP_START(name) \
413 	FOP2E(op##b, dl, al) \
414 	FOP2E(op##w, dx, ax) \
415 	FOP2E(op##l, edx, eax) \
416 	ON64(FOP2E(op##q, rdx, rax)) \
417 	FOP_END
418 
419 #define FOP3E(op,  dst, src, src2) \
420 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
421 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
422 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
423 
424 /* 3-operand, word-only, src2=cl */
425 #define FASTOP3WCL(op) \
426 	FOP_START(op) \
427 	FOPNOP() \
428 	FOP3E(op##w, ax, dx, cl) \
429 	FOP3E(op##l, eax, edx, cl) \
430 	ON64(FOP3E(op##q, rax, rdx, cl)) \
431 	FOP_END
432 
433 /* Special case for SETcc - 1 instruction per cc */
434 
435 /*
436  * Depending on .config the SETcc functions look like:
437  *
438  * ENDBR       [4 bytes; CONFIG_X86_KERNEL_IBT]
439  * SETcc %al   [3 bytes]
440  * RET         [1 byte]
441  * INT3        [1 byte; CONFIG_SLS]
442  *
443  * Which gives possible sizes 4, 5, 8 or 9.  When rounded up to the
444  * next power-of-two alignment they become 4, 8 or 16 resp.
445  */
446 #define SETCC_LENGTH	(ENDBR_INSN_SIZE + 4 + IS_ENABLED(CONFIG_SLS))
447 #define SETCC_ALIGN	(4 << IS_ENABLED(CONFIG_SLS) << HAS_KERNEL_IBT)
448 static_assert(SETCC_LENGTH <= SETCC_ALIGN);
449 
450 #define FOP_SETCC(op) \
451 	".align " __stringify(SETCC_ALIGN) " \n\t" \
452 	".type " #op ", @function \n\t" \
453 	#op ": \n\t" \
454 	ASM_ENDBR \
455 	#op " %al \n\t" \
456 	__FOP_RET(#op)
457 
458 FOP_START(setcc)
459 FOP_SETCC(seto)
460 FOP_SETCC(setno)
461 FOP_SETCC(setc)
462 FOP_SETCC(setnc)
463 FOP_SETCC(setz)
464 FOP_SETCC(setnz)
465 FOP_SETCC(setbe)
466 FOP_SETCC(setnbe)
467 FOP_SETCC(sets)
468 FOP_SETCC(setns)
469 FOP_SETCC(setp)
470 FOP_SETCC(setnp)
471 FOP_SETCC(setl)
472 FOP_SETCC(setnl)
473 FOP_SETCC(setle)
474 FOP_SETCC(setnle)
475 FOP_END;
476 
477 FOP_START(salc)
478 FOP_FUNC(salc)
479 "pushf; sbb %al, %al; popf \n\t"
480 FOP_RET(salc)
481 FOP_END;
482 
483 /*
484  * XXX: inoutclob user must know where the argument is being expanded.
485  *      Relying on CONFIG_CC_HAS_ASM_GOTO would allow us to remove _fault.
486  */
487 #define asm_safe(insn, inoutclob...) \
488 ({ \
489 	int _fault = 0; \
490  \
491 	asm volatile("1:" insn "\n" \
492 	             "2:\n" \
493 		     _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_ONE_REG, %[_fault]) \
494 	             : [_fault] "+r"(_fault) inoutclob ); \
495  \
496 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
497 })
498 
499 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
500 				    enum x86_intercept intercept,
501 				    enum x86_intercept_stage stage)
502 {
503 	struct x86_instruction_info info = {
504 		.intercept  = intercept,
505 		.rep_prefix = ctxt->rep_prefix,
506 		.modrm_mod  = ctxt->modrm_mod,
507 		.modrm_reg  = ctxt->modrm_reg,
508 		.modrm_rm   = ctxt->modrm_rm,
509 		.src_val    = ctxt->src.val64,
510 		.dst_val    = ctxt->dst.val64,
511 		.src_bytes  = ctxt->src.bytes,
512 		.dst_bytes  = ctxt->dst.bytes,
513 		.ad_bytes   = ctxt->ad_bytes,
514 		.next_rip   = ctxt->eip,
515 	};
516 
517 	return ctxt->ops->intercept(ctxt, &info, stage);
518 }
519 
520 static void assign_masked(ulong *dest, ulong src, ulong mask)
521 {
522 	*dest = (*dest & ~mask) | (src & mask);
523 }
524 
525 static void assign_register(unsigned long *reg, u64 val, int bytes)
526 {
527 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
528 	switch (bytes) {
529 	case 1:
530 		*(u8 *)reg = (u8)val;
531 		break;
532 	case 2:
533 		*(u16 *)reg = (u16)val;
534 		break;
535 	case 4:
536 		*reg = (u32)val;
537 		break;	/* 64b: zero-extend */
538 	case 8:
539 		*reg = val;
540 		break;
541 	}
542 }
543 
544 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
545 {
546 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
547 }
548 
549 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
550 {
551 	u16 sel;
552 	struct desc_struct ss;
553 
554 	if (ctxt->mode == X86EMUL_MODE_PROT64)
555 		return ~0UL;
556 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
557 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
558 }
559 
560 static int stack_size(struct x86_emulate_ctxt *ctxt)
561 {
562 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
563 }
564 
565 /* Access/update address held in a register, based on addressing mode. */
566 static inline unsigned long
567 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
568 {
569 	if (ctxt->ad_bytes == sizeof(unsigned long))
570 		return reg;
571 	else
572 		return reg & ad_mask(ctxt);
573 }
574 
575 static inline unsigned long
576 register_address(struct x86_emulate_ctxt *ctxt, int reg)
577 {
578 	return address_mask(ctxt, reg_read(ctxt, reg));
579 }
580 
581 static void masked_increment(ulong *reg, ulong mask, int inc)
582 {
583 	assign_masked(reg, *reg + inc, mask);
584 }
585 
586 static inline void
587 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
588 {
589 	ulong *preg = reg_rmw(ctxt, reg);
590 
591 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
592 }
593 
594 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
595 {
596 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
597 }
598 
599 static u32 desc_limit_scaled(struct desc_struct *desc)
600 {
601 	u32 limit = get_desc_limit(desc);
602 
603 	return desc->g ? (limit << 12) | 0xfff : limit;
604 }
605 
606 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
607 {
608 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
609 		return 0;
610 
611 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
612 }
613 
614 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
615 			     u32 error, bool valid)
616 {
617 	WARN_ON(vec > 0x1f);
618 	ctxt->exception.vector = vec;
619 	ctxt->exception.error_code = error;
620 	ctxt->exception.error_code_valid = valid;
621 	return X86EMUL_PROPAGATE_FAULT;
622 }
623 
624 static int emulate_db(struct x86_emulate_ctxt *ctxt)
625 {
626 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
627 }
628 
629 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
630 {
631 	return emulate_exception(ctxt, GP_VECTOR, err, true);
632 }
633 
634 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
635 {
636 	return emulate_exception(ctxt, SS_VECTOR, err, true);
637 }
638 
639 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
640 {
641 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
642 }
643 
644 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
645 {
646 	return emulate_exception(ctxt, TS_VECTOR, err, true);
647 }
648 
649 static int emulate_de(struct x86_emulate_ctxt *ctxt)
650 {
651 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
652 }
653 
654 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
655 {
656 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
657 }
658 
659 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
660 {
661 	u16 selector;
662 	struct desc_struct desc;
663 
664 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
665 	return selector;
666 }
667 
668 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
669 				 unsigned seg)
670 {
671 	u16 dummy;
672 	u32 base3;
673 	struct desc_struct desc;
674 
675 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
676 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
677 }
678 
679 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
680 {
681 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
682 }
683 
684 static inline bool emul_is_noncanonical_address(u64 la,
685 						struct x86_emulate_ctxt *ctxt)
686 {
687 	return !__is_canonical_address(la, ctxt_virt_addr_bits(ctxt));
688 }
689 
690 /*
691  * x86 defines three classes of vector instructions: explicitly
692  * aligned, explicitly unaligned, and the rest, which change behaviour
693  * depending on whether they're AVX encoded or not.
694  *
695  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
696  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
697  * 512 bytes of data must be aligned to a 16 byte boundary.
698  */
699 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
700 {
701 	u64 alignment = ctxt->d & AlignMask;
702 
703 	if (likely(size < 16))
704 		return 1;
705 
706 	switch (alignment) {
707 	case Unaligned:
708 	case Avx:
709 		return 1;
710 	case Aligned16:
711 		return 16;
712 	case Aligned:
713 	default:
714 		return size;
715 	}
716 }
717 
718 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
719 				       struct segmented_address addr,
720 				       unsigned *max_size, unsigned size,
721 				       bool write, bool fetch,
722 				       enum x86emul_mode mode, ulong *linear)
723 {
724 	struct desc_struct desc;
725 	bool usable;
726 	ulong la;
727 	u32 lim;
728 	u16 sel;
729 	u8  va_bits;
730 
731 	la = seg_base(ctxt, addr.seg) + addr.ea;
732 	*max_size = 0;
733 	switch (mode) {
734 	case X86EMUL_MODE_PROT64:
735 		*linear = la;
736 		va_bits = ctxt_virt_addr_bits(ctxt);
737 		if (!__is_canonical_address(la, va_bits))
738 			goto bad;
739 
740 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
741 		if (size > *max_size)
742 			goto bad;
743 		break;
744 	default:
745 		*linear = la = (u32)la;
746 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
747 						addr.seg);
748 		if (!usable)
749 			goto bad;
750 		/* code segment in protected mode or read-only data segment */
751 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
752 					|| !(desc.type & 2)) && write)
753 			goto bad;
754 		/* unreadable code segment */
755 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
756 			goto bad;
757 		lim = desc_limit_scaled(&desc);
758 		if (!(desc.type & 8) && (desc.type & 4)) {
759 			/* expand-down segment */
760 			if (addr.ea <= lim)
761 				goto bad;
762 			lim = desc.d ? 0xffffffff : 0xffff;
763 		}
764 		if (addr.ea > lim)
765 			goto bad;
766 		if (lim == 0xffffffff)
767 			*max_size = ~0u;
768 		else {
769 			*max_size = (u64)lim + 1 - addr.ea;
770 			if (size > *max_size)
771 				goto bad;
772 		}
773 		break;
774 	}
775 	if (la & (insn_alignment(ctxt, size) - 1))
776 		return emulate_gp(ctxt, 0);
777 	return X86EMUL_CONTINUE;
778 bad:
779 	if (addr.seg == VCPU_SREG_SS)
780 		return emulate_ss(ctxt, 0);
781 	else
782 		return emulate_gp(ctxt, 0);
783 }
784 
785 static int linearize(struct x86_emulate_ctxt *ctxt,
786 		     struct segmented_address addr,
787 		     unsigned size, bool write,
788 		     ulong *linear)
789 {
790 	unsigned max_size;
791 	return __linearize(ctxt, addr, &max_size, size, write, false,
792 			   ctxt->mode, linear);
793 }
794 
795 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
796 			     enum x86emul_mode mode)
797 {
798 	ulong linear;
799 	int rc;
800 	unsigned max_size;
801 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
802 					   .ea = dst };
803 
804 	if (ctxt->op_bytes != sizeof(unsigned long))
805 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
806 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
807 	if (rc == X86EMUL_CONTINUE)
808 		ctxt->_eip = addr.ea;
809 	return rc;
810 }
811 
812 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
813 {
814 	return assign_eip(ctxt, dst, ctxt->mode);
815 }
816 
817 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
818 			  const struct desc_struct *cs_desc)
819 {
820 	enum x86emul_mode mode = ctxt->mode;
821 	int rc;
822 
823 #ifdef CONFIG_X86_64
824 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
825 		if (cs_desc->l) {
826 			u64 efer = 0;
827 
828 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
829 			if (efer & EFER_LMA)
830 				mode = X86EMUL_MODE_PROT64;
831 		} else
832 			mode = X86EMUL_MODE_PROT32; /* temporary value */
833 	}
834 #endif
835 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
836 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
837 	rc = assign_eip(ctxt, dst, mode);
838 	if (rc == X86EMUL_CONTINUE)
839 		ctxt->mode = mode;
840 	return rc;
841 }
842 
843 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
844 {
845 	return assign_eip_near(ctxt, ctxt->_eip + rel);
846 }
847 
848 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear,
849 			      void *data, unsigned size)
850 {
851 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true);
852 }
853 
854 static int linear_write_system(struct x86_emulate_ctxt *ctxt,
855 			       ulong linear, void *data,
856 			       unsigned int size)
857 {
858 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true);
859 }
860 
861 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
862 			      struct segmented_address addr,
863 			      void *data,
864 			      unsigned size)
865 {
866 	int rc;
867 	ulong linear;
868 
869 	rc = linearize(ctxt, addr, size, false, &linear);
870 	if (rc != X86EMUL_CONTINUE)
871 		return rc;
872 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false);
873 }
874 
875 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
876 			       struct segmented_address addr,
877 			       void *data,
878 			       unsigned int size)
879 {
880 	int rc;
881 	ulong linear;
882 
883 	rc = linearize(ctxt, addr, size, true, &linear);
884 	if (rc != X86EMUL_CONTINUE)
885 		return rc;
886 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false);
887 }
888 
889 /*
890  * Prefetch the remaining bytes of the instruction without crossing page
891  * boundary if they are not in fetch_cache yet.
892  */
893 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
894 {
895 	int rc;
896 	unsigned size, max_size;
897 	unsigned long linear;
898 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
899 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
900 					   .ea = ctxt->eip + cur_size };
901 
902 	/*
903 	 * We do not know exactly how many bytes will be needed, and
904 	 * __linearize is expensive, so fetch as much as possible.  We
905 	 * just have to avoid going beyond the 15 byte limit, the end
906 	 * of the segment, or the end of the page.
907 	 *
908 	 * __linearize is called with size 0 so that it does not do any
909 	 * boundary check itself.  Instead, we use max_size to check
910 	 * against op_size.
911 	 */
912 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
913 			 &linear);
914 	if (unlikely(rc != X86EMUL_CONTINUE))
915 		return rc;
916 
917 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
918 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
919 
920 	/*
921 	 * One instruction can only straddle two pages,
922 	 * and one has been loaded at the beginning of
923 	 * x86_decode_insn.  So, if not enough bytes
924 	 * still, we must have hit the 15-byte boundary.
925 	 */
926 	if (unlikely(size < op_size))
927 		return emulate_gp(ctxt, 0);
928 
929 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
930 			      size, &ctxt->exception);
931 	if (unlikely(rc != X86EMUL_CONTINUE))
932 		return rc;
933 	ctxt->fetch.end += size;
934 	return X86EMUL_CONTINUE;
935 }
936 
937 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
938 					       unsigned size)
939 {
940 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
941 
942 	if (unlikely(done_size < size))
943 		return __do_insn_fetch_bytes(ctxt, size - done_size);
944 	else
945 		return X86EMUL_CONTINUE;
946 }
947 
948 /* Fetch next part of the instruction being emulated. */
949 #define insn_fetch(_type, _ctxt)					\
950 ({	_type _x;							\
951 									\
952 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
953 	if (rc != X86EMUL_CONTINUE)					\
954 		goto done;						\
955 	ctxt->_eip += sizeof(_type);					\
956 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
957 	ctxt->fetch.ptr += sizeof(_type);				\
958 	_x;								\
959 })
960 
961 #define insn_fetch_arr(_arr, _size, _ctxt)				\
962 ({									\
963 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
964 	if (rc != X86EMUL_CONTINUE)					\
965 		goto done;						\
966 	ctxt->_eip += (_size);						\
967 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
968 	ctxt->fetch.ptr += (_size);					\
969 })
970 
971 /*
972  * Given the 'reg' portion of a ModRM byte, and a register block, return a
973  * pointer into the block that addresses the relevant register.
974  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
975  */
976 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
977 			     int byteop)
978 {
979 	void *p;
980 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
981 
982 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
983 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
984 	else
985 		p = reg_rmw(ctxt, modrm_reg);
986 	return p;
987 }
988 
989 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
990 			   struct segmented_address addr,
991 			   u16 *size, unsigned long *address, int op_bytes)
992 {
993 	int rc;
994 
995 	if (op_bytes == 2)
996 		op_bytes = 3;
997 	*address = 0;
998 	rc = segmented_read_std(ctxt, addr, size, 2);
999 	if (rc != X86EMUL_CONTINUE)
1000 		return rc;
1001 	addr.ea += 2;
1002 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
1003 	return rc;
1004 }
1005 
1006 FASTOP2(add);
1007 FASTOP2(or);
1008 FASTOP2(adc);
1009 FASTOP2(sbb);
1010 FASTOP2(and);
1011 FASTOP2(sub);
1012 FASTOP2(xor);
1013 FASTOP2(cmp);
1014 FASTOP2(test);
1015 
1016 FASTOP1SRC2(mul, mul_ex);
1017 FASTOP1SRC2(imul, imul_ex);
1018 FASTOP1SRC2EX(div, div_ex);
1019 FASTOP1SRC2EX(idiv, idiv_ex);
1020 
1021 FASTOP3WCL(shld);
1022 FASTOP3WCL(shrd);
1023 
1024 FASTOP2W(imul);
1025 
1026 FASTOP1(not);
1027 FASTOP1(neg);
1028 FASTOP1(inc);
1029 FASTOP1(dec);
1030 
1031 FASTOP2CL(rol);
1032 FASTOP2CL(ror);
1033 FASTOP2CL(rcl);
1034 FASTOP2CL(rcr);
1035 FASTOP2CL(shl);
1036 FASTOP2CL(shr);
1037 FASTOP2CL(sar);
1038 
1039 FASTOP2W(bsf);
1040 FASTOP2W(bsr);
1041 FASTOP2W(bt);
1042 FASTOP2W(bts);
1043 FASTOP2W(btr);
1044 FASTOP2W(btc);
1045 
1046 FASTOP2(xadd);
1047 
1048 FASTOP2R(cmp, cmp_r);
1049 
1050 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1051 {
1052 	/* If src is zero, do not writeback, but update flags */
1053 	if (ctxt->src.val == 0)
1054 		ctxt->dst.type = OP_NONE;
1055 	return fastop(ctxt, em_bsf);
1056 }
1057 
1058 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1059 {
1060 	/* If src is zero, do not writeback, but update flags */
1061 	if (ctxt->src.val == 0)
1062 		ctxt->dst.type = OP_NONE;
1063 	return fastop(ctxt, em_bsr);
1064 }
1065 
1066 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1067 {
1068 	u8 rc;
1069 	void (*fop)(void) = (void *)em_setcc + SETCC_ALIGN * (condition & 0xf);
1070 
1071 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1072 	asm("push %[flags]; popf; " CALL_NOSPEC
1073 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1074 	return rc;
1075 }
1076 
1077 static void fetch_register_operand(struct operand *op)
1078 {
1079 	switch (op->bytes) {
1080 	case 1:
1081 		op->val = *(u8 *)op->addr.reg;
1082 		break;
1083 	case 2:
1084 		op->val = *(u16 *)op->addr.reg;
1085 		break;
1086 	case 4:
1087 		op->val = *(u32 *)op->addr.reg;
1088 		break;
1089 	case 8:
1090 		op->val = *(u64 *)op->addr.reg;
1091 		break;
1092 	}
1093 }
1094 
1095 static int em_fninit(struct x86_emulate_ctxt *ctxt)
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("fninit");
1102 	kvm_fpu_put();
1103 	return X86EMUL_CONTINUE;
1104 }
1105 
1106 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1107 {
1108 	u16 fcw;
1109 
1110 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1111 		return emulate_nm(ctxt);
1112 
1113 	kvm_fpu_get();
1114 	asm volatile("fnstcw %0": "+m"(fcw));
1115 	kvm_fpu_put();
1116 
1117 	ctxt->dst.val = fcw;
1118 
1119 	return X86EMUL_CONTINUE;
1120 }
1121 
1122 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1123 {
1124 	u16 fsw;
1125 
1126 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1127 		return emulate_nm(ctxt);
1128 
1129 	kvm_fpu_get();
1130 	asm volatile("fnstsw %0": "+m"(fsw));
1131 	kvm_fpu_put();
1132 
1133 	ctxt->dst.val = fsw;
1134 
1135 	return X86EMUL_CONTINUE;
1136 }
1137 
1138 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1139 				    struct operand *op)
1140 {
1141 	unsigned reg = ctxt->modrm_reg;
1142 
1143 	if (!(ctxt->d & ModRM))
1144 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1145 
1146 	if (ctxt->d & Sse) {
1147 		op->type = OP_XMM;
1148 		op->bytes = 16;
1149 		op->addr.xmm = reg;
1150 		kvm_read_sse_reg(reg, &op->vec_val);
1151 		return;
1152 	}
1153 	if (ctxt->d & Mmx) {
1154 		reg &= 7;
1155 		op->type = OP_MM;
1156 		op->bytes = 8;
1157 		op->addr.mm = reg;
1158 		return;
1159 	}
1160 
1161 	op->type = OP_REG;
1162 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1163 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1164 
1165 	fetch_register_operand(op);
1166 	op->orig_val = op->val;
1167 }
1168 
1169 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1170 {
1171 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1172 		ctxt->modrm_seg = VCPU_SREG_SS;
1173 }
1174 
1175 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1176 			struct operand *op)
1177 {
1178 	u8 sib;
1179 	int index_reg, base_reg, scale;
1180 	int rc = X86EMUL_CONTINUE;
1181 	ulong modrm_ea = 0;
1182 
1183 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1184 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1185 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1186 
1187 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1188 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1189 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1190 	ctxt->modrm_seg = VCPU_SREG_DS;
1191 
1192 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1193 		op->type = OP_REG;
1194 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1195 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1196 				ctxt->d & ByteOp);
1197 		if (ctxt->d & Sse) {
1198 			op->type = OP_XMM;
1199 			op->bytes = 16;
1200 			op->addr.xmm = ctxt->modrm_rm;
1201 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1202 			return rc;
1203 		}
1204 		if (ctxt->d & Mmx) {
1205 			op->type = OP_MM;
1206 			op->bytes = 8;
1207 			op->addr.mm = ctxt->modrm_rm & 7;
1208 			return rc;
1209 		}
1210 		fetch_register_operand(op);
1211 		return rc;
1212 	}
1213 
1214 	op->type = OP_MEM;
1215 
1216 	if (ctxt->ad_bytes == 2) {
1217 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1218 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1219 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1220 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1221 
1222 		/* 16-bit ModR/M decode. */
1223 		switch (ctxt->modrm_mod) {
1224 		case 0:
1225 			if (ctxt->modrm_rm == 6)
1226 				modrm_ea += insn_fetch(u16, ctxt);
1227 			break;
1228 		case 1:
1229 			modrm_ea += insn_fetch(s8, ctxt);
1230 			break;
1231 		case 2:
1232 			modrm_ea += insn_fetch(u16, ctxt);
1233 			break;
1234 		}
1235 		switch (ctxt->modrm_rm) {
1236 		case 0:
1237 			modrm_ea += bx + si;
1238 			break;
1239 		case 1:
1240 			modrm_ea += bx + di;
1241 			break;
1242 		case 2:
1243 			modrm_ea += bp + si;
1244 			break;
1245 		case 3:
1246 			modrm_ea += bp + di;
1247 			break;
1248 		case 4:
1249 			modrm_ea += si;
1250 			break;
1251 		case 5:
1252 			modrm_ea += di;
1253 			break;
1254 		case 6:
1255 			if (ctxt->modrm_mod != 0)
1256 				modrm_ea += bp;
1257 			break;
1258 		case 7:
1259 			modrm_ea += bx;
1260 			break;
1261 		}
1262 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1263 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1264 			ctxt->modrm_seg = VCPU_SREG_SS;
1265 		modrm_ea = (u16)modrm_ea;
1266 	} else {
1267 		/* 32/64-bit ModR/M decode. */
1268 		if ((ctxt->modrm_rm & 7) == 4) {
1269 			sib = insn_fetch(u8, ctxt);
1270 			index_reg |= (sib >> 3) & 7;
1271 			base_reg |= sib & 7;
1272 			scale = sib >> 6;
1273 
1274 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1275 				modrm_ea += insn_fetch(s32, ctxt);
1276 			else {
1277 				modrm_ea += reg_read(ctxt, base_reg);
1278 				adjust_modrm_seg(ctxt, base_reg);
1279 				/* Increment ESP on POP [ESP] */
1280 				if ((ctxt->d & IncSP) &&
1281 				    base_reg == VCPU_REGS_RSP)
1282 					modrm_ea += ctxt->op_bytes;
1283 			}
1284 			if (index_reg != 4)
1285 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1286 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1287 			modrm_ea += insn_fetch(s32, ctxt);
1288 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1289 				ctxt->rip_relative = 1;
1290 		} else {
1291 			base_reg = ctxt->modrm_rm;
1292 			modrm_ea += reg_read(ctxt, base_reg);
1293 			adjust_modrm_seg(ctxt, base_reg);
1294 		}
1295 		switch (ctxt->modrm_mod) {
1296 		case 1:
1297 			modrm_ea += insn_fetch(s8, ctxt);
1298 			break;
1299 		case 2:
1300 			modrm_ea += insn_fetch(s32, ctxt);
1301 			break;
1302 		}
1303 	}
1304 	op->addr.mem.ea = modrm_ea;
1305 	if (ctxt->ad_bytes != 8)
1306 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1307 
1308 done:
1309 	return rc;
1310 }
1311 
1312 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1313 		      struct operand *op)
1314 {
1315 	int rc = X86EMUL_CONTINUE;
1316 
1317 	op->type = OP_MEM;
1318 	switch (ctxt->ad_bytes) {
1319 	case 2:
1320 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1321 		break;
1322 	case 4:
1323 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1324 		break;
1325 	case 8:
1326 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1327 		break;
1328 	}
1329 done:
1330 	return rc;
1331 }
1332 
1333 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1334 {
1335 	long sv = 0, mask;
1336 
1337 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1338 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1339 
1340 		if (ctxt->src.bytes == 2)
1341 			sv = (s16)ctxt->src.val & (s16)mask;
1342 		else if (ctxt->src.bytes == 4)
1343 			sv = (s32)ctxt->src.val & (s32)mask;
1344 		else
1345 			sv = (s64)ctxt->src.val & (s64)mask;
1346 
1347 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1348 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1349 	}
1350 
1351 	/* only subword offset */
1352 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1353 }
1354 
1355 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1356 			 unsigned long addr, void *dest, unsigned size)
1357 {
1358 	int rc;
1359 	struct read_cache *mc = &ctxt->mem_read;
1360 
1361 	if (mc->pos < mc->end)
1362 		goto read_cached;
1363 
1364 	WARN_ON((mc->end + size) >= sizeof(mc->data));
1365 
1366 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1367 				      &ctxt->exception);
1368 	if (rc != X86EMUL_CONTINUE)
1369 		return rc;
1370 
1371 	mc->end += size;
1372 
1373 read_cached:
1374 	memcpy(dest, mc->data + mc->pos, size);
1375 	mc->pos += size;
1376 	return X86EMUL_CONTINUE;
1377 }
1378 
1379 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1380 			  struct segmented_address addr,
1381 			  void *data,
1382 			  unsigned size)
1383 {
1384 	int rc;
1385 	ulong linear;
1386 
1387 	rc = linearize(ctxt, addr, size, false, &linear);
1388 	if (rc != X86EMUL_CONTINUE)
1389 		return rc;
1390 	return read_emulated(ctxt, linear, data, size);
1391 }
1392 
1393 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1394 			   struct segmented_address addr,
1395 			   const void *data,
1396 			   unsigned size)
1397 {
1398 	int rc;
1399 	ulong linear;
1400 
1401 	rc = linearize(ctxt, addr, size, true, &linear);
1402 	if (rc != X86EMUL_CONTINUE)
1403 		return rc;
1404 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1405 					 &ctxt->exception);
1406 }
1407 
1408 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1409 			     struct segmented_address addr,
1410 			     const void *orig_data, const void *data,
1411 			     unsigned size)
1412 {
1413 	int rc;
1414 	ulong linear;
1415 
1416 	rc = linearize(ctxt, addr, size, true, &linear);
1417 	if (rc != X86EMUL_CONTINUE)
1418 		return rc;
1419 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1420 					   size, &ctxt->exception);
1421 }
1422 
1423 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1424 			   unsigned int size, unsigned short port,
1425 			   void *dest)
1426 {
1427 	struct read_cache *rc = &ctxt->io_read;
1428 
1429 	if (rc->pos == rc->end) { /* refill pio read ahead */
1430 		unsigned int in_page, n;
1431 		unsigned int count = ctxt->rep_prefix ?
1432 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1433 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1434 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1435 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1436 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1437 		if (n == 0)
1438 			n = 1;
1439 		rc->pos = rc->end = 0;
1440 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1441 			return 0;
1442 		rc->end = n * size;
1443 	}
1444 
1445 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1446 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1447 		ctxt->dst.data = rc->data + rc->pos;
1448 		ctxt->dst.type = OP_MEM_STR;
1449 		ctxt->dst.count = (rc->end - rc->pos) / size;
1450 		rc->pos = rc->end;
1451 	} else {
1452 		memcpy(dest, rc->data + rc->pos, size);
1453 		rc->pos += size;
1454 	}
1455 	return 1;
1456 }
1457 
1458 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1459 				     u16 index, struct desc_struct *desc)
1460 {
1461 	struct desc_ptr dt;
1462 	ulong addr;
1463 
1464 	ctxt->ops->get_idt(ctxt, &dt);
1465 
1466 	if (dt.size < index * 8 + 7)
1467 		return emulate_gp(ctxt, index << 3 | 0x2);
1468 
1469 	addr = dt.address + index * 8;
1470 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1471 }
1472 
1473 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1474 				     u16 selector, struct desc_ptr *dt)
1475 {
1476 	const struct x86_emulate_ops *ops = ctxt->ops;
1477 	u32 base3 = 0;
1478 
1479 	if (selector & 1 << 2) {
1480 		struct desc_struct desc;
1481 		u16 sel;
1482 
1483 		memset(dt, 0, sizeof(*dt));
1484 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1485 				      VCPU_SREG_LDTR))
1486 			return;
1487 
1488 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1489 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1490 	} else
1491 		ops->get_gdt(ctxt, dt);
1492 }
1493 
1494 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1495 			      u16 selector, ulong *desc_addr_p)
1496 {
1497 	struct desc_ptr dt;
1498 	u16 index = selector >> 3;
1499 	ulong addr;
1500 
1501 	get_descriptor_table_ptr(ctxt, selector, &dt);
1502 
1503 	if (dt.size < index * 8 + 7)
1504 		return emulate_gp(ctxt, selector & 0xfffc);
1505 
1506 	addr = dt.address + index * 8;
1507 
1508 #ifdef CONFIG_X86_64
1509 	if (addr >> 32 != 0) {
1510 		u64 efer = 0;
1511 
1512 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1513 		if (!(efer & EFER_LMA))
1514 			addr &= (u32)-1;
1515 	}
1516 #endif
1517 
1518 	*desc_addr_p = addr;
1519 	return X86EMUL_CONTINUE;
1520 }
1521 
1522 /* allowed just for 8 bytes segments */
1523 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1524 				   u16 selector, struct desc_struct *desc,
1525 				   ulong *desc_addr_p)
1526 {
1527 	int rc;
1528 
1529 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1530 	if (rc != X86EMUL_CONTINUE)
1531 		return rc;
1532 
1533 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1534 }
1535 
1536 /* allowed just for 8 bytes segments */
1537 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1538 				    u16 selector, struct desc_struct *desc)
1539 {
1540 	int rc;
1541 	ulong addr;
1542 
1543 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1544 	if (rc != X86EMUL_CONTINUE)
1545 		return rc;
1546 
1547 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1548 }
1549 
1550 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1551 				     u16 selector, int seg, u8 cpl,
1552 				     enum x86_transfer_type transfer,
1553 				     struct desc_struct *desc)
1554 {
1555 	struct desc_struct seg_desc, old_desc;
1556 	u8 dpl, rpl;
1557 	unsigned err_vec = GP_VECTOR;
1558 	u32 err_code = 0;
1559 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1560 	ulong desc_addr;
1561 	int ret;
1562 	u16 dummy;
1563 	u32 base3 = 0;
1564 
1565 	memset(&seg_desc, 0, sizeof(seg_desc));
1566 
1567 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1568 		/* set real mode segment descriptor (keep limit etc. for
1569 		 * unreal mode) */
1570 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1571 		set_desc_base(&seg_desc, selector << 4);
1572 		goto load;
1573 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1574 		/* VM86 needs a clean new segment descriptor */
1575 		set_desc_base(&seg_desc, selector << 4);
1576 		set_desc_limit(&seg_desc, 0xffff);
1577 		seg_desc.type = 3;
1578 		seg_desc.p = 1;
1579 		seg_desc.s = 1;
1580 		seg_desc.dpl = 3;
1581 		goto load;
1582 	}
1583 
1584 	rpl = selector & 3;
1585 
1586 	/* TR should be in GDT only */
1587 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1588 		goto exception;
1589 
1590 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1591 	if (null_selector) {
1592 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1593 			goto exception;
1594 
1595 		if (seg == VCPU_SREG_SS) {
1596 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1597 				goto exception;
1598 
1599 			/*
1600 			 * ctxt->ops->set_segment expects the CPL to be in
1601 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1602 			 */
1603 			seg_desc.type = 3;
1604 			seg_desc.p = 1;
1605 			seg_desc.s = 1;
1606 			seg_desc.dpl = cpl;
1607 			seg_desc.d = 1;
1608 			seg_desc.g = 1;
1609 		}
1610 
1611 		/* Skip all following checks */
1612 		goto load;
1613 	}
1614 
1615 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1616 	if (ret != X86EMUL_CONTINUE)
1617 		return ret;
1618 
1619 	err_code = selector & 0xfffc;
1620 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1621 							   GP_VECTOR;
1622 
1623 	/* can't load system descriptor into segment selector */
1624 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1625 		if (transfer == X86_TRANSFER_CALL_JMP)
1626 			return X86EMUL_UNHANDLEABLE;
1627 		goto exception;
1628 	}
1629 
1630 	dpl = seg_desc.dpl;
1631 
1632 	switch (seg) {
1633 	case VCPU_SREG_SS:
1634 		/*
1635 		 * segment is not a writable data segment or segment
1636 		 * selector's RPL != CPL or segment selector's RPL != CPL
1637 		 */
1638 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1639 			goto exception;
1640 		break;
1641 	case VCPU_SREG_CS:
1642 		if (!(seg_desc.type & 8))
1643 			goto exception;
1644 
1645 		if (transfer == X86_TRANSFER_RET) {
1646 			/* RET can never return to an inner privilege level. */
1647 			if (rpl < cpl)
1648 				goto exception;
1649 			/* Outer-privilege level return is not implemented */
1650 			if (rpl > cpl)
1651 				return X86EMUL_UNHANDLEABLE;
1652 		}
1653 		if (transfer == X86_TRANSFER_RET || transfer == X86_TRANSFER_TASK_SWITCH) {
1654 			if (seg_desc.type & 4) {
1655 				/* conforming */
1656 				if (dpl > rpl)
1657 					goto exception;
1658 			} else {
1659 				/* nonconforming */
1660 				if (dpl != rpl)
1661 					goto exception;
1662 			}
1663 		} else { /* X86_TRANSFER_CALL_JMP */
1664 			if (seg_desc.type & 4) {
1665 				/* conforming */
1666 				if (dpl > cpl)
1667 					goto exception;
1668 			} else {
1669 				/* nonconforming */
1670 				if (rpl > cpl || dpl != cpl)
1671 					goto exception;
1672 			}
1673 		}
1674 		/* in long-mode d/b must be clear if l is set */
1675 		if (seg_desc.d && seg_desc.l) {
1676 			u64 efer = 0;
1677 
1678 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1679 			if (efer & EFER_LMA)
1680 				goto exception;
1681 		}
1682 
1683 		/* CS(RPL) <- CPL */
1684 		selector = (selector & 0xfffc) | cpl;
1685 		break;
1686 	case VCPU_SREG_TR:
1687 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1688 			goto exception;
1689 		if (!seg_desc.p) {
1690 			err_vec = NP_VECTOR;
1691 			goto exception;
1692 		}
1693 		old_desc = seg_desc;
1694 		seg_desc.type |= 2; /* busy */
1695 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1696 						  sizeof(seg_desc), &ctxt->exception);
1697 		if (ret != X86EMUL_CONTINUE)
1698 			return ret;
1699 		break;
1700 	case VCPU_SREG_LDTR:
1701 		if (seg_desc.s || seg_desc.type != 2)
1702 			goto exception;
1703 		break;
1704 	default: /*  DS, ES, FS, or GS */
1705 		/*
1706 		 * segment is not a data or readable code segment or
1707 		 * ((segment is a data or nonconforming code segment)
1708 		 * and (both RPL and CPL > DPL))
1709 		 */
1710 		if ((seg_desc.type & 0xa) == 0x8 ||
1711 		    (((seg_desc.type & 0xc) != 0xc) &&
1712 		     (rpl > dpl && cpl > dpl)))
1713 			goto exception;
1714 		break;
1715 	}
1716 
1717 	if (!seg_desc.p) {
1718 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1719 		goto exception;
1720 	}
1721 
1722 	if (seg_desc.s) {
1723 		/* mark segment as accessed */
1724 		if (!(seg_desc.type & 1)) {
1725 			seg_desc.type |= 1;
1726 			ret = write_segment_descriptor(ctxt, selector,
1727 						       &seg_desc);
1728 			if (ret != X86EMUL_CONTINUE)
1729 				return ret;
1730 		}
1731 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1732 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1733 		if (ret != X86EMUL_CONTINUE)
1734 			return ret;
1735 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1736 				((u64)base3 << 32), ctxt))
1737 			return emulate_gp(ctxt, 0);
1738 	}
1739 load:
1740 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1741 	if (desc)
1742 		*desc = seg_desc;
1743 	return X86EMUL_CONTINUE;
1744 exception:
1745 	return emulate_exception(ctxt, err_vec, err_code, true);
1746 }
1747 
1748 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1749 				   u16 selector, int seg)
1750 {
1751 	u8 cpl = ctxt->ops->cpl(ctxt);
1752 
1753 	/*
1754 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1755 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1756 	 * but it's wrong).
1757 	 *
1758 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1759 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1760 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1761 	 * and only forbid it here.
1762 	 */
1763 	if (seg == VCPU_SREG_SS && selector == 3 &&
1764 	    ctxt->mode == X86EMUL_MODE_PROT64)
1765 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1766 
1767 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1768 					 X86_TRANSFER_NONE, NULL);
1769 }
1770 
1771 static void write_register_operand(struct operand *op)
1772 {
1773 	return assign_register(op->addr.reg, op->val, op->bytes);
1774 }
1775 
1776 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1777 {
1778 	switch (op->type) {
1779 	case OP_REG:
1780 		write_register_operand(op);
1781 		break;
1782 	case OP_MEM:
1783 		if (ctxt->lock_prefix)
1784 			return segmented_cmpxchg(ctxt,
1785 						 op->addr.mem,
1786 						 &op->orig_val,
1787 						 &op->val,
1788 						 op->bytes);
1789 		else
1790 			return segmented_write(ctxt,
1791 					       op->addr.mem,
1792 					       &op->val,
1793 					       op->bytes);
1794 		break;
1795 	case OP_MEM_STR:
1796 		return segmented_write(ctxt,
1797 				       op->addr.mem,
1798 				       op->data,
1799 				       op->bytes * op->count);
1800 		break;
1801 	case OP_XMM:
1802 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1803 		break;
1804 	case OP_MM:
1805 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1806 		break;
1807 	case OP_NONE:
1808 		/* no writeback */
1809 		break;
1810 	default:
1811 		break;
1812 	}
1813 	return X86EMUL_CONTINUE;
1814 }
1815 
1816 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1817 {
1818 	struct segmented_address addr;
1819 
1820 	rsp_increment(ctxt, -bytes);
1821 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1822 	addr.seg = VCPU_SREG_SS;
1823 
1824 	return segmented_write(ctxt, addr, data, bytes);
1825 }
1826 
1827 static int em_push(struct x86_emulate_ctxt *ctxt)
1828 {
1829 	/* Disable writeback. */
1830 	ctxt->dst.type = OP_NONE;
1831 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1832 }
1833 
1834 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1835 		       void *dest, int len)
1836 {
1837 	int rc;
1838 	struct segmented_address addr;
1839 
1840 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1841 	addr.seg = VCPU_SREG_SS;
1842 	rc = segmented_read(ctxt, addr, dest, len);
1843 	if (rc != X86EMUL_CONTINUE)
1844 		return rc;
1845 
1846 	rsp_increment(ctxt, len);
1847 	return rc;
1848 }
1849 
1850 static int em_pop(struct x86_emulate_ctxt *ctxt)
1851 {
1852 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1853 }
1854 
1855 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1856 			void *dest, int len)
1857 {
1858 	int rc;
1859 	unsigned long val, change_mask;
1860 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1861 	int cpl = ctxt->ops->cpl(ctxt);
1862 
1863 	rc = emulate_pop(ctxt, &val, len);
1864 	if (rc != X86EMUL_CONTINUE)
1865 		return rc;
1866 
1867 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1868 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1869 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1870 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1871 
1872 	switch(ctxt->mode) {
1873 	case X86EMUL_MODE_PROT64:
1874 	case X86EMUL_MODE_PROT32:
1875 	case X86EMUL_MODE_PROT16:
1876 		if (cpl == 0)
1877 			change_mask |= X86_EFLAGS_IOPL;
1878 		if (cpl <= iopl)
1879 			change_mask |= X86_EFLAGS_IF;
1880 		break;
1881 	case X86EMUL_MODE_VM86:
1882 		if (iopl < 3)
1883 			return emulate_gp(ctxt, 0);
1884 		change_mask |= X86_EFLAGS_IF;
1885 		break;
1886 	default: /* real mode */
1887 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1888 		break;
1889 	}
1890 
1891 	*(unsigned long *)dest =
1892 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1893 
1894 	return rc;
1895 }
1896 
1897 static int em_popf(struct x86_emulate_ctxt *ctxt)
1898 {
1899 	ctxt->dst.type = OP_REG;
1900 	ctxt->dst.addr.reg = &ctxt->eflags;
1901 	ctxt->dst.bytes = ctxt->op_bytes;
1902 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1903 }
1904 
1905 static int em_enter(struct x86_emulate_ctxt *ctxt)
1906 {
1907 	int rc;
1908 	unsigned frame_size = ctxt->src.val;
1909 	unsigned nesting_level = ctxt->src2.val & 31;
1910 	ulong rbp;
1911 
1912 	if (nesting_level)
1913 		return X86EMUL_UNHANDLEABLE;
1914 
1915 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1916 	rc = push(ctxt, &rbp, stack_size(ctxt));
1917 	if (rc != X86EMUL_CONTINUE)
1918 		return rc;
1919 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1920 		      stack_mask(ctxt));
1921 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1922 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1923 		      stack_mask(ctxt));
1924 	return X86EMUL_CONTINUE;
1925 }
1926 
1927 static int em_leave(struct x86_emulate_ctxt *ctxt)
1928 {
1929 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1930 		      stack_mask(ctxt));
1931 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1932 }
1933 
1934 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1935 {
1936 	int seg = ctxt->src2.val;
1937 
1938 	ctxt->src.val = get_segment_selector(ctxt, seg);
1939 	if (ctxt->op_bytes == 4) {
1940 		rsp_increment(ctxt, -2);
1941 		ctxt->op_bytes = 2;
1942 	}
1943 
1944 	return em_push(ctxt);
1945 }
1946 
1947 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1948 {
1949 	int seg = ctxt->src2.val;
1950 	unsigned long selector;
1951 	int rc;
1952 
1953 	rc = emulate_pop(ctxt, &selector, 2);
1954 	if (rc != X86EMUL_CONTINUE)
1955 		return rc;
1956 
1957 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1958 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1959 	if (ctxt->op_bytes > 2)
1960 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1961 
1962 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1963 	return rc;
1964 }
1965 
1966 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1967 {
1968 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1969 	int rc = X86EMUL_CONTINUE;
1970 	int reg = VCPU_REGS_RAX;
1971 
1972 	while (reg <= VCPU_REGS_RDI) {
1973 		(reg == VCPU_REGS_RSP) ?
1974 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1975 
1976 		rc = em_push(ctxt);
1977 		if (rc != X86EMUL_CONTINUE)
1978 			return rc;
1979 
1980 		++reg;
1981 	}
1982 
1983 	return rc;
1984 }
1985 
1986 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1987 {
1988 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1989 	return em_push(ctxt);
1990 }
1991 
1992 static int em_popa(struct x86_emulate_ctxt *ctxt)
1993 {
1994 	int rc = X86EMUL_CONTINUE;
1995 	int reg = VCPU_REGS_RDI;
1996 	u32 val;
1997 
1998 	while (reg >= VCPU_REGS_RAX) {
1999 		if (reg == VCPU_REGS_RSP) {
2000 			rsp_increment(ctxt, ctxt->op_bytes);
2001 			--reg;
2002 		}
2003 
2004 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2005 		if (rc != X86EMUL_CONTINUE)
2006 			break;
2007 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2008 		--reg;
2009 	}
2010 	return rc;
2011 }
2012 
2013 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2014 {
2015 	const struct x86_emulate_ops *ops = ctxt->ops;
2016 	int rc;
2017 	struct desc_ptr dt;
2018 	gva_t cs_addr;
2019 	gva_t eip_addr;
2020 	u16 cs, eip;
2021 
2022 	/* TODO: Add limit checks */
2023 	ctxt->src.val = ctxt->eflags;
2024 	rc = em_push(ctxt);
2025 	if (rc != X86EMUL_CONTINUE)
2026 		return rc;
2027 
2028 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2029 
2030 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2031 	rc = em_push(ctxt);
2032 	if (rc != X86EMUL_CONTINUE)
2033 		return rc;
2034 
2035 	ctxt->src.val = ctxt->_eip;
2036 	rc = em_push(ctxt);
2037 	if (rc != X86EMUL_CONTINUE)
2038 		return rc;
2039 
2040 	ops->get_idt(ctxt, &dt);
2041 
2042 	eip_addr = dt.address + (irq << 2);
2043 	cs_addr = dt.address + (irq << 2) + 2;
2044 
2045 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2046 	if (rc != X86EMUL_CONTINUE)
2047 		return rc;
2048 
2049 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2050 	if (rc != X86EMUL_CONTINUE)
2051 		return rc;
2052 
2053 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2054 	if (rc != X86EMUL_CONTINUE)
2055 		return rc;
2056 
2057 	ctxt->_eip = eip;
2058 
2059 	return rc;
2060 }
2061 
2062 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2063 {
2064 	int rc;
2065 
2066 	invalidate_registers(ctxt);
2067 	rc = __emulate_int_real(ctxt, irq);
2068 	if (rc == X86EMUL_CONTINUE)
2069 		writeback_registers(ctxt);
2070 	return rc;
2071 }
2072 
2073 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2074 {
2075 	switch(ctxt->mode) {
2076 	case X86EMUL_MODE_REAL:
2077 		return __emulate_int_real(ctxt, irq);
2078 	case X86EMUL_MODE_VM86:
2079 	case X86EMUL_MODE_PROT16:
2080 	case X86EMUL_MODE_PROT32:
2081 	case X86EMUL_MODE_PROT64:
2082 	default:
2083 		/* Protected mode interrupts unimplemented yet */
2084 		return X86EMUL_UNHANDLEABLE;
2085 	}
2086 }
2087 
2088 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2089 {
2090 	int rc = X86EMUL_CONTINUE;
2091 	unsigned long temp_eip = 0;
2092 	unsigned long temp_eflags = 0;
2093 	unsigned long cs = 0;
2094 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2095 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2096 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2097 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2098 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2099 			     X86_EFLAGS_FIXED;
2100 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2101 				  X86_EFLAGS_VIP;
2102 
2103 	/* TODO: Add stack limit check */
2104 
2105 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2106 
2107 	if (rc != X86EMUL_CONTINUE)
2108 		return rc;
2109 
2110 	if (temp_eip & ~0xffff)
2111 		return emulate_gp(ctxt, 0);
2112 
2113 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2114 
2115 	if (rc != X86EMUL_CONTINUE)
2116 		return rc;
2117 
2118 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2119 
2120 	if (rc != X86EMUL_CONTINUE)
2121 		return rc;
2122 
2123 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2124 
2125 	if (rc != X86EMUL_CONTINUE)
2126 		return rc;
2127 
2128 	ctxt->_eip = temp_eip;
2129 
2130 	if (ctxt->op_bytes == 4)
2131 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2132 	else if (ctxt->op_bytes == 2) {
2133 		ctxt->eflags &= ~0xffff;
2134 		ctxt->eflags |= temp_eflags;
2135 	}
2136 
2137 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2138 	ctxt->eflags |= X86_EFLAGS_FIXED;
2139 	ctxt->ops->set_nmi_mask(ctxt, false);
2140 
2141 	return rc;
2142 }
2143 
2144 static int em_iret(struct x86_emulate_ctxt *ctxt)
2145 {
2146 	switch(ctxt->mode) {
2147 	case X86EMUL_MODE_REAL:
2148 		return emulate_iret_real(ctxt);
2149 	case X86EMUL_MODE_VM86:
2150 	case X86EMUL_MODE_PROT16:
2151 	case X86EMUL_MODE_PROT32:
2152 	case X86EMUL_MODE_PROT64:
2153 	default:
2154 		/* iret from protected mode unimplemented yet */
2155 		return X86EMUL_UNHANDLEABLE;
2156 	}
2157 }
2158 
2159 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2160 {
2161 	int rc;
2162 	unsigned short sel;
2163 	struct desc_struct new_desc;
2164 	u8 cpl = ctxt->ops->cpl(ctxt);
2165 
2166 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2167 
2168 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2169 				       X86_TRANSFER_CALL_JMP,
2170 				       &new_desc);
2171 	if (rc != X86EMUL_CONTINUE)
2172 		return rc;
2173 
2174 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2175 	/* Error handling is not implemented. */
2176 	if (rc != X86EMUL_CONTINUE)
2177 		return X86EMUL_UNHANDLEABLE;
2178 
2179 	return rc;
2180 }
2181 
2182 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2183 {
2184 	return assign_eip_near(ctxt, ctxt->src.val);
2185 }
2186 
2187 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2188 {
2189 	int rc;
2190 	long int old_eip;
2191 
2192 	old_eip = ctxt->_eip;
2193 	rc = assign_eip_near(ctxt, ctxt->src.val);
2194 	if (rc != X86EMUL_CONTINUE)
2195 		return rc;
2196 	ctxt->src.val = old_eip;
2197 	rc = em_push(ctxt);
2198 	return rc;
2199 }
2200 
2201 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2202 {
2203 	u64 old = ctxt->dst.orig_val64;
2204 
2205 	if (ctxt->dst.bytes == 16)
2206 		return X86EMUL_UNHANDLEABLE;
2207 
2208 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2209 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2210 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2211 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2212 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2213 	} else {
2214 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2215 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2216 
2217 		ctxt->eflags |= X86_EFLAGS_ZF;
2218 	}
2219 	return X86EMUL_CONTINUE;
2220 }
2221 
2222 static int em_ret(struct x86_emulate_ctxt *ctxt)
2223 {
2224 	int rc;
2225 	unsigned long eip;
2226 
2227 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2228 	if (rc != X86EMUL_CONTINUE)
2229 		return rc;
2230 
2231 	return assign_eip_near(ctxt, eip);
2232 }
2233 
2234 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2235 {
2236 	int rc;
2237 	unsigned long eip, cs;
2238 	int cpl = ctxt->ops->cpl(ctxt);
2239 	struct desc_struct new_desc;
2240 
2241 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2242 	if (rc != X86EMUL_CONTINUE)
2243 		return rc;
2244 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2245 	if (rc != X86EMUL_CONTINUE)
2246 		return rc;
2247 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2248 				       X86_TRANSFER_RET,
2249 				       &new_desc);
2250 	if (rc != X86EMUL_CONTINUE)
2251 		return rc;
2252 	rc = assign_eip_far(ctxt, eip, &new_desc);
2253 	/* Error handling is not implemented. */
2254 	if (rc != X86EMUL_CONTINUE)
2255 		return X86EMUL_UNHANDLEABLE;
2256 
2257 	return rc;
2258 }
2259 
2260 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2261 {
2262         int rc;
2263 
2264         rc = em_ret_far(ctxt);
2265         if (rc != X86EMUL_CONTINUE)
2266                 return rc;
2267         rsp_increment(ctxt, ctxt->src.val);
2268         return X86EMUL_CONTINUE;
2269 }
2270 
2271 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2272 {
2273 	/* Save real source value, then compare EAX against destination. */
2274 	ctxt->dst.orig_val = ctxt->dst.val;
2275 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2276 	ctxt->src.orig_val = ctxt->src.val;
2277 	ctxt->src.val = ctxt->dst.orig_val;
2278 	fastop(ctxt, em_cmp);
2279 
2280 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2281 		/* Success: write back to memory; no update of EAX */
2282 		ctxt->src.type = OP_NONE;
2283 		ctxt->dst.val = ctxt->src.orig_val;
2284 	} else {
2285 		/* Failure: write the value we saw to EAX. */
2286 		ctxt->src.type = OP_REG;
2287 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2288 		ctxt->src.val = ctxt->dst.orig_val;
2289 		/* Create write-cycle to dest by writing the same value */
2290 		ctxt->dst.val = ctxt->dst.orig_val;
2291 	}
2292 	return X86EMUL_CONTINUE;
2293 }
2294 
2295 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2296 {
2297 	int seg = ctxt->src2.val;
2298 	unsigned short sel;
2299 	int rc;
2300 
2301 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2302 
2303 	rc = load_segment_descriptor(ctxt, sel, seg);
2304 	if (rc != X86EMUL_CONTINUE)
2305 		return rc;
2306 
2307 	ctxt->dst.val = ctxt->src.val;
2308 	return rc;
2309 }
2310 
2311 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2312 {
2313 #ifdef CONFIG_X86_64
2314 	return ctxt->ops->guest_has_long_mode(ctxt);
2315 #else
2316 	return false;
2317 #endif
2318 }
2319 
2320 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2321 {
2322 	desc->g    = (flags >> 23) & 1;
2323 	desc->d    = (flags >> 22) & 1;
2324 	desc->l    = (flags >> 21) & 1;
2325 	desc->avl  = (flags >> 20) & 1;
2326 	desc->p    = (flags >> 15) & 1;
2327 	desc->dpl  = (flags >> 13) & 3;
2328 	desc->s    = (flags >> 12) & 1;
2329 	desc->type = (flags >>  8) & 15;
2330 }
2331 
2332 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, const char *smstate,
2333 			   int n)
2334 {
2335 	struct desc_struct desc;
2336 	int offset;
2337 	u16 selector;
2338 
2339 	selector = GET_SMSTATE(u32, smstate, 0x7fa8 + n * 4);
2340 
2341 	if (n < 3)
2342 		offset = 0x7f84 + n * 12;
2343 	else
2344 		offset = 0x7f2c + (n - 3) * 12;
2345 
2346 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2347 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2348 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, offset));
2349 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2350 	return X86EMUL_CONTINUE;
2351 }
2352 
2353 #ifdef CONFIG_X86_64
2354 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, const char *smstate,
2355 			   int n)
2356 {
2357 	struct desc_struct desc;
2358 	int offset;
2359 	u16 selector;
2360 	u32 base3;
2361 
2362 	offset = 0x7e00 + n * 16;
2363 
2364 	selector =                GET_SMSTATE(u16, smstate, offset);
2365 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smstate, offset + 2) << 8);
2366 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2367 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2368 	base3 =                   GET_SMSTATE(u32, smstate, offset + 12);
2369 
2370 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2371 	return X86EMUL_CONTINUE;
2372 }
2373 #endif
2374 
2375 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2376 				    u64 cr0, u64 cr3, u64 cr4)
2377 {
2378 	int bad;
2379 	u64 pcid;
2380 
2381 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
2382 	pcid = 0;
2383 	if (cr4 & X86_CR4_PCIDE) {
2384 		pcid = cr3 & 0xfff;
2385 		cr3 &= ~0xfff;
2386 	}
2387 
2388 	bad = ctxt->ops->set_cr(ctxt, 3, cr3);
2389 	if (bad)
2390 		return X86EMUL_UNHANDLEABLE;
2391 
2392 	/*
2393 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2394 	 * Then enable protected mode.	However, PCID cannot be enabled
2395 	 * if EFER.LMA=0, so set it separately.
2396 	 */
2397 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2398 	if (bad)
2399 		return X86EMUL_UNHANDLEABLE;
2400 
2401 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2402 	if (bad)
2403 		return X86EMUL_UNHANDLEABLE;
2404 
2405 	if (cr4 & X86_CR4_PCIDE) {
2406 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2407 		if (bad)
2408 			return X86EMUL_UNHANDLEABLE;
2409 		if (pcid) {
2410 			bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid);
2411 			if (bad)
2412 				return X86EMUL_UNHANDLEABLE;
2413 		}
2414 
2415 	}
2416 
2417 	return X86EMUL_CONTINUE;
2418 }
2419 
2420 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
2421 			     const char *smstate)
2422 {
2423 	struct desc_struct desc;
2424 	struct desc_ptr dt;
2425 	u16 selector;
2426 	u32 val, cr0, cr3, cr4;
2427 	int i;
2428 
2429 	cr0 =                      GET_SMSTATE(u32, smstate, 0x7ffc);
2430 	cr3 =                      GET_SMSTATE(u32, smstate, 0x7ff8);
2431 	ctxt->eflags =             GET_SMSTATE(u32, smstate, 0x7ff4) | X86_EFLAGS_FIXED;
2432 	ctxt->_eip =               GET_SMSTATE(u32, smstate, 0x7ff0);
2433 
2434 	for (i = 0; i < 8; i++)
2435 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smstate, 0x7fd0 + i * 4);
2436 
2437 	val = GET_SMSTATE(u32, smstate, 0x7fcc);
2438 
2439 	if (ctxt->ops->set_dr(ctxt, 6, val))
2440 		return X86EMUL_UNHANDLEABLE;
2441 
2442 	val = GET_SMSTATE(u32, smstate, 0x7fc8);
2443 
2444 	if (ctxt->ops->set_dr(ctxt, 7, val))
2445 		return X86EMUL_UNHANDLEABLE;
2446 
2447 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc4);
2448 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f64));
2449 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f60));
2450 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f5c));
2451 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2452 
2453 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc0);
2454 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f80));
2455 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f7c));
2456 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f78));
2457 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2458 
2459 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f74);
2460 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f70);
2461 	ctxt->ops->set_gdt(ctxt, &dt);
2462 
2463 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f58);
2464 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f54);
2465 	ctxt->ops->set_idt(ctxt, &dt);
2466 
2467 	for (i = 0; i < 6; i++) {
2468 		int r = rsm_load_seg_32(ctxt, smstate, i);
2469 		if (r != X86EMUL_CONTINUE)
2470 			return r;
2471 	}
2472 
2473 	cr4 = GET_SMSTATE(u32, smstate, 0x7f14);
2474 
2475 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7ef8));
2476 
2477 	return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2478 }
2479 
2480 #ifdef CONFIG_X86_64
2481 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
2482 			     const char *smstate)
2483 {
2484 	struct desc_struct desc;
2485 	struct desc_ptr dt;
2486 	u64 val, cr0, cr3, cr4;
2487 	u32 base3;
2488 	u16 selector;
2489 	int i, r;
2490 
2491 	for (i = 0; i < 16; i++)
2492 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smstate, 0x7ff8 - i * 8);
2493 
2494 	ctxt->_eip   = GET_SMSTATE(u64, smstate, 0x7f78);
2495 	ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7f70) | X86_EFLAGS_FIXED;
2496 
2497 	val = GET_SMSTATE(u64, smstate, 0x7f68);
2498 
2499 	if (ctxt->ops->set_dr(ctxt, 6, val))
2500 		return X86EMUL_UNHANDLEABLE;
2501 
2502 	val = GET_SMSTATE(u64, smstate, 0x7f60);
2503 
2504 	if (ctxt->ops->set_dr(ctxt, 7, val))
2505 		return X86EMUL_UNHANDLEABLE;
2506 
2507 	cr0 =                       GET_SMSTATE(u64, smstate, 0x7f58);
2508 	cr3 =                       GET_SMSTATE(u64, smstate, 0x7f50);
2509 	cr4 =                       GET_SMSTATE(u64, smstate, 0x7f48);
2510 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7f00));
2511 	val =                       GET_SMSTATE(u64, smstate, 0x7ed0);
2512 
2513 	if (ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA))
2514 		return X86EMUL_UNHANDLEABLE;
2515 
2516 	selector =                  GET_SMSTATE(u32, smstate, 0x7e90);
2517 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e92) << 8);
2518 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e94));
2519 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e98));
2520 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e9c);
2521 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2522 
2523 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e84);
2524 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e88);
2525 	ctxt->ops->set_idt(ctxt, &dt);
2526 
2527 	selector =                  GET_SMSTATE(u32, smstate, 0x7e70);
2528 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e72) << 8);
2529 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e74));
2530 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e78));
2531 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e7c);
2532 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2533 
2534 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e64);
2535 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e68);
2536 	ctxt->ops->set_gdt(ctxt, &dt);
2537 
2538 	r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2539 	if (r != X86EMUL_CONTINUE)
2540 		return r;
2541 
2542 	for (i = 0; i < 6; i++) {
2543 		r = rsm_load_seg_64(ctxt, smstate, i);
2544 		if (r != X86EMUL_CONTINUE)
2545 			return r;
2546 	}
2547 
2548 	return X86EMUL_CONTINUE;
2549 }
2550 #endif
2551 
2552 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2553 {
2554 	unsigned long cr0, cr4, efer;
2555 	char buf[512];
2556 	u64 smbase;
2557 	int ret;
2558 
2559 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2560 		return emulate_ud(ctxt);
2561 
2562 	smbase = ctxt->ops->get_smbase(ctxt);
2563 
2564 	ret = ctxt->ops->read_phys(ctxt, smbase + 0xfe00, buf, sizeof(buf));
2565 	if (ret != X86EMUL_CONTINUE)
2566 		return X86EMUL_UNHANDLEABLE;
2567 
2568 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2569 		ctxt->ops->set_nmi_mask(ctxt, false);
2570 
2571 	ctxt->ops->exiting_smm(ctxt);
2572 
2573 	/*
2574 	 * Get back to real mode, to prepare a safe state in which to load
2575 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2576 	 * supports long mode.
2577 	 */
2578 	if (emulator_has_longmode(ctxt)) {
2579 		struct desc_struct cs_desc;
2580 
2581 		/* Zero CR4.PCIDE before CR0.PG.  */
2582 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2583 		if (cr4 & X86_CR4_PCIDE)
2584 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2585 
2586 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2587 		memset(&cs_desc, 0, sizeof(cs_desc));
2588 		cs_desc.type = 0xb;
2589 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2590 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2591 	}
2592 
2593 	/* For the 64-bit case, this will clear EFER.LMA.  */
2594 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2595 	if (cr0 & X86_CR0_PE)
2596 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2597 
2598 	if (emulator_has_longmode(ctxt)) {
2599 		/* Clear CR4.PAE before clearing EFER.LME. */
2600 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2601 		if (cr4 & X86_CR4_PAE)
2602 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2603 
2604 		/* And finally go back to 32-bit mode.  */
2605 		efer = 0;
2606 		ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2607 	}
2608 
2609 	/*
2610 	 * Give leave_smm() a chance to make ISA-specific changes to the vCPU
2611 	 * state (e.g. enter guest mode) before loading state from the SMM
2612 	 * state-save area.
2613 	 */
2614 	if (ctxt->ops->leave_smm(ctxt, buf))
2615 		goto emulate_shutdown;
2616 
2617 #ifdef CONFIG_X86_64
2618 	if (emulator_has_longmode(ctxt))
2619 		ret = rsm_load_state_64(ctxt, buf);
2620 	else
2621 #endif
2622 		ret = rsm_load_state_32(ctxt, buf);
2623 
2624 	if (ret != X86EMUL_CONTINUE)
2625 		goto emulate_shutdown;
2626 
2627 	/*
2628 	 * Note, the ctxt->ops callbacks are responsible for handling side
2629 	 * effects when writing MSRs and CRs, e.g. MMU context resets, CPUID
2630 	 * runtime updates, etc...  If that changes, e.g. this flow is moved
2631 	 * out of the emulator to make it look more like enter_smm(), then
2632 	 * those side effects need to be explicitly handled for both success
2633 	 * and shutdown.
2634 	 */
2635 	return X86EMUL_CONTINUE;
2636 
2637 emulate_shutdown:
2638 	ctxt->ops->triple_fault(ctxt);
2639 	return X86EMUL_CONTINUE;
2640 }
2641 
2642 static void
2643 setup_syscalls_segments(struct desc_struct *cs, struct desc_struct *ss)
2644 {
2645 	cs->l = 0;		/* will be adjusted later */
2646 	set_desc_base(cs, 0);	/* flat segment */
2647 	cs->g = 1;		/* 4kb granularity */
2648 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2649 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2650 	cs->s = 1;
2651 	cs->dpl = 0;		/* will be adjusted later */
2652 	cs->p = 1;
2653 	cs->d = 1;
2654 	cs->avl = 0;
2655 
2656 	set_desc_base(ss, 0);	/* flat segment */
2657 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2658 	ss->g = 1;		/* 4kb granularity */
2659 	ss->s = 1;
2660 	ss->type = 0x03;	/* Read/Write, Accessed */
2661 	ss->d = 1;		/* 32bit stack segment */
2662 	ss->dpl = 0;
2663 	ss->p = 1;
2664 	ss->l = 0;
2665 	ss->avl = 0;
2666 }
2667 
2668 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2669 {
2670 	u32 eax, ebx, ecx, edx;
2671 
2672 	eax = ecx = 0;
2673 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2674 	return is_guest_vendor_intel(ebx, ecx, edx);
2675 }
2676 
2677 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2678 {
2679 	const struct x86_emulate_ops *ops = ctxt->ops;
2680 	u32 eax, ebx, ecx, edx;
2681 
2682 	/*
2683 	 * syscall should always be enabled in longmode - so only become
2684 	 * vendor specific (cpuid) if other modes are active...
2685 	 */
2686 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2687 		return true;
2688 
2689 	eax = 0x00000000;
2690 	ecx = 0x00000000;
2691 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2692 	/*
2693 	 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a
2694 	 * 64bit guest with a 32bit compat-app running will #UD !! While this
2695 	 * behaviour can be fixed (by emulating) into AMD response - CPUs of
2696 	 * AMD can't behave like Intel.
2697 	 */
2698 	if (is_guest_vendor_intel(ebx, ecx, edx))
2699 		return false;
2700 
2701 	if (is_guest_vendor_amd(ebx, ecx, edx) ||
2702 	    is_guest_vendor_hygon(ebx, ecx, edx))
2703 		return true;
2704 
2705 	/*
2706 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2707 	 * stricter rules...
2708 	 */
2709 	return false;
2710 }
2711 
2712 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2713 {
2714 	const struct x86_emulate_ops *ops = ctxt->ops;
2715 	struct desc_struct cs, ss;
2716 	u64 msr_data;
2717 	u16 cs_sel, ss_sel;
2718 	u64 efer = 0;
2719 
2720 	/* syscall is not available in real mode */
2721 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2722 	    ctxt->mode == X86EMUL_MODE_VM86)
2723 		return emulate_ud(ctxt);
2724 
2725 	if (!(em_syscall_is_enabled(ctxt)))
2726 		return emulate_ud(ctxt);
2727 
2728 	ops->get_msr(ctxt, MSR_EFER, &efer);
2729 	if (!(efer & EFER_SCE))
2730 		return emulate_ud(ctxt);
2731 
2732 	setup_syscalls_segments(&cs, &ss);
2733 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2734 	msr_data >>= 32;
2735 	cs_sel = (u16)(msr_data & 0xfffc);
2736 	ss_sel = (u16)(msr_data + 8);
2737 
2738 	if (efer & EFER_LMA) {
2739 		cs.d = 0;
2740 		cs.l = 1;
2741 	}
2742 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2743 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2744 
2745 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2746 	if (efer & EFER_LMA) {
2747 #ifdef CONFIG_X86_64
2748 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2749 
2750 		ops->get_msr(ctxt,
2751 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2752 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2753 		ctxt->_eip = msr_data;
2754 
2755 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2756 		ctxt->eflags &= ~msr_data;
2757 		ctxt->eflags |= X86_EFLAGS_FIXED;
2758 #endif
2759 	} else {
2760 		/* legacy mode */
2761 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2762 		ctxt->_eip = (u32)msr_data;
2763 
2764 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2765 	}
2766 
2767 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2768 	return X86EMUL_CONTINUE;
2769 }
2770 
2771 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2772 {
2773 	const struct x86_emulate_ops *ops = ctxt->ops;
2774 	struct desc_struct cs, ss;
2775 	u64 msr_data;
2776 	u16 cs_sel, ss_sel;
2777 	u64 efer = 0;
2778 
2779 	ops->get_msr(ctxt, MSR_EFER, &efer);
2780 	/* inject #GP if in real mode */
2781 	if (ctxt->mode == X86EMUL_MODE_REAL)
2782 		return emulate_gp(ctxt, 0);
2783 
2784 	/*
2785 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2786 	 * mode).
2787 	 */
2788 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2789 	    && !vendor_intel(ctxt))
2790 		return emulate_ud(ctxt);
2791 
2792 	/* sysenter/sysexit have not been tested in 64bit mode. */
2793 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2794 		return X86EMUL_UNHANDLEABLE;
2795 
2796 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2797 	if ((msr_data & 0xfffc) == 0x0)
2798 		return emulate_gp(ctxt, 0);
2799 
2800 	setup_syscalls_segments(&cs, &ss);
2801 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2802 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2803 	ss_sel = cs_sel + 8;
2804 	if (efer & EFER_LMA) {
2805 		cs.d = 0;
2806 		cs.l = 1;
2807 	}
2808 
2809 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2810 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2811 
2812 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2813 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2814 
2815 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2816 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2817 							      (u32)msr_data;
2818 	if (efer & EFER_LMA)
2819 		ctxt->mode = X86EMUL_MODE_PROT64;
2820 
2821 	return X86EMUL_CONTINUE;
2822 }
2823 
2824 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2825 {
2826 	const struct x86_emulate_ops *ops = ctxt->ops;
2827 	struct desc_struct cs, ss;
2828 	u64 msr_data, rcx, rdx;
2829 	int usermode;
2830 	u16 cs_sel = 0, ss_sel = 0;
2831 
2832 	/* inject #GP if in real mode or Virtual 8086 mode */
2833 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2834 	    ctxt->mode == X86EMUL_MODE_VM86)
2835 		return emulate_gp(ctxt, 0);
2836 
2837 	setup_syscalls_segments(&cs, &ss);
2838 
2839 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2840 		usermode = X86EMUL_MODE_PROT64;
2841 	else
2842 		usermode = X86EMUL_MODE_PROT32;
2843 
2844 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2845 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2846 
2847 	cs.dpl = 3;
2848 	ss.dpl = 3;
2849 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2850 	switch (usermode) {
2851 	case X86EMUL_MODE_PROT32:
2852 		cs_sel = (u16)(msr_data + 16);
2853 		if ((msr_data & 0xfffc) == 0x0)
2854 			return emulate_gp(ctxt, 0);
2855 		ss_sel = (u16)(msr_data + 24);
2856 		rcx = (u32)rcx;
2857 		rdx = (u32)rdx;
2858 		break;
2859 	case X86EMUL_MODE_PROT64:
2860 		cs_sel = (u16)(msr_data + 32);
2861 		if (msr_data == 0x0)
2862 			return emulate_gp(ctxt, 0);
2863 		ss_sel = cs_sel + 8;
2864 		cs.d = 0;
2865 		cs.l = 1;
2866 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2867 		    emul_is_noncanonical_address(rdx, ctxt))
2868 			return emulate_gp(ctxt, 0);
2869 		break;
2870 	}
2871 	cs_sel |= SEGMENT_RPL_MASK;
2872 	ss_sel |= SEGMENT_RPL_MASK;
2873 
2874 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2875 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2876 
2877 	ctxt->_eip = rdx;
2878 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2879 
2880 	return X86EMUL_CONTINUE;
2881 }
2882 
2883 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2884 {
2885 	int iopl;
2886 	if (ctxt->mode == X86EMUL_MODE_REAL)
2887 		return false;
2888 	if (ctxt->mode == X86EMUL_MODE_VM86)
2889 		return true;
2890 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2891 	return ctxt->ops->cpl(ctxt) > iopl;
2892 }
2893 
2894 #define VMWARE_PORT_VMPORT	(0x5658)
2895 #define VMWARE_PORT_VMRPC	(0x5659)
2896 
2897 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2898 					    u16 port, u16 len)
2899 {
2900 	const struct x86_emulate_ops *ops = ctxt->ops;
2901 	struct desc_struct tr_seg;
2902 	u32 base3;
2903 	int r;
2904 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2905 	unsigned mask = (1 << len) - 1;
2906 	unsigned long base;
2907 
2908 	/*
2909 	 * VMware allows access to these ports even if denied
2910 	 * by TSS I/O permission bitmap. Mimic behavior.
2911 	 */
2912 	if (enable_vmware_backdoor &&
2913 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2914 		return true;
2915 
2916 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2917 	if (!tr_seg.p)
2918 		return false;
2919 	if (desc_limit_scaled(&tr_seg) < 103)
2920 		return false;
2921 	base = get_desc_base(&tr_seg);
2922 #ifdef CONFIG_X86_64
2923 	base |= ((u64)base3) << 32;
2924 #endif
2925 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2926 	if (r != X86EMUL_CONTINUE)
2927 		return false;
2928 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2929 		return false;
2930 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2931 	if (r != X86EMUL_CONTINUE)
2932 		return false;
2933 	if ((perm >> bit_idx) & mask)
2934 		return false;
2935 	return true;
2936 }
2937 
2938 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2939 				 u16 port, u16 len)
2940 {
2941 	if (ctxt->perm_ok)
2942 		return true;
2943 
2944 	if (emulator_bad_iopl(ctxt))
2945 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2946 			return false;
2947 
2948 	ctxt->perm_ok = true;
2949 
2950 	return true;
2951 }
2952 
2953 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2954 {
2955 	/*
2956 	 * Intel CPUs mask the counter and pointers in quite strange
2957 	 * manner when ECX is zero due to REP-string optimizations.
2958 	 */
2959 #ifdef CONFIG_X86_64
2960 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2961 		return;
2962 
2963 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2964 
2965 	switch (ctxt->b) {
2966 	case 0xa4:	/* movsb */
2967 	case 0xa5:	/* movsd/w */
2968 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2969 		fallthrough;
2970 	case 0xaa:	/* stosb */
2971 	case 0xab:	/* stosd/w */
2972 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2973 	}
2974 #endif
2975 }
2976 
2977 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2978 				struct tss_segment_16 *tss)
2979 {
2980 	tss->ip = ctxt->_eip;
2981 	tss->flag = ctxt->eflags;
2982 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2983 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2984 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2985 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2986 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2987 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2988 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2989 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2990 
2991 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2992 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2993 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2994 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2995 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2996 }
2997 
2998 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2999 				 struct tss_segment_16 *tss)
3000 {
3001 	int ret;
3002 	u8 cpl;
3003 
3004 	ctxt->_eip = tss->ip;
3005 	ctxt->eflags = tss->flag | 2;
3006 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
3007 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
3008 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
3009 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
3010 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
3011 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
3012 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
3013 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
3014 
3015 	/*
3016 	 * SDM says that segment selectors are loaded before segment
3017 	 * descriptors
3018 	 */
3019 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
3020 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3021 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3022 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3023 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3024 
3025 	cpl = tss->cs & 3;
3026 
3027 	/*
3028 	 * Now load segment descriptors. If fault happens at this stage
3029 	 * it is handled in a context of new task
3030 	 */
3031 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
3032 					X86_TRANSFER_TASK_SWITCH, NULL);
3033 	if (ret != X86EMUL_CONTINUE)
3034 		return ret;
3035 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3036 					X86_TRANSFER_TASK_SWITCH, NULL);
3037 	if (ret != X86EMUL_CONTINUE)
3038 		return ret;
3039 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3040 					X86_TRANSFER_TASK_SWITCH, NULL);
3041 	if (ret != X86EMUL_CONTINUE)
3042 		return ret;
3043 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3044 					X86_TRANSFER_TASK_SWITCH, NULL);
3045 	if (ret != X86EMUL_CONTINUE)
3046 		return ret;
3047 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3048 					X86_TRANSFER_TASK_SWITCH, NULL);
3049 	if (ret != X86EMUL_CONTINUE)
3050 		return ret;
3051 
3052 	return X86EMUL_CONTINUE;
3053 }
3054 
3055 static int task_switch_16(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
3056 			  ulong old_tss_base, struct desc_struct *new_desc)
3057 {
3058 	struct tss_segment_16 tss_seg;
3059 	int ret;
3060 	u32 new_tss_base = get_desc_base(new_desc);
3061 
3062 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3063 	if (ret != X86EMUL_CONTINUE)
3064 		return ret;
3065 
3066 	save_state_to_tss16(ctxt, &tss_seg);
3067 
3068 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3069 	if (ret != X86EMUL_CONTINUE)
3070 		return ret;
3071 
3072 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3073 	if (ret != X86EMUL_CONTINUE)
3074 		return ret;
3075 
3076 	if (old_tss_sel != 0xffff) {
3077 		tss_seg.prev_task_link = old_tss_sel;
3078 
3079 		ret = linear_write_system(ctxt, new_tss_base,
3080 					  &tss_seg.prev_task_link,
3081 					  sizeof(tss_seg.prev_task_link));
3082 		if (ret != X86EMUL_CONTINUE)
3083 			return ret;
3084 	}
3085 
3086 	return load_state_from_tss16(ctxt, &tss_seg);
3087 }
3088 
3089 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3090 				struct tss_segment_32 *tss)
3091 {
3092 	/* CR3 and ldt selector are not saved intentionally */
3093 	tss->eip = ctxt->_eip;
3094 	tss->eflags = ctxt->eflags;
3095 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3096 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3097 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3098 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3099 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3100 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3101 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3102 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3103 
3104 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3105 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3106 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3107 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3108 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3109 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3110 }
3111 
3112 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3113 				 struct tss_segment_32 *tss)
3114 {
3115 	int ret;
3116 	u8 cpl;
3117 
3118 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3119 		return emulate_gp(ctxt, 0);
3120 	ctxt->_eip = tss->eip;
3121 	ctxt->eflags = tss->eflags | 2;
3122 
3123 	/* General purpose registers */
3124 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3125 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3126 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3127 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3128 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3129 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3130 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3131 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3132 
3133 	/*
3134 	 * SDM says that segment selectors are loaded before segment
3135 	 * descriptors.  This is important because CPL checks will
3136 	 * use CS.RPL.
3137 	 */
3138 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3139 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3140 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3141 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3142 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3143 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3144 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3145 
3146 	/*
3147 	 * If we're switching between Protected Mode and VM86, we need to make
3148 	 * sure to update the mode before loading the segment descriptors so
3149 	 * that the selectors are interpreted correctly.
3150 	 */
3151 	if (ctxt->eflags & X86_EFLAGS_VM) {
3152 		ctxt->mode = X86EMUL_MODE_VM86;
3153 		cpl = 3;
3154 	} else {
3155 		ctxt->mode = X86EMUL_MODE_PROT32;
3156 		cpl = tss->cs & 3;
3157 	}
3158 
3159 	/*
3160 	 * Now load segment descriptors. If fault happens at this stage
3161 	 * it is handled in a context of new task
3162 	 */
3163 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3164 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3165 	if (ret != X86EMUL_CONTINUE)
3166 		return ret;
3167 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3168 					X86_TRANSFER_TASK_SWITCH, NULL);
3169 	if (ret != X86EMUL_CONTINUE)
3170 		return ret;
3171 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3172 					X86_TRANSFER_TASK_SWITCH, NULL);
3173 	if (ret != X86EMUL_CONTINUE)
3174 		return ret;
3175 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3176 					X86_TRANSFER_TASK_SWITCH, NULL);
3177 	if (ret != X86EMUL_CONTINUE)
3178 		return ret;
3179 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3180 					X86_TRANSFER_TASK_SWITCH, NULL);
3181 	if (ret != X86EMUL_CONTINUE)
3182 		return ret;
3183 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3184 					X86_TRANSFER_TASK_SWITCH, NULL);
3185 	if (ret != X86EMUL_CONTINUE)
3186 		return ret;
3187 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3188 					X86_TRANSFER_TASK_SWITCH, NULL);
3189 
3190 	return ret;
3191 }
3192 
3193 static int task_switch_32(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
3194 			  ulong old_tss_base, struct desc_struct *new_desc)
3195 {
3196 	struct tss_segment_32 tss_seg;
3197 	int ret;
3198 	u32 new_tss_base = get_desc_base(new_desc);
3199 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3200 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3201 
3202 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3203 	if (ret != X86EMUL_CONTINUE)
3204 		return ret;
3205 
3206 	save_state_to_tss32(ctxt, &tss_seg);
3207 
3208 	/* Only GP registers and segment selectors are saved */
3209 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3210 				  ldt_sel_offset - eip_offset);
3211 	if (ret != X86EMUL_CONTINUE)
3212 		return ret;
3213 
3214 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3215 	if (ret != X86EMUL_CONTINUE)
3216 		return ret;
3217 
3218 	if (old_tss_sel != 0xffff) {
3219 		tss_seg.prev_task_link = old_tss_sel;
3220 
3221 		ret = linear_write_system(ctxt, new_tss_base,
3222 					  &tss_seg.prev_task_link,
3223 					  sizeof(tss_seg.prev_task_link));
3224 		if (ret != X86EMUL_CONTINUE)
3225 			return ret;
3226 	}
3227 
3228 	return load_state_from_tss32(ctxt, &tss_seg);
3229 }
3230 
3231 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3232 				   u16 tss_selector, int idt_index, int reason,
3233 				   bool has_error_code, u32 error_code)
3234 {
3235 	const struct x86_emulate_ops *ops = ctxt->ops;
3236 	struct desc_struct curr_tss_desc, next_tss_desc;
3237 	int ret;
3238 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3239 	ulong old_tss_base =
3240 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3241 	u32 desc_limit;
3242 	ulong desc_addr, dr7;
3243 
3244 	/* FIXME: old_tss_base == ~0 ? */
3245 
3246 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3247 	if (ret != X86EMUL_CONTINUE)
3248 		return ret;
3249 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3250 	if (ret != X86EMUL_CONTINUE)
3251 		return ret;
3252 
3253 	/* FIXME: check that next_tss_desc is tss */
3254 
3255 	/*
3256 	 * Check privileges. The three cases are task switch caused by...
3257 	 *
3258 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3259 	 * 2. Exception/IRQ/iret: No check is performed
3260 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3261 	 *    hardware checks it before exiting.
3262 	 */
3263 	if (reason == TASK_SWITCH_GATE) {
3264 		if (idt_index != -1) {
3265 			/* Software interrupts */
3266 			struct desc_struct task_gate_desc;
3267 			int dpl;
3268 
3269 			ret = read_interrupt_descriptor(ctxt, idt_index,
3270 							&task_gate_desc);
3271 			if (ret != X86EMUL_CONTINUE)
3272 				return ret;
3273 
3274 			dpl = task_gate_desc.dpl;
3275 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3276 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3277 		}
3278 	}
3279 
3280 	desc_limit = desc_limit_scaled(&next_tss_desc);
3281 	if (!next_tss_desc.p ||
3282 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3283 	     desc_limit < 0x2b)) {
3284 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3285 	}
3286 
3287 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3288 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3289 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3290 	}
3291 
3292 	if (reason == TASK_SWITCH_IRET)
3293 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3294 
3295 	/* set back link to prev task only if NT bit is set in eflags
3296 	   note that old_tss_sel is not used after this point */
3297 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3298 		old_tss_sel = 0xffff;
3299 
3300 	if (next_tss_desc.type & 8)
3301 		ret = task_switch_32(ctxt, old_tss_sel, old_tss_base, &next_tss_desc);
3302 	else
3303 		ret = task_switch_16(ctxt, old_tss_sel,
3304 				     old_tss_base, &next_tss_desc);
3305 	if (ret != X86EMUL_CONTINUE)
3306 		return ret;
3307 
3308 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3309 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3310 
3311 	if (reason != TASK_SWITCH_IRET) {
3312 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3313 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3314 	}
3315 
3316 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3317 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3318 
3319 	if (has_error_code) {
3320 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3321 		ctxt->lock_prefix = 0;
3322 		ctxt->src.val = (unsigned long) error_code;
3323 		ret = em_push(ctxt);
3324 	}
3325 
3326 	ops->get_dr(ctxt, 7, &dr7);
3327 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3328 
3329 	return ret;
3330 }
3331 
3332 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3333 			 u16 tss_selector, int idt_index, int reason,
3334 			 bool has_error_code, u32 error_code)
3335 {
3336 	int rc;
3337 
3338 	invalidate_registers(ctxt);
3339 	ctxt->_eip = ctxt->eip;
3340 	ctxt->dst.type = OP_NONE;
3341 
3342 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3343 				     has_error_code, error_code);
3344 
3345 	if (rc == X86EMUL_CONTINUE) {
3346 		ctxt->eip = ctxt->_eip;
3347 		writeback_registers(ctxt);
3348 	}
3349 
3350 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3351 }
3352 
3353 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3354 		struct operand *op)
3355 {
3356 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3357 
3358 	register_address_increment(ctxt, reg, df * op->bytes);
3359 	op->addr.mem.ea = register_address(ctxt, reg);
3360 }
3361 
3362 static int em_das(struct x86_emulate_ctxt *ctxt)
3363 {
3364 	u8 al, old_al;
3365 	bool af, cf, old_cf;
3366 
3367 	cf = ctxt->eflags & X86_EFLAGS_CF;
3368 	al = ctxt->dst.val;
3369 
3370 	old_al = al;
3371 	old_cf = cf;
3372 	cf = false;
3373 	af = ctxt->eflags & X86_EFLAGS_AF;
3374 	if ((al & 0x0f) > 9 || af) {
3375 		al -= 6;
3376 		cf = old_cf | (al >= 250);
3377 		af = true;
3378 	} else {
3379 		af = false;
3380 	}
3381 	if (old_al > 0x99 || old_cf) {
3382 		al -= 0x60;
3383 		cf = true;
3384 	}
3385 
3386 	ctxt->dst.val = al;
3387 	/* Set PF, ZF, SF */
3388 	ctxt->src.type = OP_IMM;
3389 	ctxt->src.val = 0;
3390 	ctxt->src.bytes = 1;
3391 	fastop(ctxt, em_or);
3392 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3393 	if (cf)
3394 		ctxt->eflags |= X86_EFLAGS_CF;
3395 	if (af)
3396 		ctxt->eflags |= X86_EFLAGS_AF;
3397 	return X86EMUL_CONTINUE;
3398 }
3399 
3400 static int em_aam(struct x86_emulate_ctxt *ctxt)
3401 {
3402 	u8 al, ah;
3403 
3404 	if (ctxt->src.val == 0)
3405 		return emulate_de(ctxt);
3406 
3407 	al = ctxt->dst.val & 0xff;
3408 	ah = al / ctxt->src.val;
3409 	al %= ctxt->src.val;
3410 
3411 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3412 
3413 	/* Set PF, ZF, SF */
3414 	ctxt->src.type = OP_IMM;
3415 	ctxt->src.val = 0;
3416 	ctxt->src.bytes = 1;
3417 	fastop(ctxt, em_or);
3418 
3419 	return X86EMUL_CONTINUE;
3420 }
3421 
3422 static int em_aad(struct x86_emulate_ctxt *ctxt)
3423 {
3424 	u8 al = ctxt->dst.val & 0xff;
3425 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3426 
3427 	al = (al + (ah * ctxt->src.val)) & 0xff;
3428 
3429 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3430 
3431 	/* Set PF, ZF, SF */
3432 	ctxt->src.type = OP_IMM;
3433 	ctxt->src.val = 0;
3434 	ctxt->src.bytes = 1;
3435 	fastop(ctxt, em_or);
3436 
3437 	return X86EMUL_CONTINUE;
3438 }
3439 
3440 static int em_call(struct x86_emulate_ctxt *ctxt)
3441 {
3442 	int rc;
3443 	long rel = ctxt->src.val;
3444 
3445 	ctxt->src.val = (unsigned long)ctxt->_eip;
3446 	rc = jmp_rel(ctxt, rel);
3447 	if (rc != X86EMUL_CONTINUE)
3448 		return rc;
3449 	return em_push(ctxt);
3450 }
3451 
3452 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3453 {
3454 	u16 sel, old_cs;
3455 	ulong old_eip;
3456 	int rc;
3457 	struct desc_struct old_desc, new_desc;
3458 	const struct x86_emulate_ops *ops = ctxt->ops;
3459 	int cpl = ctxt->ops->cpl(ctxt);
3460 	enum x86emul_mode prev_mode = ctxt->mode;
3461 
3462 	old_eip = ctxt->_eip;
3463 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3464 
3465 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3466 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3467 				       X86_TRANSFER_CALL_JMP, &new_desc);
3468 	if (rc != X86EMUL_CONTINUE)
3469 		return rc;
3470 
3471 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3472 	if (rc != X86EMUL_CONTINUE)
3473 		goto fail;
3474 
3475 	ctxt->src.val = old_cs;
3476 	rc = em_push(ctxt);
3477 	if (rc != X86EMUL_CONTINUE)
3478 		goto fail;
3479 
3480 	ctxt->src.val = old_eip;
3481 	rc = em_push(ctxt);
3482 	/* If we failed, we tainted the memory, but the very least we should
3483 	   restore cs */
3484 	if (rc != X86EMUL_CONTINUE) {
3485 		pr_warn_once("faulting far call emulation tainted memory\n");
3486 		goto fail;
3487 	}
3488 	return rc;
3489 fail:
3490 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3491 	ctxt->mode = prev_mode;
3492 	return rc;
3493 
3494 }
3495 
3496 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3497 {
3498 	int rc;
3499 	unsigned long eip;
3500 
3501 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3502 	if (rc != X86EMUL_CONTINUE)
3503 		return rc;
3504 	rc = assign_eip_near(ctxt, eip);
3505 	if (rc != X86EMUL_CONTINUE)
3506 		return rc;
3507 	rsp_increment(ctxt, ctxt->src.val);
3508 	return X86EMUL_CONTINUE;
3509 }
3510 
3511 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3512 {
3513 	/* Write back the register source. */
3514 	ctxt->src.val = ctxt->dst.val;
3515 	write_register_operand(&ctxt->src);
3516 
3517 	/* Write back the memory destination with implicit LOCK prefix. */
3518 	ctxt->dst.val = ctxt->src.orig_val;
3519 	ctxt->lock_prefix = 1;
3520 	return X86EMUL_CONTINUE;
3521 }
3522 
3523 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3524 {
3525 	ctxt->dst.val = ctxt->src2.val;
3526 	return fastop(ctxt, em_imul);
3527 }
3528 
3529 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3530 {
3531 	ctxt->dst.type = OP_REG;
3532 	ctxt->dst.bytes = ctxt->src.bytes;
3533 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3534 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3535 
3536 	return X86EMUL_CONTINUE;
3537 }
3538 
3539 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3540 {
3541 	u64 tsc_aux = 0;
3542 
3543 	if (!ctxt->ops->guest_has_rdpid(ctxt))
3544 		return emulate_ud(ctxt);
3545 
3546 	ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux);
3547 	ctxt->dst.val = tsc_aux;
3548 	return X86EMUL_CONTINUE;
3549 }
3550 
3551 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3552 {
3553 	u64 tsc = 0;
3554 
3555 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3556 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3557 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3558 	return X86EMUL_CONTINUE;
3559 }
3560 
3561 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3562 {
3563 	u64 pmc;
3564 
3565 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3566 		return emulate_gp(ctxt, 0);
3567 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3568 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3569 	return X86EMUL_CONTINUE;
3570 }
3571 
3572 static int em_mov(struct x86_emulate_ctxt *ctxt)
3573 {
3574 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3575 	return X86EMUL_CONTINUE;
3576 }
3577 
3578 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3579 {
3580 	u16 tmp;
3581 
3582 	if (!ctxt->ops->guest_has_movbe(ctxt))
3583 		return emulate_ud(ctxt);
3584 
3585 	switch (ctxt->op_bytes) {
3586 	case 2:
3587 		/*
3588 		 * From MOVBE definition: "...When the operand size is 16 bits,
3589 		 * the upper word of the destination register remains unchanged
3590 		 * ..."
3591 		 *
3592 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3593 		 * rules so we have to do the operation almost per hand.
3594 		 */
3595 		tmp = (u16)ctxt->src.val;
3596 		ctxt->dst.val &= ~0xffffUL;
3597 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3598 		break;
3599 	case 4:
3600 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3601 		break;
3602 	case 8:
3603 		ctxt->dst.val = swab64(ctxt->src.val);
3604 		break;
3605 	default:
3606 		BUG();
3607 	}
3608 	return X86EMUL_CONTINUE;
3609 }
3610 
3611 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3612 {
3613 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3614 		return emulate_gp(ctxt, 0);
3615 
3616 	/* Disable writeback. */
3617 	ctxt->dst.type = OP_NONE;
3618 	return X86EMUL_CONTINUE;
3619 }
3620 
3621 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3622 {
3623 	unsigned long val;
3624 
3625 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3626 		val = ctxt->src.val & ~0ULL;
3627 	else
3628 		val = ctxt->src.val & ~0U;
3629 
3630 	/* #UD condition is already handled. */
3631 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3632 		return emulate_gp(ctxt, 0);
3633 
3634 	/* Disable writeback. */
3635 	ctxt->dst.type = OP_NONE;
3636 	return X86EMUL_CONTINUE;
3637 }
3638 
3639 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3640 {
3641 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3642 	u64 msr_data;
3643 	int r;
3644 
3645 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3646 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3647 	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);
3648 
3649 	if (r == X86EMUL_IO_NEEDED)
3650 		return r;
3651 
3652 	if (r > 0)
3653 		return emulate_gp(ctxt, 0);
3654 
3655 	return r < 0 ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
3656 }
3657 
3658 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3659 {
3660 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3661 	u64 msr_data;
3662 	int r;
3663 
3664 	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);
3665 
3666 	if (r == X86EMUL_IO_NEEDED)
3667 		return r;
3668 
3669 	if (r)
3670 		return emulate_gp(ctxt, 0);
3671 
3672 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3673 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3674 	return X86EMUL_CONTINUE;
3675 }
3676 
3677 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3678 {
3679 	if (segment > VCPU_SREG_GS &&
3680 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3681 	    ctxt->ops->cpl(ctxt) > 0)
3682 		return emulate_gp(ctxt, 0);
3683 
3684 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3685 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3686 		ctxt->dst.bytes = 2;
3687 	return X86EMUL_CONTINUE;
3688 }
3689 
3690 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3691 {
3692 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3693 		return emulate_ud(ctxt);
3694 
3695 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3696 }
3697 
3698 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3699 {
3700 	u16 sel = ctxt->src.val;
3701 
3702 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3703 		return emulate_ud(ctxt);
3704 
3705 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3706 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3707 
3708 	/* Disable writeback. */
3709 	ctxt->dst.type = OP_NONE;
3710 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3711 }
3712 
3713 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3714 {
3715 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3716 }
3717 
3718 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3719 {
3720 	u16 sel = ctxt->src.val;
3721 
3722 	/* Disable writeback. */
3723 	ctxt->dst.type = OP_NONE;
3724 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3725 }
3726 
3727 static int em_str(struct x86_emulate_ctxt *ctxt)
3728 {
3729 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3730 }
3731 
3732 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3733 {
3734 	u16 sel = ctxt->src.val;
3735 
3736 	/* Disable writeback. */
3737 	ctxt->dst.type = OP_NONE;
3738 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3739 }
3740 
3741 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3742 {
3743 	int rc;
3744 	ulong linear;
3745 
3746 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3747 	if (rc == X86EMUL_CONTINUE)
3748 		ctxt->ops->invlpg(ctxt, linear);
3749 	/* Disable writeback. */
3750 	ctxt->dst.type = OP_NONE;
3751 	return X86EMUL_CONTINUE;
3752 }
3753 
3754 static int em_clts(struct x86_emulate_ctxt *ctxt)
3755 {
3756 	ulong cr0;
3757 
3758 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3759 	cr0 &= ~X86_CR0_TS;
3760 	ctxt->ops->set_cr(ctxt, 0, cr0);
3761 	return X86EMUL_CONTINUE;
3762 }
3763 
3764 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3765 {
3766 	int rc = ctxt->ops->fix_hypercall(ctxt);
3767 
3768 	if (rc != X86EMUL_CONTINUE)
3769 		return rc;
3770 
3771 	/* Let the processor re-execute the fixed hypercall */
3772 	ctxt->_eip = ctxt->eip;
3773 	/* Disable writeback. */
3774 	ctxt->dst.type = OP_NONE;
3775 	return X86EMUL_CONTINUE;
3776 }
3777 
3778 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3779 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3780 					      struct desc_ptr *ptr))
3781 {
3782 	struct desc_ptr desc_ptr;
3783 
3784 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3785 	    ctxt->ops->cpl(ctxt) > 0)
3786 		return emulate_gp(ctxt, 0);
3787 
3788 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3789 		ctxt->op_bytes = 8;
3790 	get(ctxt, &desc_ptr);
3791 	if (ctxt->op_bytes == 2) {
3792 		ctxt->op_bytes = 4;
3793 		desc_ptr.address &= 0x00ffffff;
3794 	}
3795 	/* Disable writeback. */
3796 	ctxt->dst.type = OP_NONE;
3797 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3798 				   &desc_ptr, 2 + ctxt->op_bytes);
3799 }
3800 
3801 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3802 {
3803 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3804 }
3805 
3806 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3807 {
3808 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3809 }
3810 
3811 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3812 {
3813 	struct desc_ptr desc_ptr;
3814 	int rc;
3815 
3816 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3817 		ctxt->op_bytes = 8;
3818 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3819 			     &desc_ptr.size, &desc_ptr.address,
3820 			     ctxt->op_bytes);
3821 	if (rc != X86EMUL_CONTINUE)
3822 		return rc;
3823 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3824 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3825 		return emulate_gp(ctxt, 0);
3826 	if (lgdt)
3827 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3828 	else
3829 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3830 	/* Disable writeback. */
3831 	ctxt->dst.type = OP_NONE;
3832 	return X86EMUL_CONTINUE;
3833 }
3834 
3835 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3836 {
3837 	return em_lgdt_lidt(ctxt, true);
3838 }
3839 
3840 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3841 {
3842 	return em_lgdt_lidt(ctxt, false);
3843 }
3844 
3845 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3846 {
3847 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3848 	    ctxt->ops->cpl(ctxt) > 0)
3849 		return emulate_gp(ctxt, 0);
3850 
3851 	if (ctxt->dst.type == OP_MEM)
3852 		ctxt->dst.bytes = 2;
3853 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3854 	return X86EMUL_CONTINUE;
3855 }
3856 
3857 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3858 {
3859 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3860 			  | (ctxt->src.val & 0x0f));
3861 	ctxt->dst.type = OP_NONE;
3862 	return X86EMUL_CONTINUE;
3863 }
3864 
3865 static int em_loop(struct x86_emulate_ctxt *ctxt)
3866 {
3867 	int rc = X86EMUL_CONTINUE;
3868 
3869 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3870 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3871 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3872 		rc = jmp_rel(ctxt, ctxt->src.val);
3873 
3874 	return rc;
3875 }
3876 
3877 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3878 {
3879 	int rc = X86EMUL_CONTINUE;
3880 
3881 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3882 		rc = jmp_rel(ctxt, ctxt->src.val);
3883 
3884 	return rc;
3885 }
3886 
3887 static int em_in(struct x86_emulate_ctxt *ctxt)
3888 {
3889 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3890 			     &ctxt->dst.val))
3891 		return X86EMUL_IO_NEEDED;
3892 
3893 	return X86EMUL_CONTINUE;
3894 }
3895 
3896 static int em_out(struct x86_emulate_ctxt *ctxt)
3897 {
3898 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3899 				    &ctxt->src.val, 1);
3900 	/* Disable writeback. */
3901 	ctxt->dst.type = OP_NONE;
3902 	return X86EMUL_CONTINUE;
3903 }
3904 
3905 static int em_cli(struct x86_emulate_ctxt *ctxt)
3906 {
3907 	if (emulator_bad_iopl(ctxt))
3908 		return emulate_gp(ctxt, 0);
3909 
3910 	ctxt->eflags &= ~X86_EFLAGS_IF;
3911 	return X86EMUL_CONTINUE;
3912 }
3913 
3914 static int em_sti(struct x86_emulate_ctxt *ctxt)
3915 {
3916 	if (emulator_bad_iopl(ctxt))
3917 		return emulate_gp(ctxt, 0);
3918 
3919 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3920 	ctxt->eflags |= X86_EFLAGS_IF;
3921 	return X86EMUL_CONTINUE;
3922 }
3923 
3924 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3925 {
3926 	u32 eax, ebx, ecx, edx;
3927 	u64 msr = 0;
3928 
3929 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3930 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3931 	    ctxt->ops->cpl(ctxt)) {
3932 		return emulate_gp(ctxt, 0);
3933 	}
3934 
3935 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3936 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3937 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3938 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3939 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3940 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3941 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3942 	return X86EMUL_CONTINUE;
3943 }
3944 
3945 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3946 {
3947 	u32 flags;
3948 
3949 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3950 		X86_EFLAGS_SF;
3951 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3952 
3953 	ctxt->eflags &= ~0xffUL;
3954 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3955 	return X86EMUL_CONTINUE;
3956 }
3957 
3958 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3959 {
3960 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3961 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3962 	return X86EMUL_CONTINUE;
3963 }
3964 
3965 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3966 {
3967 	switch (ctxt->op_bytes) {
3968 #ifdef CONFIG_X86_64
3969 	case 8:
3970 		asm("bswap %0" : "+r"(ctxt->dst.val));
3971 		break;
3972 #endif
3973 	default:
3974 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3975 		break;
3976 	}
3977 	return X86EMUL_CONTINUE;
3978 }
3979 
3980 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3981 {
3982 	/* emulating clflush regardless of cpuid */
3983 	return X86EMUL_CONTINUE;
3984 }
3985 
3986 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
3987 {
3988 	/* emulating clflushopt regardless of cpuid */
3989 	return X86EMUL_CONTINUE;
3990 }
3991 
3992 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3993 {
3994 	ctxt->dst.val = (s32) ctxt->src.val;
3995 	return X86EMUL_CONTINUE;
3996 }
3997 
3998 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3999 {
4000 	if (!ctxt->ops->guest_has_fxsr(ctxt))
4001 		return emulate_ud(ctxt);
4002 
4003 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
4004 		return emulate_nm(ctxt);
4005 
4006 	/*
4007 	 * Don't emulate a case that should never be hit, instead of working
4008 	 * around a lack of fxsave64/fxrstor64 on old compilers.
4009 	 */
4010 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
4011 		return X86EMUL_UNHANDLEABLE;
4012 
4013 	return X86EMUL_CONTINUE;
4014 }
4015 
4016 /*
4017  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
4018  * and restore MXCSR.
4019  */
4020 static size_t __fxstate_size(int nregs)
4021 {
4022 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
4023 }
4024 
4025 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
4026 {
4027 	bool cr4_osfxsr;
4028 	if (ctxt->mode == X86EMUL_MODE_PROT64)
4029 		return __fxstate_size(16);
4030 
4031 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
4032 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
4033 }
4034 
4035 /*
4036  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
4037  *  1) 16 bit mode
4038  *  2) 32 bit mode
4039  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
4040  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
4041  *       save and restore
4042  *  3) 64-bit mode with REX.W prefix
4043  *     - like (2), but XMM 8-15 are being saved and restored
4044  *  4) 64-bit mode without REX.W prefix
4045  *     - like (3), but FIP and FDP are 64 bit
4046  *
4047  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
4048  * desired result.  (4) is not emulated.
4049  *
4050  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
4051  * and FPU DS) should match.
4052  */
4053 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
4054 {
4055 	struct fxregs_state fx_state;
4056 	int rc;
4057 
4058 	rc = check_fxsr(ctxt);
4059 	if (rc != X86EMUL_CONTINUE)
4060 		return rc;
4061 
4062 	kvm_fpu_get();
4063 
4064 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4065 
4066 	kvm_fpu_put();
4067 
4068 	if (rc != X86EMUL_CONTINUE)
4069 		return rc;
4070 
4071 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4072 		                   fxstate_size(ctxt));
4073 }
4074 
4075 /*
4076  * FXRSTOR might restore XMM registers not provided by the guest. Fill
4077  * in the host registers (via FXSAVE) instead, so they won't be modified.
4078  * (preemption has to stay disabled until FXRSTOR).
4079  *
4080  * Use noinline to keep the stack for other functions called by callers small.
4081  */
4082 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
4083 				 const size_t used_size)
4084 {
4085 	struct fxregs_state fx_tmp;
4086 	int rc;
4087 
4088 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
4089 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
4090 	       __fxstate_size(16) - used_size);
4091 
4092 	return rc;
4093 }
4094 
4095 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4096 {
4097 	struct fxregs_state fx_state;
4098 	int rc;
4099 	size_t size;
4100 
4101 	rc = check_fxsr(ctxt);
4102 	if (rc != X86EMUL_CONTINUE)
4103 		return rc;
4104 
4105 	size = fxstate_size(ctxt);
4106 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4107 	if (rc != X86EMUL_CONTINUE)
4108 		return rc;
4109 
4110 	kvm_fpu_get();
4111 
4112 	if (size < __fxstate_size(16)) {
4113 		rc = fxregs_fixup(&fx_state, size);
4114 		if (rc != X86EMUL_CONTINUE)
4115 			goto out;
4116 	}
4117 
4118 	if (fx_state.mxcsr >> 16) {
4119 		rc = emulate_gp(ctxt, 0);
4120 		goto out;
4121 	}
4122 
4123 	if (rc == X86EMUL_CONTINUE)
4124 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4125 
4126 out:
4127 	kvm_fpu_put();
4128 
4129 	return rc;
4130 }
4131 
4132 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
4133 {
4134 	u32 eax, ecx, edx;
4135 
4136 	eax = reg_read(ctxt, VCPU_REGS_RAX);
4137 	edx = reg_read(ctxt, VCPU_REGS_RDX);
4138 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
4139 
4140 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
4141 		return emulate_gp(ctxt, 0);
4142 
4143 	return X86EMUL_CONTINUE;
4144 }
4145 
4146 static bool valid_cr(int nr)
4147 {
4148 	switch (nr) {
4149 	case 0:
4150 	case 2 ... 4:
4151 	case 8:
4152 		return true;
4153 	default:
4154 		return false;
4155 	}
4156 }
4157 
4158 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
4159 {
4160 	if (!valid_cr(ctxt->modrm_reg))
4161 		return emulate_ud(ctxt);
4162 
4163 	return X86EMUL_CONTINUE;
4164 }
4165 
4166 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4167 {
4168 	unsigned long dr7;
4169 
4170 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4171 
4172 	/* Check if DR7.Global_Enable is set */
4173 	return dr7 & (1 << 13);
4174 }
4175 
4176 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4177 {
4178 	int dr = ctxt->modrm_reg;
4179 	u64 cr4;
4180 
4181 	if (dr > 7)
4182 		return emulate_ud(ctxt);
4183 
4184 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4185 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4186 		return emulate_ud(ctxt);
4187 
4188 	if (check_dr7_gd(ctxt)) {
4189 		ulong dr6;
4190 
4191 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4192 		dr6 &= ~DR_TRAP_BITS;
4193 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
4194 		ctxt->ops->set_dr(ctxt, 6, dr6);
4195 		return emulate_db(ctxt);
4196 	}
4197 
4198 	return X86EMUL_CONTINUE;
4199 }
4200 
4201 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4202 {
4203 	u64 new_val = ctxt->src.val64;
4204 	int dr = ctxt->modrm_reg;
4205 
4206 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4207 		return emulate_gp(ctxt, 0);
4208 
4209 	return check_dr_read(ctxt);
4210 }
4211 
4212 static int check_svme(struct x86_emulate_ctxt *ctxt)
4213 {
4214 	u64 efer = 0;
4215 
4216 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4217 
4218 	if (!(efer & EFER_SVME))
4219 		return emulate_ud(ctxt);
4220 
4221 	return X86EMUL_CONTINUE;
4222 }
4223 
4224 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4225 {
4226 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4227 
4228 	/* Valid physical address? */
4229 	if (rax & 0xffff000000000000ULL)
4230 		return emulate_gp(ctxt, 0);
4231 
4232 	return check_svme(ctxt);
4233 }
4234 
4235 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4236 {
4237 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4238 
4239 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4240 		return emulate_gp(ctxt, 0);
4241 
4242 	return X86EMUL_CONTINUE;
4243 }
4244 
4245 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4246 {
4247 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4248 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4249 
4250 	/*
4251 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
4252 	 * in Ring3 when CR4.PCE=0.
4253 	 */
4254 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
4255 		return X86EMUL_CONTINUE;
4256 
4257 	/*
4258 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
4259 	 * check however is unnecessary because CPL is always 0 outside
4260 	 * protected mode.
4261 	 */
4262 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4263 	    ctxt->ops->check_pmc(ctxt, rcx))
4264 		return emulate_gp(ctxt, 0);
4265 
4266 	return X86EMUL_CONTINUE;
4267 }
4268 
4269 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4270 {
4271 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4272 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4273 		return emulate_gp(ctxt, 0);
4274 
4275 	return X86EMUL_CONTINUE;
4276 }
4277 
4278 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4279 {
4280 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4281 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4282 		return emulate_gp(ctxt, 0);
4283 
4284 	return X86EMUL_CONTINUE;
4285 }
4286 
4287 #define D(_y) { .flags = (_y) }
4288 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4289 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4290 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4291 #define N    D(NotImpl)
4292 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4293 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4294 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4295 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4296 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4297 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4298 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4299 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4300 #define II(_f, _e, _i) \
4301 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4302 #define IIP(_f, _e, _i, _p) \
4303 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4304 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4305 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4306 
4307 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4308 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4309 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4310 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4311 #define I2bvIP(_f, _e, _i, _p) \
4312 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4313 
4314 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4315 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4316 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4317 
4318 static const struct opcode group7_rm0[] = {
4319 	N,
4320 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4321 	N, N, N, N, N, N,
4322 };
4323 
4324 static const struct opcode group7_rm1[] = {
4325 	DI(SrcNone | Priv, monitor),
4326 	DI(SrcNone | Priv, mwait),
4327 	N, N, N, N, N, N,
4328 };
4329 
4330 static const struct opcode group7_rm2[] = {
4331 	N,
4332 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4333 	N, N, N, N, N, N,
4334 };
4335 
4336 static const struct opcode group7_rm3[] = {
4337 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4338 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4339 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4340 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4341 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4342 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4343 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4344 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4345 };
4346 
4347 static const struct opcode group7_rm7[] = {
4348 	N,
4349 	DIP(SrcNone, rdtscp, check_rdtsc),
4350 	N, N, N, N, N, N,
4351 };
4352 
4353 static const struct opcode group1[] = {
4354 	F(Lock, em_add),
4355 	F(Lock | PageTable, em_or),
4356 	F(Lock, em_adc),
4357 	F(Lock, em_sbb),
4358 	F(Lock | PageTable, em_and),
4359 	F(Lock, em_sub),
4360 	F(Lock, em_xor),
4361 	F(NoWrite, em_cmp),
4362 };
4363 
4364 static const struct opcode group1A[] = {
4365 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4366 };
4367 
4368 static const struct opcode group2[] = {
4369 	F(DstMem | ModRM, em_rol),
4370 	F(DstMem | ModRM, em_ror),
4371 	F(DstMem | ModRM, em_rcl),
4372 	F(DstMem | ModRM, em_rcr),
4373 	F(DstMem | ModRM, em_shl),
4374 	F(DstMem | ModRM, em_shr),
4375 	F(DstMem | ModRM, em_shl),
4376 	F(DstMem | ModRM, em_sar),
4377 };
4378 
4379 static const struct opcode group3[] = {
4380 	F(DstMem | SrcImm | NoWrite, em_test),
4381 	F(DstMem | SrcImm | NoWrite, em_test),
4382 	F(DstMem | SrcNone | Lock, em_not),
4383 	F(DstMem | SrcNone | Lock, em_neg),
4384 	F(DstXacc | Src2Mem, em_mul_ex),
4385 	F(DstXacc | Src2Mem, em_imul_ex),
4386 	F(DstXacc | Src2Mem, em_div_ex),
4387 	F(DstXacc | Src2Mem, em_idiv_ex),
4388 };
4389 
4390 static const struct opcode group4[] = {
4391 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4392 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4393 	N, N, N, N, N, N,
4394 };
4395 
4396 static const struct opcode group5[] = {
4397 	F(DstMem | SrcNone | Lock,		em_inc),
4398 	F(DstMem | SrcNone | Lock,		em_dec),
4399 	I(SrcMem | NearBranch | IsBranch,       em_call_near_abs),
4400 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_call_far),
4401 	I(SrcMem | NearBranch | IsBranch,       em_jmp_abs),
4402 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_jmp_far),
4403 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4404 };
4405 
4406 static const struct opcode group6[] = {
4407 	II(Prot | DstMem,	   em_sldt, sldt),
4408 	II(Prot | DstMem,	   em_str, str),
4409 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4410 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4411 	N, N, N, N,
4412 };
4413 
4414 static const struct group_dual group7 = { {
4415 	II(Mov | DstMem,			em_sgdt, sgdt),
4416 	II(Mov | DstMem,			em_sidt, sidt),
4417 	II(SrcMem | Priv,			em_lgdt, lgdt),
4418 	II(SrcMem | Priv,			em_lidt, lidt),
4419 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4420 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4421 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4422 }, {
4423 	EXT(0, group7_rm0),
4424 	EXT(0, group7_rm1),
4425 	EXT(0, group7_rm2),
4426 	EXT(0, group7_rm3),
4427 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4428 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4429 	EXT(0, group7_rm7),
4430 } };
4431 
4432 static const struct opcode group8[] = {
4433 	N, N, N, N,
4434 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4435 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4436 	F(DstMem | SrcImmByte | Lock,			em_btr),
4437 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4438 };
4439 
4440 /*
4441  * The "memory" destination is actually always a register, since we come
4442  * from the register case of group9.
4443  */
4444 static const struct gprefix pfx_0f_c7_7 = {
4445 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4446 };
4447 
4448 
4449 static const struct group_dual group9 = { {
4450 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4451 }, {
4452 	N, N, N, N, N, N, N,
4453 	GP(0, &pfx_0f_c7_7),
4454 } };
4455 
4456 static const struct opcode group11[] = {
4457 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4458 	X7(D(Undefined)),
4459 };
4460 
4461 static const struct gprefix pfx_0f_ae_7 = {
4462 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4463 };
4464 
4465 static const struct group_dual group15 = { {
4466 	I(ModRM | Aligned16, em_fxsave),
4467 	I(ModRM | Aligned16, em_fxrstor),
4468 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4469 }, {
4470 	N, N, N, N, N, N, N, N,
4471 } };
4472 
4473 static const struct gprefix pfx_0f_6f_0f_7f = {
4474 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4475 };
4476 
4477 static const struct instr_dual instr_dual_0f_2b = {
4478 	I(0, em_mov), N
4479 };
4480 
4481 static const struct gprefix pfx_0f_2b = {
4482 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4483 };
4484 
4485 static const struct gprefix pfx_0f_10_0f_11 = {
4486 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4487 };
4488 
4489 static const struct gprefix pfx_0f_28_0f_29 = {
4490 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4491 };
4492 
4493 static const struct gprefix pfx_0f_e7 = {
4494 	N, I(Sse, em_mov), N, N,
4495 };
4496 
4497 static const struct escape escape_d9 = { {
4498 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4499 }, {
4500 	/* 0xC0 - 0xC7 */
4501 	N, N, N, N, N, N, N, N,
4502 	/* 0xC8 - 0xCF */
4503 	N, N, N, N, N, N, N, N,
4504 	/* 0xD0 - 0xC7 */
4505 	N, N, N, N, N, N, N, N,
4506 	/* 0xD8 - 0xDF */
4507 	N, N, N, N, N, N, N, N,
4508 	/* 0xE0 - 0xE7 */
4509 	N, N, N, N, N, N, N, N,
4510 	/* 0xE8 - 0xEF */
4511 	N, N, N, N, N, N, N, N,
4512 	/* 0xF0 - 0xF7 */
4513 	N, N, N, N, N, N, N, N,
4514 	/* 0xF8 - 0xFF */
4515 	N, N, N, N, N, N, N, N,
4516 } };
4517 
4518 static const struct escape escape_db = { {
4519 	N, N, N, N, N, N, N, N,
4520 }, {
4521 	/* 0xC0 - 0xC7 */
4522 	N, N, N, N, N, N, N, N,
4523 	/* 0xC8 - 0xCF */
4524 	N, N, N, N, N, N, N, N,
4525 	/* 0xD0 - 0xC7 */
4526 	N, N, N, N, N, N, N, N,
4527 	/* 0xD8 - 0xDF */
4528 	N, N, N, N, N, N, N, N,
4529 	/* 0xE0 - 0xE7 */
4530 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4531 	/* 0xE8 - 0xEF */
4532 	N, N, N, N, N, N, N, N,
4533 	/* 0xF0 - 0xF7 */
4534 	N, N, N, N, N, N, N, N,
4535 	/* 0xF8 - 0xFF */
4536 	N, N, N, N, N, N, N, N,
4537 } };
4538 
4539 static const struct escape escape_dd = { {
4540 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4541 }, {
4542 	/* 0xC0 - 0xC7 */
4543 	N, N, N, N, N, N, N, N,
4544 	/* 0xC8 - 0xCF */
4545 	N, N, N, N, N, N, N, N,
4546 	/* 0xD0 - 0xC7 */
4547 	N, N, N, N, N, N, N, N,
4548 	/* 0xD8 - 0xDF */
4549 	N, N, N, N, N, N, N, N,
4550 	/* 0xE0 - 0xE7 */
4551 	N, N, N, N, N, N, N, N,
4552 	/* 0xE8 - 0xEF */
4553 	N, N, N, N, N, N, N, N,
4554 	/* 0xF0 - 0xF7 */
4555 	N, N, N, N, N, N, N, N,
4556 	/* 0xF8 - 0xFF */
4557 	N, N, N, N, N, N, N, N,
4558 } };
4559 
4560 static const struct instr_dual instr_dual_0f_c3 = {
4561 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4562 };
4563 
4564 static const struct mode_dual mode_dual_63 = {
4565 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4566 };
4567 
4568 static const struct opcode opcode_table[256] = {
4569 	/* 0x00 - 0x07 */
4570 	F6ALU(Lock, em_add),
4571 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4572 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4573 	/* 0x08 - 0x0F */
4574 	F6ALU(Lock | PageTable, em_or),
4575 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4576 	N,
4577 	/* 0x10 - 0x17 */
4578 	F6ALU(Lock, em_adc),
4579 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4580 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4581 	/* 0x18 - 0x1F */
4582 	F6ALU(Lock, em_sbb),
4583 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4584 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4585 	/* 0x20 - 0x27 */
4586 	F6ALU(Lock | PageTable, em_and), N, N,
4587 	/* 0x28 - 0x2F */
4588 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4589 	/* 0x30 - 0x37 */
4590 	F6ALU(Lock, em_xor), N, N,
4591 	/* 0x38 - 0x3F */
4592 	F6ALU(NoWrite, em_cmp), N, N,
4593 	/* 0x40 - 0x4F */
4594 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4595 	/* 0x50 - 0x57 */
4596 	X8(I(SrcReg | Stack, em_push)),
4597 	/* 0x58 - 0x5F */
4598 	X8(I(DstReg | Stack, em_pop)),
4599 	/* 0x60 - 0x67 */
4600 	I(ImplicitOps | Stack | No64, em_pusha),
4601 	I(ImplicitOps | Stack | No64, em_popa),
4602 	N, MD(ModRM, &mode_dual_63),
4603 	N, N, N, N,
4604 	/* 0x68 - 0x6F */
4605 	I(SrcImm | Mov | Stack, em_push),
4606 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4607 	I(SrcImmByte | Mov | Stack, em_push),
4608 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4609 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4610 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4611 	/* 0x70 - 0x7F */
4612 	X16(D(SrcImmByte | NearBranch | IsBranch)),
4613 	/* 0x80 - 0x87 */
4614 	G(ByteOp | DstMem | SrcImm, group1),
4615 	G(DstMem | SrcImm, group1),
4616 	G(ByteOp | DstMem | SrcImm | No64, group1),
4617 	G(DstMem | SrcImmByte, group1),
4618 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4619 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4620 	/* 0x88 - 0x8F */
4621 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4622 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4623 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4624 	D(ModRM | SrcMem | NoAccess | DstReg),
4625 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4626 	G(0, group1A),
4627 	/* 0x90 - 0x97 */
4628 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4629 	/* 0x98 - 0x9F */
4630 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4631 	I(SrcImmFAddr | No64 | IsBranch, em_call_far), N,
4632 	II(ImplicitOps | Stack, em_pushf, pushf),
4633 	II(ImplicitOps | Stack, em_popf, popf),
4634 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4635 	/* 0xA0 - 0xA7 */
4636 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4637 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4638 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4639 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4640 	/* 0xA8 - 0xAF */
4641 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4642 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4643 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4644 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4645 	/* 0xB0 - 0xB7 */
4646 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4647 	/* 0xB8 - 0xBF */
4648 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4649 	/* 0xC0 - 0xC7 */
4650 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4651 	I(ImplicitOps | NearBranch | SrcImmU16 | IsBranch, em_ret_near_imm),
4652 	I(ImplicitOps | NearBranch | IsBranch, em_ret),
4653 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4654 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4655 	G(ByteOp, group11), G(0, group11),
4656 	/* 0xC8 - 0xCF */
4657 	I(Stack | SrcImmU16 | Src2ImmByte | IsBranch, em_enter),
4658 	I(Stack | IsBranch, em_leave),
4659 	I(ImplicitOps | SrcImmU16 | IsBranch, em_ret_far_imm),
4660 	I(ImplicitOps | IsBranch, em_ret_far),
4661 	D(ImplicitOps | IsBranch), DI(SrcImmByte | IsBranch, intn),
4662 	D(ImplicitOps | No64 | IsBranch),
4663 	II(ImplicitOps | IsBranch, em_iret, iret),
4664 	/* 0xD0 - 0xD7 */
4665 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4666 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4667 	I(DstAcc | SrcImmUByte | No64, em_aam),
4668 	I(DstAcc | SrcImmUByte | No64, em_aad),
4669 	F(DstAcc | ByteOp | No64, em_salc),
4670 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4671 	/* 0xD8 - 0xDF */
4672 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4673 	/* 0xE0 - 0xE7 */
4674 	X3(I(SrcImmByte | NearBranch | IsBranch, em_loop)),
4675 	I(SrcImmByte | NearBranch | IsBranch, em_jcxz),
4676 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4677 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4678 	/* 0xE8 - 0xEF */
4679 	I(SrcImm | NearBranch | IsBranch, em_call),
4680 	D(SrcImm | ImplicitOps | NearBranch | IsBranch),
4681 	I(SrcImmFAddr | No64 | IsBranch, em_jmp_far),
4682 	D(SrcImmByte | ImplicitOps | NearBranch | IsBranch),
4683 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4684 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4685 	/* 0xF0 - 0xF7 */
4686 	N, DI(ImplicitOps, icebp), N, N,
4687 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4688 	G(ByteOp, group3), G(0, group3),
4689 	/* 0xF8 - 0xFF */
4690 	D(ImplicitOps), D(ImplicitOps),
4691 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4692 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4693 };
4694 
4695 static const struct opcode twobyte_table[256] = {
4696 	/* 0x00 - 0x0F */
4697 	G(0, group6), GD(0, &group7), N, N,
4698 	N, I(ImplicitOps | EmulateOnUD | IsBranch, em_syscall),
4699 	II(ImplicitOps | Priv, em_clts, clts), N,
4700 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4701 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4702 	/* 0x10 - 0x1F */
4703 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4704 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4705 	N, N, N, N, N, N,
4706 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4707 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4708 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4709 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4710 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4711 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4712 	/* 0x20 - 0x2F */
4713 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4714 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4715 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4716 						check_cr_access),
4717 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4718 						check_dr_write),
4719 	N, N, N, N,
4720 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4721 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4722 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4723 	N, N, N, N,
4724 	/* 0x30 - 0x3F */
4725 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4726 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4727 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4728 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4729 	I(ImplicitOps | EmulateOnUD | IsBranch, em_sysenter),
4730 	I(ImplicitOps | Priv | EmulateOnUD | IsBranch, em_sysexit),
4731 	N, N,
4732 	N, N, N, N, N, N, N, N,
4733 	/* 0x40 - 0x4F */
4734 	X16(D(DstReg | SrcMem | ModRM)),
4735 	/* 0x50 - 0x5F */
4736 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4737 	/* 0x60 - 0x6F */
4738 	N, N, N, N,
4739 	N, N, N, N,
4740 	N, N, N, N,
4741 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4742 	/* 0x70 - 0x7F */
4743 	N, N, N, N,
4744 	N, N, N, N,
4745 	N, N, N, N,
4746 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4747 	/* 0x80 - 0x8F */
4748 	X16(D(SrcImm | NearBranch | IsBranch)),
4749 	/* 0x90 - 0x9F */
4750 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4751 	/* 0xA0 - 0xA7 */
4752 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4753 	II(ImplicitOps, em_cpuid, cpuid),
4754 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4755 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4756 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4757 	/* 0xA8 - 0xAF */
4758 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4759 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4760 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4761 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4762 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4763 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4764 	/* 0xB0 - 0xB7 */
4765 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4766 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4767 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4768 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4769 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4770 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4771 	/* 0xB8 - 0xBF */
4772 	N, N,
4773 	G(BitOp, group8),
4774 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4775 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4776 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4777 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4778 	/* 0xC0 - 0xC7 */
4779 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4780 	N, ID(0, &instr_dual_0f_c3),
4781 	N, N, N, GD(0, &group9),
4782 	/* 0xC8 - 0xCF */
4783 	X8(I(DstReg, em_bswap)),
4784 	/* 0xD0 - 0xDF */
4785 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4786 	/* 0xE0 - 0xEF */
4787 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4788 	N, N, N, N, N, N, N, N,
4789 	/* 0xF0 - 0xFF */
4790 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4791 };
4792 
4793 static const struct instr_dual instr_dual_0f_38_f0 = {
4794 	I(DstReg | SrcMem | Mov, em_movbe), N
4795 };
4796 
4797 static const struct instr_dual instr_dual_0f_38_f1 = {
4798 	I(DstMem | SrcReg | Mov, em_movbe), N
4799 };
4800 
4801 static const struct gprefix three_byte_0f_38_f0 = {
4802 	ID(0, &instr_dual_0f_38_f0), N, N, N
4803 };
4804 
4805 static const struct gprefix three_byte_0f_38_f1 = {
4806 	ID(0, &instr_dual_0f_38_f1), N, N, N
4807 };
4808 
4809 /*
4810  * Insns below are selected by the prefix which indexed by the third opcode
4811  * byte.
4812  */
4813 static const struct opcode opcode_map_0f_38[256] = {
4814 	/* 0x00 - 0x7f */
4815 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4816 	/* 0x80 - 0xef */
4817 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4818 	/* 0xf0 - 0xf1 */
4819 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4820 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4821 	/* 0xf2 - 0xff */
4822 	N, N, X4(N), X8(N)
4823 };
4824 
4825 #undef D
4826 #undef N
4827 #undef G
4828 #undef GD
4829 #undef I
4830 #undef GP
4831 #undef EXT
4832 #undef MD
4833 #undef ID
4834 
4835 #undef D2bv
4836 #undef D2bvIP
4837 #undef I2bv
4838 #undef I2bvIP
4839 #undef I6ALU
4840 
4841 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4842 {
4843 	unsigned size;
4844 
4845 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4846 	if (size == 8)
4847 		size = 4;
4848 	return size;
4849 }
4850 
4851 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4852 		      unsigned size, bool sign_extension)
4853 {
4854 	int rc = X86EMUL_CONTINUE;
4855 
4856 	op->type = OP_IMM;
4857 	op->bytes = size;
4858 	op->addr.mem.ea = ctxt->_eip;
4859 	/* NB. Immediates are sign-extended as necessary. */
4860 	switch (op->bytes) {
4861 	case 1:
4862 		op->val = insn_fetch(s8, ctxt);
4863 		break;
4864 	case 2:
4865 		op->val = insn_fetch(s16, ctxt);
4866 		break;
4867 	case 4:
4868 		op->val = insn_fetch(s32, ctxt);
4869 		break;
4870 	case 8:
4871 		op->val = insn_fetch(s64, ctxt);
4872 		break;
4873 	}
4874 	if (!sign_extension) {
4875 		switch (op->bytes) {
4876 		case 1:
4877 			op->val &= 0xff;
4878 			break;
4879 		case 2:
4880 			op->val &= 0xffff;
4881 			break;
4882 		case 4:
4883 			op->val &= 0xffffffff;
4884 			break;
4885 		}
4886 	}
4887 done:
4888 	return rc;
4889 }
4890 
4891 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4892 			  unsigned d)
4893 {
4894 	int rc = X86EMUL_CONTINUE;
4895 
4896 	switch (d) {
4897 	case OpReg:
4898 		decode_register_operand(ctxt, op);
4899 		break;
4900 	case OpImmUByte:
4901 		rc = decode_imm(ctxt, op, 1, false);
4902 		break;
4903 	case OpMem:
4904 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4905 	mem_common:
4906 		*op = ctxt->memop;
4907 		ctxt->memopp = op;
4908 		if (ctxt->d & BitOp)
4909 			fetch_bit_operand(ctxt);
4910 		op->orig_val = op->val;
4911 		break;
4912 	case OpMem64:
4913 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4914 		goto mem_common;
4915 	case OpAcc:
4916 		op->type = OP_REG;
4917 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4918 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4919 		fetch_register_operand(op);
4920 		op->orig_val = op->val;
4921 		break;
4922 	case OpAccLo:
4923 		op->type = OP_REG;
4924 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4925 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4926 		fetch_register_operand(op);
4927 		op->orig_val = op->val;
4928 		break;
4929 	case OpAccHi:
4930 		if (ctxt->d & ByteOp) {
4931 			op->type = OP_NONE;
4932 			break;
4933 		}
4934 		op->type = OP_REG;
4935 		op->bytes = ctxt->op_bytes;
4936 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4937 		fetch_register_operand(op);
4938 		op->orig_val = op->val;
4939 		break;
4940 	case OpDI:
4941 		op->type = OP_MEM;
4942 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4943 		op->addr.mem.ea =
4944 			register_address(ctxt, VCPU_REGS_RDI);
4945 		op->addr.mem.seg = VCPU_SREG_ES;
4946 		op->val = 0;
4947 		op->count = 1;
4948 		break;
4949 	case OpDX:
4950 		op->type = OP_REG;
4951 		op->bytes = 2;
4952 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4953 		fetch_register_operand(op);
4954 		break;
4955 	case OpCL:
4956 		op->type = OP_IMM;
4957 		op->bytes = 1;
4958 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4959 		break;
4960 	case OpImmByte:
4961 		rc = decode_imm(ctxt, op, 1, true);
4962 		break;
4963 	case OpOne:
4964 		op->type = OP_IMM;
4965 		op->bytes = 1;
4966 		op->val = 1;
4967 		break;
4968 	case OpImm:
4969 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4970 		break;
4971 	case OpImm64:
4972 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4973 		break;
4974 	case OpMem8:
4975 		ctxt->memop.bytes = 1;
4976 		if (ctxt->memop.type == OP_REG) {
4977 			ctxt->memop.addr.reg = decode_register(ctxt,
4978 					ctxt->modrm_rm, true);
4979 			fetch_register_operand(&ctxt->memop);
4980 		}
4981 		goto mem_common;
4982 	case OpMem16:
4983 		ctxt->memop.bytes = 2;
4984 		goto mem_common;
4985 	case OpMem32:
4986 		ctxt->memop.bytes = 4;
4987 		goto mem_common;
4988 	case OpImmU16:
4989 		rc = decode_imm(ctxt, op, 2, false);
4990 		break;
4991 	case OpImmU:
4992 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4993 		break;
4994 	case OpSI:
4995 		op->type = OP_MEM;
4996 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4997 		op->addr.mem.ea =
4998 			register_address(ctxt, VCPU_REGS_RSI);
4999 		op->addr.mem.seg = ctxt->seg_override;
5000 		op->val = 0;
5001 		op->count = 1;
5002 		break;
5003 	case OpXLat:
5004 		op->type = OP_MEM;
5005 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5006 		op->addr.mem.ea =
5007 			address_mask(ctxt,
5008 				reg_read(ctxt, VCPU_REGS_RBX) +
5009 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
5010 		op->addr.mem.seg = ctxt->seg_override;
5011 		op->val = 0;
5012 		break;
5013 	case OpImmFAddr:
5014 		op->type = OP_IMM;
5015 		op->addr.mem.ea = ctxt->_eip;
5016 		op->bytes = ctxt->op_bytes + 2;
5017 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
5018 		break;
5019 	case OpMemFAddr:
5020 		ctxt->memop.bytes = ctxt->op_bytes + 2;
5021 		goto mem_common;
5022 	case OpES:
5023 		op->type = OP_IMM;
5024 		op->val = VCPU_SREG_ES;
5025 		break;
5026 	case OpCS:
5027 		op->type = OP_IMM;
5028 		op->val = VCPU_SREG_CS;
5029 		break;
5030 	case OpSS:
5031 		op->type = OP_IMM;
5032 		op->val = VCPU_SREG_SS;
5033 		break;
5034 	case OpDS:
5035 		op->type = OP_IMM;
5036 		op->val = VCPU_SREG_DS;
5037 		break;
5038 	case OpFS:
5039 		op->type = OP_IMM;
5040 		op->val = VCPU_SREG_FS;
5041 		break;
5042 	case OpGS:
5043 		op->type = OP_IMM;
5044 		op->val = VCPU_SREG_GS;
5045 		break;
5046 	case OpImplicit:
5047 		/* Special instructions do their own operand decoding. */
5048 	default:
5049 		op->type = OP_NONE; /* Disable writeback. */
5050 		break;
5051 	}
5052 
5053 done:
5054 	return rc;
5055 }
5056 
5057 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
5058 {
5059 	int rc = X86EMUL_CONTINUE;
5060 	int mode = ctxt->mode;
5061 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
5062 	bool op_prefix = false;
5063 	bool has_seg_override = false;
5064 	struct opcode opcode;
5065 	u16 dummy;
5066 	struct desc_struct desc;
5067 
5068 	ctxt->memop.type = OP_NONE;
5069 	ctxt->memopp = NULL;
5070 	ctxt->_eip = ctxt->eip;
5071 	ctxt->fetch.ptr = ctxt->fetch.data;
5072 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
5073 	ctxt->opcode_len = 1;
5074 	ctxt->intercept = x86_intercept_none;
5075 	if (insn_len > 0)
5076 		memcpy(ctxt->fetch.data, insn, insn_len);
5077 	else {
5078 		rc = __do_insn_fetch_bytes(ctxt, 1);
5079 		if (rc != X86EMUL_CONTINUE)
5080 			goto done;
5081 	}
5082 
5083 	switch (mode) {
5084 	case X86EMUL_MODE_REAL:
5085 	case X86EMUL_MODE_VM86:
5086 		def_op_bytes = def_ad_bytes = 2;
5087 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
5088 		if (desc.d)
5089 			def_op_bytes = def_ad_bytes = 4;
5090 		break;
5091 	case X86EMUL_MODE_PROT16:
5092 		def_op_bytes = def_ad_bytes = 2;
5093 		break;
5094 	case X86EMUL_MODE_PROT32:
5095 		def_op_bytes = def_ad_bytes = 4;
5096 		break;
5097 #ifdef CONFIG_X86_64
5098 	case X86EMUL_MODE_PROT64:
5099 		def_op_bytes = 4;
5100 		def_ad_bytes = 8;
5101 		break;
5102 #endif
5103 	default:
5104 		return EMULATION_FAILED;
5105 	}
5106 
5107 	ctxt->op_bytes = def_op_bytes;
5108 	ctxt->ad_bytes = def_ad_bytes;
5109 
5110 	/* Legacy prefixes. */
5111 	for (;;) {
5112 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5113 		case 0x66:	/* operand-size override */
5114 			op_prefix = true;
5115 			/* switch between 2/4 bytes */
5116 			ctxt->op_bytes = def_op_bytes ^ 6;
5117 			break;
5118 		case 0x67:	/* address-size override */
5119 			if (mode == X86EMUL_MODE_PROT64)
5120 				/* switch between 4/8 bytes */
5121 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5122 			else
5123 				/* switch between 2/4 bytes */
5124 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5125 			break;
5126 		case 0x26:	/* ES override */
5127 			has_seg_override = true;
5128 			ctxt->seg_override = VCPU_SREG_ES;
5129 			break;
5130 		case 0x2e:	/* CS override */
5131 			has_seg_override = true;
5132 			ctxt->seg_override = VCPU_SREG_CS;
5133 			break;
5134 		case 0x36:	/* SS override */
5135 			has_seg_override = true;
5136 			ctxt->seg_override = VCPU_SREG_SS;
5137 			break;
5138 		case 0x3e:	/* DS override */
5139 			has_seg_override = true;
5140 			ctxt->seg_override = VCPU_SREG_DS;
5141 			break;
5142 		case 0x64:	/* FS override */
5143 			has_seg_override = true;
5144 			ctxt->seg_override = VCPU_SREG_FS;
5145 			break;
5146 		case 0x65:	/* GS override */
5147 			has_seg_override = true;
5148 			ctxt->seg_override = VCPU_SREG_GS;
5149 			break;
5150 		case 0x40 ... 0x4f: /* REX */
5151 			if (mode != X86EMUL_MODE_PROT64)
5152 				goto done_prefixes;
5153 			ctxt->rex_prefix = ctxt->b;
5154 			continue;
5155 		case 0xf0:	/* LOCK */
5156 			ctxt->lock_prefix = 1;
5157 			break;
5158 		case 0xf2:	/* REPNE/REPNZ */
5159 		case 0xf3:	/* REP/REPE/REPZ */
5160 			ctxt->rep_prefix = ctxt->b;
5161 			break;
5162 		default:
5163 			goto done_prefixes;
5164 		}
5165 
5166 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5167 
5168 		ctxt->rex_prefix = 0;
5169 	}
5170 
5171 done_prefixes:
5172 
5173 	/* REX prefix. */
5174 	if (ctxt->rex_prefix & 8)
5175 		ctxt->op_bytes = 8;	/* REX.W */
5176 
5177 	/* Opcode byte(s). */
5178 	opcode = opcode_table[ctxt->b];
5179 	/* Two-byte opcode? */
5180 	if (ctxt->b == 0x0f) {
5181 		ctxt->opcode_len = 2;
5182 		ctxt->b = insn_fetch(u8, ctxt);
5183 		opcode = twobyte_table[ctxt->b];
5184 
5185 		/* 0F_38 opcode map */
5186 		if (ctxt->b == 0x38) {
5187 			ctxt->opcode_len = 3;
5188 			ctxt->b = insn_fetch(u8, ctxt);
5189 			opcode = opcode_map_0f_38[ctxt->b];
5190 		}
5191 	}
5192 	ctxt->d = opcode.flags;
5193 
5194 	if (ctxt->d & ModRM)
5195 		ctxt->modrm = insn_fetch(u8, ctxt);
5196 
5197 	/* vex-prefix instructions are not implemented */
5198 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5199 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5200 		ctxt->d = NotImpl;
5201 	}
5202 
5203 	while (ctxt->d & GroupMask) {
5204 		switch (ctxt->d & GroupMask) {
5205 		case Group:
5206 			goffset = (ctxt->modrm >> 3) & 7;
5207 			opcode = opcode.u.group[goffset];
5208 			break;
5209 		case GroupDual:
5210 			goffset = (ctxt->modrm >> 3) & 7;
5211 			if ((ctxt->modrm >> 6) == 3)
5212 				opcode = opcode.u.gdual->mod3[goffset];
5213 			else
5214 				opcode = opcode.u.gdual->mod012[goffset];
5215 			break;
5216 		case RMExt:
5217 			goffset = ctxt->modrm & 7;
5218 			opcode = opcode.u.group[goffset];
5219 			break;
5220 		case Prefix:
5221 			if (ctxt->rep_prefix && op_prefix)
5222 				return EMULATION_FAILED;
5223 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5224 			switch (simd_prefix) {
5225 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5226 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5227 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5228 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5229 			}
5230 			break;
5231 		case Escape:
5232 			if (ctxt->modrm > 0xbf) {
5233 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
5234 				u32 index = array_index_nospec(
5235 					ctxt->modrm - 0xc0, size);
5236 
5237 				opcode = opcode.u.esc->high[index];
5238 			} else {
5239 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5240 			}
5241 			break;
5242 		case InstrDual:
5243 			if ((ctxt->modrm >> 6) == 3)
5244 				opcode = opcode.u.idual->mod3;
5245 			else
5246 				opcode = opcode.u.idual->mod012;
5247 			break;
5248 		case ModeDual:
5249 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5250 				opcode = opcode.u.mdual->mode64;
5251 			else
5252 				opcode = opcode.u.mdual->mode32;
5253 			break;
5254 		default:
5255 			return EMULATION_FAILED;
5256 		}
5257 
5258 		ctxt->d &= ~(u64)GroupMask;
5259 		ctxt->d |= opcode.flags;
5260 	}
5261 
5262 	ctxt->is_branch = opcode.flags & IsBranch;
5263 
5264 	/* Unrecognised? */
5265 	if (ctxt->d == 0)
5266 		return EMULATION_FAILED;
5267 
5268 	ctxt->execute = opcode.u.execute;
5269 
5270 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
5271 	    likely(!(ctxt->d & EmulateOnUD)))
5272 		return EMULATION_FAILED;
5273 
5274 	if (unlikely(ctxt->d &
5275 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5276 	     No16))) {
5277 		/*
5278 		 * These are copied unconditionally here, and checked unconditionally
5279 		 * in x86_emulate_insn.
5280 		 */
5281 		ctxt->check_perm = opcode.check_perm;
5282 		ctxt->intercept = opcode.intercept;
5283 
5284 		if (ctxt->d & NotImpl)
5285 			return EMULATION_FAILED;
5286 
5287 		if (mode == X86EMUL_MODE_PROT64) {
5288 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5289 				ctxt->op_bytes = 8;
5290 			else if (ctxt->d & NearBranch)
5291 				ctxt->op_bytes = 8;
5292 		}
5293 
5294 		if (ctxt->d & Op3264) {
5295 			if (mode == X86EMUL_MODE_PROT64)
5296 				ctxt->op_bytes = 8;
5297 			else
5298 				ctxt->op_bytes = 4;
5299 		}
5300 
5301 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5302 			ctxt->op_bytes = 4;
5303 
5304 		if (ctxt->d & Sse)
5305 			ctxt->op_bytes = 16;
5306 		else if (ctxt->d & Mmx)
5307 			ctxt->op_bytes = 8;
5308 	}
5309 
5310 	/* ModRM and SIB bytes. */
5311 	if (ctxt->d & ModRM) {
5312 		rc = decode_modrm(ctxt, &ctxt->memop);
5313 		if (!has_seg_override) {
5314 			has_seg_override = true;
5315 			ctxt->seg_override = ctxt->modrm_seg;
5316 		}
5317 	} else if (ctxt->d & MemAbs)
5318 		rc = decode_abs(ctxt, &ctxt->memop);
5319 	if (rc != X86EMUL_CONTINUE)
5320 		goto done;
5321 
5322 	if (!has_seg_override)
5323 		ctxt->seg_override = VCPU_SREG_DS;
5324 
5325 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5326 
5327 	/*
5328 	 * Decode and fetch the source operand: register, memory
5329 	 * or immediate.
5330 	 */
5331 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5332 	if (rc != X86EMUL_CONTINUE)
5333 		goto done;
5334 
5335 	/*
5336 	 * Decode and fetch the second source operand: register, memory
5337 	 * or immediate.
5338 	 */
5339 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5340 	if (rc != X86EMUL_CONTINUE)
5341 		goto done;
5342 
5343 	/* Decode and fetch the destination operand: register or memory. */
5344 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5345 
5346 	if (ctxt->rip_relative && likely(ctxt->memopp))
5347 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5348 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5349 
5350 done:
5351 	if (rc == X86EMUL_PROPAGATE_FAULT)
5352 		ctxt->have_exception = true;
5353 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5354 }
5355 
5356 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5357 {
5358 	return ctxt->d & PageTable;
5359 }
5360 
5361 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5362 {
5363 	/* The second termination condition only applies for REPE
5364 	 * and REPNE. Test if the repeat string operation prefix is
5365 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5366 	 * corresponding termination condition according to:
5367 	 * 	- if REPE/REPZ and ZF = 0 then done
5368 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5369 	 */
5370 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5371 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5372 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5373 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5374 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5375 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5376 		return true;
5377 
5378 	return false;
5379 }
5380 
5381 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5382 {
5383 	int rc;
5384 
5385 	kvm_fpu_get();
5386 	rc = asm_safe("fwait");
5387 	kvm_fpu_put();
5388 
5389 	if (unlikely(rc != X86EMUL_CONTINUE))
5390 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5391 
5392 	return X86EMUL_CONTINUE;
5393 }
5394 
5395 static void fetch_possible_mmx_operand(struct operand *op)
5396 {
5397 	if (op->type == OP_MM)
5398 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5399 }
5400 
5401 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5402 {
5403 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5404 
5405 	if (!(ctxt->d & ByteOp))
5406 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5407 
5408 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5409 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5410 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5411 	    : "c"(ctxt->src2.val));
5412 
5413 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5414 	if (!fop) /* exception is returned in fop variable */
5415 		return emulate_de(ctxt);
5416 	return X86EMUL_CONTINUE;
5417 }
5418 
5419 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5420 {
5421 	/* Clear fields that are set conditionally but read without a guard. */
5422 	ctxt->rip_relative = false;
5423 	ctxt->rex_prefix = 0;
5424 	ctxt->lock_prefix = 0;
5425 	ctxt->rep_prefix = 0;
5426 	ctxt->regs_valid = 0;
5427 	ctxt->regs_dirty = 0;
5428 
5429 	ctxt->io_read.pos = 0;
5430 	ctxt->io_read.end = 0;
5431 	ctxt->mem_read.end = 0;
5432 }
5433 
5434 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5435 {
5436 	const struct x86_emulate_ops *ops = ctxt->ops;
5437 	int rc = X86EMUL_CONTINUE;
5438 	int saved_dst_type = ctxt->dst.type;
5439 	unsigned emul_flags;
5440 
5441 	ctxt->mem_read.pos = 0;
5442 
5443 	/* LOCK prefix is allowed only with some instructions */
5444 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5445 		rc = emulate_ud(ctxt);
5446 		goto done;
5447 	}
5448 
5449 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5450 		rc = emulate_ud(ctxt);
5451 		goto done;
5452 	}
5453 
5454 	emul_flags = ctxt->ops->get_hflags(ctxt);
5455 	if (unlikely(ctxt->d &
5456 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5457 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5458 				(ctxt->d & Undefined)) {
5459 			rc = emulate_ud(ctxt);
5460 			goto done;
5461 		}
5462 
5463 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5464 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5465 			rc = emulate_ud(ctxt);
5466 			goto done;
5467 		}
5468 
5469 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5470 			rc = emulate_nm(ctxt);
5471 			goto done;
5472 		}
5473 
5474 		if (ctxt->d & Mmx) {
5475 			rc = flush_pending_x87_faults(ctxt);
5476 			if (rc != X86EMUL_CONTINUE)
5477 				goto done;
5478 			/*
5479 			 * Now that we know the fpu is exception safe, we can fetch
5480 			 * operands from it.
5481 			 */
5482 			fetch_possible_mmx_operand(&ctxt->src);
5483 			fetch_possible_mmx_operand(&ctxt->src2);
5484 			if (!(ctxt->d & Mov))
5485 				fetch_possible_mmx_operand(&ctxt->dst);
5486 		}
5487 
5488 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5489 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5490 						      X86_ICPT_PRE_EXCEPT);
5491 			if (rc != X86EMUL_CONTINUE)
5492 				goto done;
5493 		}
5494 
5495 		/* Instruction can only be executed in protected mode */
5496 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5497 			rc = emulate_ud(ctxt);
5498 			goto done;
5499 		}
5500 
5501 		/* Privileged instruction can be executed only in CPL=0 */
5502 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5503 			if (ctxt->d & PrivUD)
5504 				rc = emulate_ud(ctxt);
5505 			else
5506 				rc = emulate_gp(ctxt, 0);
5507 			goto done;
5508 		}
5509 
5510 		/* Do instruction specific permission checks */
5511 		if (ctxt->d & CheckPerm) {
5512 			rc = ctxt->check_perm(ctxt);
5513 			if (rc != X86EMUL_CONTINUE)
5514 				goto done;
5515 		}
5516 
5517 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5518 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5519 						      X86_ICPT_POST_EXCEPT);
5520 			if (rc != X86EMUL_CONTINUE)
5521 				goto done;
5522 		}
5523 
5524 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5525 			/* All REP prefixes have the same first termination condition */
5526 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5527 				string_registers_quirk(ctxt);
5528 				ctxt->eip = ctxt->_eip;
5529 				ctxt->eflags &= ~X86_EFLAGS_RF;
5530 				goto done;
5531 			}
5532 		}
5533 	}
5534 
5535 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5536 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5537 				    ctxt->src.valptr, ctxt->src.bytes);
5538 		if (rc != X86EMUL_CONTINUE)
5539 			goto done;
5540 		ctxt->src.orig_val64 = ctxt->src.val64;
5541 	}
5542 
5543 	if (ctxt->src2.type == OP_MEM) {
5544 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5545 				    &ctxt->src2.val, ctxt->src2.bytes);
5546 		if (rc != X86EMUL_CONTINUE)
5547 			goto done;
5548 	}
5549 
5550 	if ((ctxt->d & DstMask) == ImplicitOps)
5551 		goto special_insn;
5552 
5553 
5554 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5555 		/* optimisation - avoid slow emulated read if Mov */
5556 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5557 				   &ctxt->dst.val, ctxt->dst.bytes);
5558 		if (rc != X86EMUL_CONTINUE) {
5559 			if (!(ctxt->d & NoWrite) &&
5560 			    rc == X86EMUL_PROPAGATE_FAULT &&
5561 			    ctxt->exception.vector == PF_VECTOR)
5562 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5563 			goto done;
5564 		}
5565 	}
5566 	/* Copy full 64-bit value for CMPXCHG8B.  */
5567 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5568 
5569 special_insn:
5570 
5571 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5572 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5573 					      X86_ICPT_POST_MEMACCESS);
5574 		if (rc != X86EMUL_CONTINUE)
5575 			goto done;
5576 	}
5577 
5578 	if (ctxt->rep_prefix && (ctxt->d & String))
5579 		ctxt->eflags |= X86_EFLAGS_RF;
5580 	else
5581 		ctxt->eflags &= ~X86_EFLAGS_RF;
5582 
5583 	if (ctxt->execute) {
5584 		if (ctxt->d & Fastop)
5585 			rc = fastop(ctxt, ctxt->fop);
5586 		else
5587 			rc = ctxt->execute(ctxt);
5588 		if (rc != X86EMUL_CONTINUE)
5589 			goto done;
5590 		goto writeback;
5591 	}
5592 
5593 	if (ctxt->opcode_len == 2)
5594 		goto twobyte_insn;
5595 	else if (ctxt->opcode_len == 3)
5596 		goto threebyte_insn;
5597 
5598 	switch (ctxt->b) {
5599 	case 0x70 ... 0x7f: /* jcc (short) */
5600 		if (test_cc(ctxt->b, ctxt->eflags))
5601 			rc = jmp_rel(ctxt, ctxt->src.val);
5602 		break;
5603 	case 0x8d: /* lea r16/r32, m */
5604 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5605 		break;
5606 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5607 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5608 			ctxt->dst.type = OP_NONE;
5609 		else
5610 			rc = em_xchg(ctxt);
5611 		break;
5612 	case 0x98: /* cbw/cwde/cdqe */
5613 		switch (ctxt->op_bytes) {
5614 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5615 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5616 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5617 		}
5618 		break;
5619 	case 0xcc:		/* int3 */
5620 		rc = emulate_int(ctxt, 3);
5621 		break;
5622 	case 0xcd:		/* int n */
5623 		rc = emulate_int(ctxt, ctxt->src.val);
5624 		break;
5625 	case 0xce:		/* into */
5626 		if (ctxt->eflags & X86_EFLAGS_OF)
5627 			rc = emulate_int(ctxt, 4);
5628 		break;
5629 	case 0xe9: /* jmp rel */
5630 	case 0xeb: /* jmp rel short */
5631 		rc = jmp_rel(ctxt, ctxt->src.val);
5632 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5633 		break;
5634 	case 0xf4:              /* hlt */
5635 		ctxt->ops->halt(ctxt);
5636 		break;
5637 	case 0xf5:	/* cmc */
5638 		/* complement carry flag from eflags reg */
5639 		ctxt->eflags ^= X86_EFLAGS_CF;
5640 		break;
5641 	case 0xf8: /* clc */
5642 		ctxt->eflags &= ~X86_EFLAGS_CF;
5643 		break;
5644 	case 0xf9: /* stc */
5645 		ctxt->eflags |= X86_EFLAGS_CF;
5646 		break;
5647 	case 0xfc: /* cld */
5648 		ctxt->eflags &= ~X86_EFLAGS_DF;
5649 		break;
5650 	case 0xfd: /* std */
5651 		ctxt->eflags |= X86_EFLAGS_DF;
5652 		break;
5653 	default:
5654 		goto cannot_emulate;
5655 	}
5656 
5657 	if (rc != X86EMUL_CONTINUE)
5658 		goto done;
5659 
5660 writeback:
5661 	if (ctxt->d & SrcWrite) {
5662 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5663 		rc = writeback(ctxt, &ctxt->src);
5664 		if (rc != X86EMUL_CONTINUE)
5665 			goto done;
5666 	}
5667 	if (!(ctxt->d & NoWrite)) {
5668 		rc = writeback(ctxt, &ctxt->dst);
5669 		if (rc != X86EMUL_CONTINUE)
5670 			goto done;
5671 	}
5672 
5673 	/*
5674 	 * restore dst type in case the decoding will be reused
5675 	 * (happens for string instruction )
5676 	 */
5677 	ctxt->dst.type = saved_dst_type;
5678 
5679 	if ((ctxt->d & SrcMask) == SrcSI)
5680 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5681 
5682 	if ((ctxt->d & DstMask) == DstDI)
5683 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5684 
5685 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5686 		unsigned int count;
5687 		struct read_cache *r = &ctxt->io_read;
5688 		if ((ctxt->d & SrcMask) == SrcSI)
5689 			count = ctxt->src.count;
5690 		else
5691 			count = ctxt->dst.count;
5692 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5693 
5694 		if (!string_insn_completed(ctxt)) {
5695 			/*
5696 			 * Re-enter guest when pio read ahead buffer is empty
5697 			 * or, if it is not used, after each 1024 iteration.
5698 			 */
5699 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5700 			    (r->end == 0 || r->end != r->pos)) {
5701 				/*
5702 				 * Reset read cache. Usually happens before
5703 				 * decode, but since instruction is restarted
5704 				 * we have to do it here.
5705 				 */
5706 				ctxt->mem_read.end = 0;
5707 				writeback_registers(ctxt);
5708 				return EMULATION_RESTART;
5709 			}
5710 			goto done; /* skip rip writeback */
5711 		}
5712 		ctxt->eflags &= ~X86_EFLAGS_RF;
5713 	}
5714 
5715 	ctxt->eip = ctxt->_eip;
5716 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5717 		ctxt->eip = (u32)ctxt->_eip;
5718 
5719 done:
5720 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5721 		WARN_ON(ctxt->exception.vector > 0x1f);
5722 		ctxt->have_exception = true;
5723 	}
5724 	if (rc == X86EMUL_INTERCEPTED)
5725 		return EMULATION_INTERCEPTED;
5726 
5727 	if (rc == X86EMUL_CONTINUE)
5728 		writeback_registers(ctxt);
5729 
5730 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5731 
5732 twobyte_insn:
5733 	switch (ctxt->b) {
5734 	case 0x09:		/* wbinvd */
5735 		(ctxt->ops->wbinvd)(ctxt);
5736 		break;
5737 	case 0x08:		/* invd */
5738 	case 0x0d:		/* GrpP (prefetch) */
5739 	case 0x18:		/* Grp16 (prefetch/nop) */
5740 	case 0x1f:		/* nop */
5741 		break;
5742 	case 0x20: /* mov cr, reg */
5743 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5744 		break;
5745 	case 0x21: /* mov from dr to reg */
5746 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5747 		break;
5748 	case 0x40 ... 0x4f:	/* cmov */
5749 		if (test_cc(ctxt->b, ctxt->eflags))
5750 			ctxt->dst.val = ctxt->src.val;
5751 		else if (ctxt->op_bytes != 4)
5752 			ctxt->dst.type = OP_NONE; /* no writeback */
5753 		break;
5754 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5755 		if (test_cc(ctxt->b, ctxt->eflags))
5756 			rc = jmp_rel(ctxt, ctxt->src.val);
5757 		break;
5758 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5759 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5760 		break;
5761 	case 0xb6 ... 0xb7:	/* movzx */
5762 		ctxt->dst.bytes = ctxt->op_bytes;
5763 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5764 						       : (u16) ctxt->src.val;
5765 		break;
5766 	case 0xbe ... 0xbf:	/* movsx */
5767 		ctxt->dst.bytes = ctxt->op_bytes;
5768 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5769 							(s16) ctxt->src.val;
5770 		break;
5771 	default:
5772 		goto cannot_emulate;
5773 	}
5774 
5775 threebyte_insn:
5776 
5777 	if (rc != X86EMUL_CONTINUE)
5778 		goto done;
5779 
5780 	goto writeback;
5781 
5782 cannot_emulate:
5783 	return EMULATION_FAILED;
5784 }
5785 
5786 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5787 {
5788 	invalidate_registers(ctxt);
5789 }
5790 
5791 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5792 {
5793 	writeback_registers(ctxt);
5794 }
5795 
5796 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5797 {
5798 	if (ctxt->rep_prefix && (ctxt->d & String))
5799 		return false;
5800 
5801 	if (ctxt->d & TwoMemOp)
5802 		return false;
5803 
5804 	return true;
5805 }
5806