xref: /linux/arch/x86/kvm/emulate.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include <linux/kvm_host.h>
23 #include "kvm_cache_regs.h"
24 #include "kvm_emulate.h"
25 #include <linux/stringify.h>
26 #include <asm/debugreg.h>
27 #include <asm/nospec-branch.h>
28 #include <asm/ibt.h>
29 
30 #include "x86.h"
31 #include "tss.h"
32 #include "mmu.h"
33 #include "pmu.h"
34 
35 /*
36  * Operand types
37  */
38 #define OpNone             0ull
39 #define OpImplicit         1ull  /* No generic decode */
40 #define OpReg              2ull  /* Register */
41 #define OpMem              3ull  /* Memory */
42 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
43 #define OpDI               5ull  /* ES:DI/EDI/RDI */
44 #define OpMem64            6ull  /* Memory, 64-bit */
45 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
46 #define OpDX               8ull  /* DX register */
47 #define OpCL               9ull  /* CL register (for shifts) */
48 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
49 #define OpOne             11ull  /* Implied 1 */
50 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
51 #define OpMem16           13ull  /* Memory operand (16-bit). */
52 #define OpMem32           14ull  /* Memory operand (32-bit). */
53 #define OpImmU            15ull  /* Immediate operand, zero extended */
54 #define OpSI              16ull  /* SI/ESI/RSI */
55 #define OpImmFAddr        17ull  /* Immediate far address */
56 #define OpMemFAddr        18ull  /* Far address in memory */
57 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
58 #define OpES              20ull  /* ES */
59 #define OpCS              21ull  /* CS */
60 #define OpSS              22ull  /* SS */
61 #define OpDS              23ull  /* DS */
62 #define OpFS              24ull  /* FS */
63 #define OpGS              25ull  /* GS */
64 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
65 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
66 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
67 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
68 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
69 
70 #define OpBits             5  /* Width of operand field */
71 #define OpMask             ((1ull << OpBits) - 1)
72 
73 /*
74  * Opcode effective-address decode tables.
75  * Note that we only emulate instructions that have at least one memory
76  * operand (excluding implicit stack references). We assume that stack
77  * references and instruction fetches will never occur in special memory
78  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
79  * not be handled.
80  */
81 
82 /* Operand sizes: 8-bit operands or specified/overridden size. */
83 #define ByteOp      (1<<0)	/* 8-bit operands. */
84 /* Destination operand type. */
85 #define DstShift    1
86 #define ImplicitOps (OpImplicit << DstShift)
87 #define DstReg      (OpReg << DstShift)
88 #define DstMem      (OpMem << DstShift)
89 #define DstAcc      (OpAcc << DstShift)
90 #define DstDI       (OpDI << DstShift)
91 #define DstMem64    (OpMem64 << DstShift)
92 #define DstMem16    (OpMem16 << DstShift)
93 #define DstImmUByte (OpImmUByte << DstShift)
94 #define DstDX       (OpDX << DstShift)
95 #define DstAccLo    (OpAccLo << DstShift)
96 #define DstMask     (OpMask << DstShift)
97 /* Source operand type. */
98 #define SrcShift    6
99 #define SrcNone     (OpNone << SrcShift)
100 #define SrcReg      (OpReg << SrcShift)
101 #define SrcMem      (OpMem << SrcShift)
102 #define SrcMem16    (OpMem16 << SrcShift)
103 #define SrcMem32    (OpMem32 << SrcShift)
104 #define SrcImm      (OpImm << SrcShift)
105 #define SrcImmByte  (OpImmByte << SrcShift)
106 #define SrcOne      (OpOne << SrcShift)
107 #define SrcImmUByte (OpImmUByte << SrcShift)
108 #define SrcImmU     (OpImmU << SrcShift)
109 #define SrcSI       (OpSI << SrcShift)
110 #define SrcXLat     (OpXLat << SrcShift)
111 #define SrcImmFAddr (OpImmFAddr << SrcShift)
112 #define SrcMemFAddr (OpMemFAddr << SrcShift)
113 #define SrcAcc      (OpAcc << SrcShift)
114 #define SrcImmU16   (OpImmU16 << SrcShift)
115 #define SrcImm64    (OpImm64 << SrcShift)
116 #define SrcDX       (OpDX << SrcShift)
117 #define SrcMem8     (OpMem8 << SrcShift)
118 #define SrcAccHi    (OpAccHi << SrcShift)
119 #define SrcMask     (OpMask << SrcShift)
120 #define BitOp       (1<<11)
121 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
122 #define String      (1<<13)     /* String instruction (rep capable) */
123 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
124 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
125 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
126 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
127 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
128 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
129 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
130 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
131 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
132 #define Sse         (1<<18)     /* SSE Vector instruction */
133 /* Generic ModRM decode. */
134 #define ModRM       (1<<19)
135 /* Destination is only written; never read. */
136 #define Mov         (1<<20)
137 /* Misc flags */
138 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
139 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
140 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
141 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
142 #define Undefined   (1<<25) /* No Such Instruction */
143 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
144 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
145 #define No64	    (1<<28)
146 #define PageTable   (1 << 29)   /* instruction used to write page table */
147 #define NotImpl     (1 << 30)   /* instruction is not implemented */
148 /* Source 2 operand type */
149 #define Src2Shift   (31)
150 #define Src2None    (OpNone << Src2Shift)
151 #define Src2Mem     (OpMem << Src2Shift)
152 #define Src2CL      (OpCL << Src2Shift)
153 #define Src2ImmByte (OpImmByte << Src2Shift)
154 #define Src2One     (OpOne << Src2Shift)
155 #define Src2Imm     (OpImm << Src2Shift)
156 #define Src2ES      (OpES << Src2Shift)
157 #define Src2CS      (OpCS << Src2Shift)
158 #define Src2SS      (OpSS << Src2Shift)
159 #define Src2DS      (OpDS << Src2Shift)
160 #define Src2FS      (OpFS << Src2Shift)
161 #define Src2GS      (OpGS << Src2Shift)
162 #define Src2Mask    (OpMask << Src2Shift)
163 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
164 #define AlignMask   ((u64)7 << 41)
165 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
166 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
167 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
168 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
169 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
170 #define NoWrite     ((u64)1 << 45)  /* No writeback */
171 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
172 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
173 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
174 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
175 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
176 #define NearBranch  ((u64)1 << 52)  /* Near branches */
177 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
178 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
179 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
180 #define IsBranch    ((u64)1 << 56)  /* Instruction is considered a branch. */
181 
182 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
183 
184 #define X2(x...) x, x
185 #define X3(x...) X2(x), x
186 #define X4(x...) X2(x), X2(x)
187 #define X5(x...) X4(x), x
188 #define X6(x...) X4(x), X2(x)
189 #define X7(x...) X4(x), X3(x)
190 #define X8(x...) X4(x), X4(x)
191 #define X16(x...) X8(x), X8(x)
192 
193 struct opcode {
194 	u64 flags;
195 	u8 intercept;
196 	u8 pad[7];
197 	union {
198 		int (*execute)(struct x86_emulate_ctxt *ctxt);
199 		const struct opcode *group;
200 		const struct group_dual *gdual;
201 		const struct gprefix *gprefix;
202 		const struct escape *esc;
203 		const struct instr_dual *idual;
204 		const struct mode_dual *mdual;
205 		void (*fastop)(struct fastop *fake);
206 	} u;
207 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
208 };
209 
210 struct group_dual {
211 	struct opcode mod012[8];
212 	struct opcode mod3[8];
213 };
214 
215 struct gprefix {
216 	struct opcode pfx_no;
217 	struct opcode pfx_66;
218 	struct opcode pfx_f2;
219 	struct opcode pfx_f3;
220 };
221 
222 struct escape {
223 	struct opcode op[8];
224 	struct opcode high[64];
225 };
226 
227 struct instr_dual {
228 	struct opcode mod012;
229 	struct opcode mod3;
230 };
231 
232 struct mode_dual {
233 	struct opcode mode32;
234 	struct opcode mode64;
235 };
236 
237 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
238 
239 enum x86_transfer_type {
240 	X86_TRANSFER_NONE,
241 	X86_TRANSFER_CALL_JMP,
242 	X86_TRANSFER_RET,
243 	X86_TRANSFER_TASK_SWITCH,
244 };
245 
246 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
247 {
248 	unsigned long dirty = ctxt->regs_dirty;
249 	unsigned reg;
250 
251 	for_each_set_bit(reg, &dirty, NR_EMULATOR_GPRS)
252 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
253 }
254 
255 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
256 {
257 	ctxt->regs_dirty = 0;
258 	ctxt->regs_valid = 0;
259 }
260 
261 /*
262  * These EFLAGS bits are restored from saved value during emulation, and
263  * any changes are written back to the saved value after emulation.
264  */
265 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
266 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
267 
268 #ifdef CONFIG_X86_64
269 #define ON64(x) x
270 #else
271 #define ON64(x)
272 #endif
273 
274 /*
275  * fastop functions have a special calling convention:
276  *
277  * dst:    rax        (in/out)
278  * src:    rdx        (in/out)
279  * src2:   rcx        (in)
280  * flags:  rflags     (in/out)
281  * ex:     rsi        (in:fastop pointer, out:zero if exception)
282  *
283  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
284  * different operand sizes can be reached by calculation, rather than a jump
285  * table (which would be bigger than the code).
286  *
287  * The 16 byte alignment, considering 5 bytes for the RET thunk, 3 for ENDBR
288  * and 1 for the straight line speculation INT3, leaves 7 bytes for the
289  * body of the function.  Currently none is larger than 4.
290  */
291 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
292 
293 #define FASTOP_SIZE	16
294 
295 #define __FOP_FUNC(name) \
296 	".align " __stringify(FASTOP_SIZE) " \n\t" \
297 	".type " name ", @function \n\t" \
298 	name ":\n\t" \
299 	ASM_ENDBR \
300 	IBT_NOSEAL(name)
301 
302 #define FOP_FUNC(name) \
303 	__FOP_FUNC(#name)
304 
305 #define __FOP_RET(name) \
306 	"11: " ASM_RET \
307 	".size " name ", .-" name "\n\t"
308 
309 #define FOP_RET(name) \
310 	__FOP_RET(#name)
311 
312 #define __FOP_START(op, align) \
313 	extern void em_##op(struct fastop *fake); \
314 	asm(".pushsection .text, \"ax\" \n\t" \
315 	    ".global em_" #op " \n\t" \
316 	    ".align " __stringify(align) " \n\t" \
317 	    "em_" #op ":\n\t"
318 
319 #define FOP_START(op) __FOP_START(op, FASTOP_SIZE)
320 
321 #define FOP_END \
322 	    ".popsection")
323 
324 #define __FOPNOP(name) \
325 	__FOP_FUNC(name) \
326 	__FOP_RET(name)
327 
328 #define FOPNOP() \
329 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
330 
331 #define FOP1E(op,  dst) \
332 	__FOP_FUNC(#op "_" #dst) \
333 	"10: " #op " %" #dst " \n\t" \
334 	__FOP_RET(#op "_" #dst)
335 
336 #define FOP1EEX(op,  dst) \
337 	FOP1E(op, dst) _ASM_EXTABLE_TYPE_REG(10b, 11b, EX_TYPE_ZERO_REG, %%esi)
338 
339 #define FASTOP1(op) \
340 	FOP_START(op) \
341 	FOP1E(op##b, al) \
342 	FOP1E(op##w, ax) \
343 	FOP1E(op##l, eax) \
344 	ON64(FOP1E(op##q, rax))	\
345 	FOP_END
346 
347 /* 1-operand, using src2 (for MUL/DIV r/m) */
348 #define FASTOP1SRC2(op, name) \
349 	FOP_START(name) \
350 	FOP1E(op, cl) \
351 	FOP1E(op, cx) \
352 	FOP1E(op, ecx) \
353 	ON64(FOP1E(op, rcx)) \
354 	FOP_END
355 
356 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
357 #define FASTOP1SRC2EX(op, name) \
358 	FOP_START(name) \
359 	FOP1EEX(op, cl) \
360 	FOP1EEX(op, cx) \
361 	FOP1EEX(op, ecx) \
362 	ON64(FOP1EEX(op, rcx)) \
363 	FOP_END
364 
365 #define FOP2E(op,  dst, src)	   \
366 	__FOP_FUNC(#op "_" #dst "_" #src) \
367 	#op " %" #src ", %" #dst " \n\t" \
368 	__FOP_RET(#op "_" #dst "_" #src)
369 
370 #define FASTOP2(op) \
371 	FOP_START(op) \
372 	FOP2E(op##b, al, dl) \
373 	FOP2E(op##w, ax, dx) \
374 	FOP2E(op##l, eax, edx) \
375 	ON64(FOP2E(op##q, rax, rdx)) \
376 	FOP_END
377 
378 /* 2 operand, word only */
379 #define FASTOP2W(op) \
380 	FOP_START(op) \
381 	FOPNOP() \
382 	FOP2E(op##w, ax, dx) \
383 	FOP2E(op##l, eax, edx) \
384 	ON64(FOP2E(op##q, rax, rdx)) \
385 	FOP_END
386 
387 /* 2 operand, src is CL */
388 #define FASTOP2CL(op) \
389 	FOP_START(op) \
390 	FOP2E(op##b, al, cl) \
391 	FOP2E(op##w, ax, cl) \
392 	FOP2E(op##l, eax, cl) \
393 	ON64(FOP2E(op##q, rax, cl)) \
394 	FOP_END
395 
396 /* 2 operand, src and dest are reversed */
397 #define FASTOP2R(op, name) \
398 	FOP_START(name) \
399 	FOP2E(op##b, dl, al) \
400 	FOP2E(op##w, dx, ax) \
401 	FOP2E(op##l, edx, eax) \
402 	ON64(FOP2E(op##q, rdx, rax)) \
403 	FOP_END
404 
405 #define FOP3E(op,  dst, src, src2) \
406 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
407 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
408 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
409 
410 /* 3-operand, word-only, src2=cl */
411 #define FASTOP3WCL(op) \
412 	FOP_START(op) \
413 	FOPNOP() \
414 	FOP3E(op##w, ax, dx, cl) \
415 	FOP3E(op##l, eax, edx, cl) \
416 	ON64(FOP3E(op##q, rax, rdx, cl)) \
417 	FOP_END
418 
419 /* Special case for SETcc - 1 instruction per cc */
420 #define FOP_SETCC(op) \
421 	FOP_FUNC(op) \
422 	#op " %al \n\t" \
423 	FOP_RET(op)
424 
425 FOP_START(setcc)
426 FOP_SETCC(seto)
427 FOP_SETCC(setno)
428 FOP_SETCC(setc)
429 FOP_SETCC(setnc)
430 FOP_SETCC(setz)
431 FOP_SETCC(setnz)
432 FOP_SETCC(setbe)
433 FOP_SETCC(setnbe)
434 FOP_SETCC(sets)
435 FOP_SETCC(setns)
436 FOP_SETCC(setp)
437 FOP_SETCC(setnp)
438 FOP_SETCC(setl)
439 FOP_SETCC(setnl)
440 FOP_SETCC(setle)
441 FOP_SETCC(setnle)
442 FOP_END;
443 
444 FOP_START(salc)
445 FOP_FUNC(salc)
446 "pushf; sbb %al, %al; popf \n\t"
447 FOP_RET(salc)
448 FOP_END;
449 
450 /*
451  * XXX: inoutclob user must know where the argument is being expanded.
452  *      Using asm goto would allow us to remove _fault.
453  */
454 #define asm_safe(insn, inoutclob...) \
455 ({ \
456 	int _fault = 0; \
457  \
458 	asm volatile("1:" insn "\n" \
459 	             "2:\n" \
460 		     _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_ONE_REG, %[_fault]) \
461 	             : [_fault] "+r"(_fault) inoutclob ); \
462  \
463 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
464 })
465 
466 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
467 				    enum x86_intercept intercept,
468 				    enum x86_intercept_stage stage)
469 {
470 	struct x86_instruction_info info = {
471 		.intercept  = intercept,
472 		.rep_prefix = ctxt->rep_prefix,
473 		.modrm_mod  = ctxt->modrm_mod,
474 		.modrm_reg  = ctxt->modrm_reg,
475 		.modrm_rm   = ctxt->modrm_rm,
476 		.src_val    = ctxt->src.val64,
477 		.dst_val    = ctxt->dst.val64,
478 		.src_bytes  = ctxt->src.bytes,
479 		.dst_bytes  = ctxt->dst.bytes,
480 		.ad_bytes   = ctxt->ad_bytes,
481 		.next_rip   = ctxt->eip,
482 	};
483 
484 	return ctxt->ops->intercept(ctxt, &info, stage);
485 }
486 
487 static void assign_masked(ulong *dest, ulong src, ulong mask)
488 {
489 	*dest = (*dest & ~mask) | (src & mask);
490 }
491 
492 static void assign_register(unsigned long *reg, u64 val, int bytes)
493 {
494 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
495 	switch (bytes) {
496 	case 1:
497 		*(u8 *)reg = (u8)val;
498 		break;
499 	case 2:
500 		*(u16 *)reg = (u16)val;
501 		break;
502 	case 4:
503 		*reg = (u32)val;
504 		break;	/* 64b: zero-extend */
505 	case 8:
506 		*reg = val;
507 		break;
508 	}
509 }
510 
511 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
512 {
513 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
514 }
515 
516 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
517 {
518 	u16 sel;
519 	struct desc_struct ss;
520 
521 	if (ctxt->mode == X86EMUL_MODE_PROT64)
522 		return ~0UL;
523 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
524 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
525 }
526 
527 static int stack_size(struct x86_emulate_ctxt *ctxt)
528 {
529 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
530 }
531 
532 /* Access/update address held in a register, based on addressing mode. */
533 static inline unsigned long
534 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
535 {
536 	if (ctxt->ad_bytes == sizeof(unsigned long))
537 		return reg;
538 	else
539 		return reg & ad_mask(ctxt);
540 }
541 
542 static inline unsigned long
543 register_address(struct x86_emulate_ctxt *ctxt, int reg)
544 {
545 	return address_mask(ctxt, reg_read(ctxt, reg));
546 }
547 
548 static void masked_increment(ulong *reg, ulong mask, int inc)
549 {
550 	assign_masked(reg, *reg + inc, mask);
551 }
552 
553 static inline void
554 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
555 {
556 	ulong *preg = reg_rmw(ctxt, reg);
557 
558 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
559 }
560 
561 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
562 {
563 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
564 }
565 
566 static u32 desc_limit_scaled(struct desc_struct *desc)
567 {
568 	u32 limit = get_desc_limit(desc);
569 
570 	return desc->g ? (limit << 12) | 0xfff : limit;
571 }
572 
573 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
574 {
575 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
576 		return 0;
577 
578 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
579 }
580 
581 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
582 			     u32 error, bool valid)
583 {
584 	if (KVM_EMULATOR_BUG_ON(vec > 0x1f, ctxt))
585 		return X86EMUL_UNHANDLEABLE;
586 
587 	ctxt->exception.vector = vec;
588 	ctxt->exception.error_code = error;
589 	ctxt->exception.error_code_valid = valid;
590 	return X86EMUL_PROPAGATE_FAULT;
591 }
592 
593 static int emulate_db(struct x86_emulate_ctxt *ctxt)
594 {
595 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
596 }
597 
598 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
599 {
600 	return emulate_exception(ctxt, GP_VECTOR, err, true);
601 }
602 
603 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
604 {
605 	return emulate_exception(ctxt, SS_VECTOR, err, true);
606 }
607 
608 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
609 {
610 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
611 }
612 
613 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
614 {
615 	return emulate_exception(ctxt, TS_VECTOR, err, true);
616 }
617 
618 static int emulate_de(struct x86_emulate_ctxt *ctxt)
619 {
620 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
621 }
622 
623 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
624 {
625 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
626 }
627 
628 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
629 {
630 	u16 selector;
631 	struct desc_struct desc;
632 
633 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
634 	return selector;
635 }
636 
637 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
638 				 unsigned seg)
639 {
640 	u16 dummy;
641 	u32 base3;
642 	struct desc_struct desc;
643 
644 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
645 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
646 }
647 
648 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
649 {
650 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
651 }
652 
653 static inline bool emul_is_noncanonical_address(u64 la,
654 						struct x86_emulate_ctxt *ctxt,
655 						unsigned int flags)
656 {
657 	return !ctxt->ops->is_canonical_addr(ctxt, la, flags);
658 }
659 
660 /*
661  * x86 defines three classes of vector instructions: explicitly
662  * aligned, explicitly unaligned, and the rest, which change behaviour
663  * depending on whether they're AVX encoded or not.
664  *
665  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
666  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
667  * 512 bytes of data must be aligned to a 16 byte boundary.
668  */
669 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
670 {
671 	u64 alignment = ctxt->d & AlignMask;
672 
673 	if (likely(size < 16))
674 		return 1;
675 
676 	switch (alignment) {
677 	case Unaligned:
678 	case Avx:
679 		return 1;
680 	case Aligned16:
681 		return 16;
682 	case Aligned:
683 	default:
684 		return size;
685 	}
686 }
687 
688 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
689 				       struct segmented_address addr,
690 				       unsigned *max_size, unsigned size,
691 				       enum x86emul_mode mode, ulong *linear,
692 				       unsigned int flags)
693 {
694 	struct desc_struct desc;
695 	bool usable;
696 	ulong la;
697 	u32 lim;
698 	u16 sel;
699 	u8  va_bits;
700 
701 	la = seg_base(ctxt, addr.seg) + addr.ea;
702 	*max_size = 0;
703 	switch (mode) {
704 	case X86EMUL_MODE_PROT64:
705 		*linear = la = ctxt->ops->get_untagged_addr(ctxt, la, flags);
706 		va_bits = ctxt_virt_addr_bits(ctxt);
707 		if (!__is_canonical_address(la, va_bits))
708 			goto bad;
709 
710 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
711 		if (size > *max_size)
712 			goto bad;
713 		break;
714 	default:
715 		*linear = la = (u32)la;
716 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
717 						addr.seg);
718 		if (!usable)
719 			goto bad;
720 		/* code segment in protected mode or read-only data segment */
721 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8)) || !(desc.type & 2)) &&
722 		    (flags & X86EMUL_F_WRITE))
723 			goto bad;
724 		/* unreadable code segment */
725 		if (!(flags & X86EMUL_F_FETCH) && (desc.type & 8) && !(desc.type & 2))
726 			goto bad;
727 		lim = desc_limit_scaled(&desc);
728 		if (!(desc.type & 8) && (desc.type & 4)) {
729 			/* expand-down segment */
730 			if (addr.ea <= lim)
731 				goto bad;
732 			lim = desc.d ? 0xffffffff : 0xffff;
733 		}
734 		if (addr.ea > lim)
735 			goto bad;
736 		if (lim == 0xffffffff)
737 			*max_size = ~0u;
738 		else {
739 			*max_size = (u64)lim + 1 - addr.ea;
740 			if (size > *max_size)
741 				goto bad;
742 		}
743 		break;
744 	}
745 	if (la & (insn_alignment(ctxt, size) - 1))
746 		return emulate_gp(ctxt, 0);
747 	return X86EMUL_CONTINUE;
748 bad:
749 	if (addr.seg == VCPU_SREG_SS)
750 		return emulate_ss(ctxt, 0);
751 	else
752 		return emulate_gp(ctxt, 0);
753 }
754 
755 static int linearize(struct x86_emulate_ctxt *ctxt,
756 		     struct segmented_address addr,
757 		     unsigned size, bool write,
758 		     ulong *linear)
759 {
760 	unsigned max_size;
761 	return __linearize(ctxt, addr, &max_size, size, ctxt->mode, linear,
762 			   write ? X86EMUL_F_WRITE : 0);
763 }
764 
765 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst)
766 {
767 	ulong linear;
768 	int rc;
769 	unsigned max_size;
770 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
771 					   .ea = dst };
772 
773 	if (ctxt->op_bytes != sizeof(unsigned long))
774 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
775 	rc = __linearize(ctxt, addr, &max_size, 1, ctxt->mode, &linear,
776 			 X86EMUL_F_FETCH);
777 	if (rc == X86EMUL_CONTINUE)
778 		ctxt->_eip = addr.ea;
779 	return rc;
780 }
781 
782 static inline int emulator_recalc_and_set_mode(struct x86_emulate_ctxt *ctxt)
783 {
784 	u64 efer;
785 	struct desc_struct cs;
786 	u16 selector;
787 	u32 base3;
788 
789 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
790 
791 	if (!(ctxt->ops->get_cr(ctxt, 0) & X86_CR0_PE)) {
792 		/* Real mode. cpu must not have long mode active */
793 		if (efer & EFER_LMA)
794 			return X86EMUL_UNHANDLEABLE;
795 		ctxt->mode = X86EMUL_MODE_REAL;
796 		return X86EMUL_CONTINUE;
797 	}
798 
799 	if (ctxt->eflags & X86_EFLAGS_VM) {
800 		/* Protected/VM86 mode. cpu must not have long mode active */
801 		if (efer & EFER_LMA)
802 			return X86EMUL_UNHANDLEABLE;
803 		ctxt->mode = X86EMUL_MODE_VM86;
804 		return X86EMUL_CONTINUE;
805 	}
806 
807 	if (!ctxt->ops->get_segment(ctxt, &selector, &cs, &base3, VCPU_SREG_CS))
808 		return X86EMUL_UNHANDLEABLE;
809 
810 	if (efer & EFER_LMA) {
811 		if (cs.l) {
812 			/* Proper long mode */
813 			ctxt->mode = X86EMUL_MODE_PROT64;
814 		} else if (cs.d) {
815 			/* 32 bit compatibility mode*/
816 			ctxt->mode = X86EMUL_MODE_PROT32;
817 		} else {
818 			ctxt->mode = X86EMUL_MODE_PROT16;
819 		}
820 	} else {
821 		/* Legacy 32 bit / 16 bit mode */
822 		ctxt->mode = cs.d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
823 	}
824 
825 	return X86EMUL_CONTINUE;
826 }
827 
828 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
829 {
830 	return assign_eip(ctxt, dst);
831 }
832 
833 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst)
834 {
835 	int rc = emulator_recalc_and_set_mode(ctxt);
836 
837 	if (rc != X86EMUL_CONTINUE)
838 		return rc;
839 
840 	return assign_eip(ctxt, dst);
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, ctxt->mode, &linear,
913 			 X86EMUL_F_FETCH);
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 + FASTOP_SIZE * (condition & 0xf);
1070 
1071 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1072 	asm("push %[flags]; popf; " CALL_NOSPEC
1073 	    : "=a"(rc), ASM_CALL_CONSTRAINT : [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 int reg;
1142 
1143 	if (ctxt->d & ModRM)
1144 		reg = ctxt->modrm_reg;
1145 	else
1146 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1147 
1148 	if (ctxt->d & Sse) {
1149 		op->type = OP_XMM;
1150 		op->bytes = 16;
1151 		op->addr.xmm = reg;
1152 		kvm_read_sse_reg(reg, &op->vec_val);
1153 		return;
1154 	}
1155 	if (ctxt->d & Mmx) {
1156 		reg &= 7;
1157 		op->type = OP_MM;
1158 		op->bytes = 8;
1159 		op->addr.mm = reg;
1160 		return;
1161 	}
1162 
1163 	op->type = OP_REG;
1164 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1165 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1166 
1167 	fetch_register_operand(op);
1168 	op->orig_val = op->val;
1169 }
1170 
1171 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1172 {
1173 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1174 		ctxt->modrm_seg = VCPU_SREG_SS;
1175 }
1176 
1177 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1178 			struct operand *op)
1179 {
1180 	u8 sib;
1181 	int index_reg, base_reg, scale;
1182 	int rc = X86EMUL_CONTINUE;
1183 	ulong modrm_ea = 0;
1184 
1185 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1186 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1187 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1188 
1189 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1190 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1191 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1192 	ctxt->modrm_seg = VCPU_SREG_DS;
1193 
1194 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1195 		op->type = OP_REG;
1196 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1197 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1198 				ctxt->d & ByteOp);
1199 		if (ctxt->d & Sse) {
1200 			op->type = OP_XMM;
1201 			op->bytes = 16;
1202 			op->addr.xmm = ctxt->modrm_rm;
1203 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1204 			return rc;
1205 		}
1206 		if (ctxt->d & Mmx) {
1207 			op->type = OP_MM;
1208 			op->bytes = 8;
1209 			op->addr.mm = ctxt->modrm_rm & 7;
1210 			return rc;
1211 		}
1212 		fetch_register_operand(op);
1213 		return rc;
1214 	}
1215 
1216 	op->type = OP_MEM;
1217 
1218 	if (ctxt->ad_bytes == 2) {
1219 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1220 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1221 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1222 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1223 
1224 		/* 16-bit ModR/M decode. */
1225 		switch (ctxt->modrm_mod) {
1226 		case 0:
1227 			if (ctxt->modrm_rm == 6)
1228 				modrm_ea += insn_fetch(u16, ctxt);
1229 			break;
1230 		case 1:
1231 			modrm_ea += insn_fetch(s8, ctxt);
1232 			break;
1233 		case 2:
1234 			modrm_ea += insn_fetch(u16, ctxt);
1235 			break;
1236 		}
1237 		switch (ctxt->modrm_rm) {
1238 		case 0:
1239 			modrm_ea += bx + si;
1240 			break;
1241 		case 1:
1242 			modrm_ea += bx + di;
1243 			break;
1244 		case 2:
1245 			modrm_ea += bp + si;
1246 			break;
1247 		case 3:
1248 			modrm_ea += bp + di;
1249 			break;
1250 		case 4:
1251 			modrm_ea += si;
1252 			break;
1253 		case 5:
1254 			modrm_ea += di;
1255 			break;
1256 		case 6:
1257 			if (ctxt->modrm_mod != 0)
1258 				modrm_ea += bp;
1259 			break;
1260 		case 7:
1261 			modrm_ea += bx;
1262 			break;
1263 		}
1264 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1265 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1266 			ctxt->modrm_seg = VCPU_SREG_SS;
1267 		modrm_ea = (u16)modrm_ea;
1268 	} else {
1269 		/* 32/64-bit ModR/M decode. */
1270 		if ((ctxt->modrm_rm & 7) == 4) {
1271 			sib = insn_fetch(u8, ctxt);
1272 			index_reg |= (sib >> 3) & 7;
1273 			base_reg |= sib & 7;
1274 			scale = sib >> 6;
1275 
1276 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1277 				modrm_ea += insn_fetch(s32, ctxt);
1278 			else {
1279 				modrm_ea += reg_read(ctxt, base_reg);
1280 				adjust_modrm_seg(ctxt, base_reg);
1281 				/* Increment ESP on POP [ESP] */
1282 				if ((ctxt->d & IncSP) &&
1283 				    base_reg == VCPU_REGS_RSP)
1284 					modrm_ea += ctxt->op_bytes;
1285 			}
1286 			if (index_reg != 4)
1287 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1288 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1289 			modrm_ea += insn_fetch(s32, ctxt);
1290 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1291 				ctxt->rip_relative = 1;
1292 		} else {
1293 			base_reg = ctxt->modrm_rm;
1294 			modrm_ea += reg_read(ctxt, base_reg);
1295 			adjust_modrm_seg(ctxt, base_reg);
1296 		}
1297 		switch (ctxt->modrm_mod) {
1298 		case 1:
1299 			modrm_ea += insn_fetch(s8, ctxt);
1300 			break;
1301 		case 2:
1302 			modrm_ea += insn_fetch(s32, ctxt);
1303 			break;
1304 		}
1305 	}
1306 	op->addr.mem.ea = modrm_ea;
1307 	if (ctxt->ad_bytes != 8)
1308 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1309 
1310 done:
1311 	return rc;
1312 }
1313 
1314 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1315 		      struct operand *op)
1316 {
1317 	int rc = X86EMUL_CONTINUE;
1318 
1319 	op->type = OP_MEM;
1320 	switch (ctxt->ad_bytes) {
1321 	case 2:
1322 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1323 		break;
1324 	case 4:
1325 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1326 		break;
1327 	case 8:
1328 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1329 		break;
1330 	}
1331 done:
1332 	return rc;
1333 }
1334 
1335 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1336 {
1337 	long sv = 0, mask;
1338 
1339 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1340 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1341 
1342 		if (ctxt->src.bytes == 2)
1343 			sv = (s16)ctxt->src.val & (s16)mask;
1344 		else if (ctxt->src.bytes == 4)
1345 			sv = (s32)ctxt->src.val & (s32)mask;
1346 		else
1347 			sv = (s64)ctxt->src.val & (s64)mask;
1348 
1349 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1350 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1351 	}
1352 
1353 	/* only subword offset */
1354 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1355 }
1356 
1357 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1358 			 unsigned long addr, void *dest, unsigned size)
1359 {
1360 	int rc;
1361 	struct read_cache *mc = &ctxt->mem_read;
1362 
1363 	if (mc->pos < mc->end)
1364 		goto read_cached;
1365 
1366 	if (KVM_EMULATOR_BUG_ON((mc->end + size) >= sizeof(mc->data), ctxt))
1367 		return X86EMUL_UNHANDLEABLE;
1368 
1369 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1370 				      &ctxt->exception);
1371 	if (rc != X86EMUL_CONTINUE)
1372 		return rc;
1373 
1374 	mc->end += size;
1375 
1376 read_cached:
1377 	memcpy(dest, mc->data + mc->pos, size);
1378 	mc->pos += size;
1379 	return X86EMUL_CONTINUE;
1380 }
1381 
1382 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1383 			  struct segmented_address addr,
1384 			  void *data,
1385 			  unsigned size)
1386 {
1387 	int rc;
1388 	ulong linear;
1389 
1390 	rc = linearize(ctxt, addr, size, false, &linear);
1391 	if (rc != X86EMUL_CONTINUE)
1392 		return rc;
1393 	return read_emulated(ctxt, linear, data, size);
1394 }
1395 
1396 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1397 			   struct segmented_address addr,
1398 			   const void *data,
1399 			   unsigned size)
1400 {
1401 	int rc;
1402 	ulong linear;
1403 
1404 	rc = linearize(ctxt, addr, size, true, &linear);
1405 	if (rc != X86EMUL_CONTINUE)
1406 		return rc;
1407 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1408 					 &ctxt->exception);
1409 }
1410 
1411 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1412 			     struct segmented_address addr,
1413 			     const void *orig_data, const void *data,
1414 			     unsigned size)
1415 {
1416 	int rc;
1417 	ulong linear;
1418 
1419 	rc = linearize(ctxt, addr, size, true, &linear);
1420 	if (rc != X86EMUL_CONTINUE)
1421 		return rc;
1422 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1423 					   size, &ctxt->exception);
1424 }
1425 
1426 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1427 			   unsigned int size, unsigned short port,
1428 			   void *dest)
1429 {
1430 	struct read_cache *rc = &ctxt->io_read;
1431 
1432 	if (rc->pos == rc->end) { /* refill pio read ahead */
1433 		unsigned int in_page, n;
1434 		unsigned int count = ctxt->rep_prefix ?
1435 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1436 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1437 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1438 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1439 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1440 		if (n == 0)
1441 			n = 1;
1442 		rc->pos = rc->end = 0;
1443 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1444 			return 0;
1445 		rc->end = n * size;
1446 	}
1447 
1448 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1449 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1450 		ctxt->dst.data = rc->data + rc->pos;
1451 		ctxt->dst.type = OP_MEM_STR;
1452 		ctxt->dst.count = (rc->end - rc->pos) / size;
1453 		rc->pos = rc->end;
1454 	} else {
1455 		memcpy(dest, rc->data + rc->pos, size);
1456 		rc->pos += size;
1457 	}
1458 	return 1;
1459 }
1460 
1461 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1462 				     u16 index, struct desc_struct *desc)
1463 {
1464 	struct desc_ptr dt;
1465 	ulong addr;
1466 
1467 	ctxt->ops->get_idt(ctxt, &dt);
1468 
1469 	if (dt.size < index * 8 + 7)
1470 		return emulate_gp(ctxt, index << 3 | 0x2);
1471 
1472 	addr = dt.address + index * 8;
1473 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1474 }
1475 
1476 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1477 				     u16 selector, struct desc_ptr *dt)
1478 {
1479 	const struct x86_emulate_ops *ops = ctxt->ops;
1480 	u32 base3 = 0;
1481 
1482 	if (selector & 1 << 2) {
1483 		struct desc_struct desc;
1484 		u16 sel;
1485 
1486 		memset(dt, 0, sizeof(*dt));
1487 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1488 				      VCPU_SREG_LDTR))
1489 			return;
1490 
1491 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1492 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1493 	} else
1494 		ops->get_gdt(ctxt, dt);
1495 }
1496 
1497 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1498 			      u16 selector, ulong *desc_addr_p)
1499 {
1500 	struct desc_ptr dt;
1501 	u16 index = selector >> 3;
1502 	ulong addr;
1503 
1504 	get_descriptor_table_ptr(ctxt, selector, &dt);
1505 
1506 	if (dt.size < index * 8 + 7)
1507 		return emulate_gp(ctxt, selector & 0xfffc);
1508 
1509 	addr = dt.address + index * 8;
1510 
1511 #ifdef CONFIG_X86_64
1512 	if (addr >> 32 != 0) {
1513 		u64 efer = 0;
1514 
1515 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1516 		if (!(efer & EFER_LMA))
1517 			addr &= (u32)-1;
1518 	}
1519 #endif
1520 
1521 	*desc_addr_p = addr;
1522 	return X86EMUL_CONTINUE;
1523 }
1524 
1525 /* allowed just for 8 bytes segments */
1526 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1527 				   u16 selector, struct desc_struct *desc,
1528 				   ulong *desc_addr_p)
1529 {
1530 	int rc;
1531 
1532 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1533 	if (rc != X86EMUL_CONTINUE)
1534 		return rc;
1535 
1536 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1537 }
1538 
1539 /* allowed just for 8 bytes segments */
1540 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1541 				    u16 selector, struct desc_struct *desc)
1542 {
1543 	int rc;
1544 	ulong addr;
1545 
1546 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1547 	if (rc != X86EMUL_CONTINUE)
1548 		return rc;
1549 
1550 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1551 }
1552 
1553 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1554 				     u16 selector, int seg, u8 cpl,
1555 				     enum x86_transfer_type transfer,
1556 				     struct desc_struct *desc)
1557 {
1558 	struct desc_struct seg_desc, old_desc;
1559 	u8 dpl, rpl;
1560 	unsigned err_vec = GP_VECTOR;
1561 	u32 err_code = 0;
1562 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1563 	ulong desc_addr;
1564 	int ret;
1565 	u16 dummy;
1566 	u32 base3 = 0;
1567 
1568 	memset(&seg_desc, 0, sizeof(seg_desc));
1569 
1570 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1571 		/* set real mode segment descriptor (keep limit etc. for
1572 		 * unreal mode) */
1573 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1574 		set_desc_base(&seg_desc, selector << 4);
1575 		goto load;
1576 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1577 		/* VM86 needs a clean new segment descriptor */
1578 		set_desc_base(&seg_desc, selector << 4);
1579 		set_desc_limit(&seg_desc, 0xffff);
1580 		seg_desc.type = 3;
1581 		seg_desc.p = 1;
1582 		seg_desc.s = 1;
1583 		seg_desc.dpl = 3;
1584 		goto load;
1585 	}
1586 
1587 	rpl = selector & 3;
1588 
1589 	/* TR should be in GDT only */
1590 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1591 		goto exception;
1592 
1593 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1594 	if (null_selector) {
1595 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1596 			goto exception;
1597 
1598 		if (seg == VCPU_SREG_SS) {
1599 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1600 				goto exception;
1601 
1602 			/*
1603 			 * ctxt->ops->set_segment expects the CPL to be in
1604 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1605 			 */
1606 			seg_desc.type = 3;
1607 			seg_desc.p = 1;
1608 			seg_desc.s = 1;
1609 			seg_desc.dpl = cpl;
1610 			seg_desc.d = 1;
1611 			seg_desc.g = 1;
1612 		}
1613 
1614 		/* Skip all following checks */
1615 		goto load;
1616 	}
1617 
1618 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1619 	if (ret != X86EMUL_CONTINUE)
1620 		return ret;
1621 
1622 	err_code = selector & 0xfffc;
1623 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1624 							   GP_VECTOR;
1625 
1626 	/* can't load system descriptor into segment selector */
1627 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1628 		if (transfer == X86_TRANSFER_CALL_JMP)
1629 			return X86EMUL_UNHANDLEABLE;
1630 		goto exception;
1631 	}
1632 
1633 	dpl = seg_desc.dpl;
1634 
1635 	switch (seg) {
1636 	case VCPU_SREG_SS:
1637 		/*
1638 		 * segment is not a writable data segment or segment
1639 		 * selector's RPL != CPL or DPL != CPL
1640 		 */
1641 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1642 			goto exception;
1643 		break;
1644 	case VCPU_SREG_CS:
1645 		/*
1646 		 * KVM uses "none" when loading CS as part of emulating Real
1647 		 * Mode exceptions and IRET (handled above).  In all other
1648 		 * cases, loading CS without a control transfer is a KVM bug.
1649 		 */
1650 		if (WARN_ON_ONCE(transfer == X86_TRANSFER_NONE))
1651 			goto exception;
1652 
1653 		if (!(seg_desc.type & 8))
1654 			goto exception;
1655 
1656 		if (transfer == X86_TRANSFER_RET) {
1657 			/* RET can never return to an inner privilege level. */
1658 			if (rpl < cpl)
1659 				goto exception;
1660 			/* Outer-privilege level return is not implemented */
1661 			if (rpl > cpl)
1662 				return X86EMUL_UNHANDLEABLE;
1663 		}
1664 		if (transfer == X86_TRANSFER_RET || transfer == X86_TRANSFER_TASK_SWITCH) {
1665 			if (seg_desc.type & 4) {
1666 				/* conforming */
1667 				if (dpl > rpl)
1668 					goto exception;
1669 			} else {
1670 				/* nonconforming */
1671 				if (dpl != rpl)
1672 					goto exception;
1673 			}
1674 		} else { /* X86_TRANSFER_CALL_JMP */
1675 			if (seg_desc.type & 4) {
1676 				/* conforming */
1677 				if (dpl > cpl)
1678 					goto exception;
1679 			} else {
1680 				/* nonconforming */
1681 				if (rpl > cpl || dpl != cpl)
1682 					goto exception;
1683 			}
1684 		}
1685 		/* in long-mode d/b must be clear if l is set */
1686 		if (seg_desc.d && seg_desc.l) {
1687 			u64 efer = 0;
1688 
1689 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1690 			if (efer & EFER_LMA)
1691 				goto exception;
1692 		}
1693 
1694 		/* CS(RPL) <- CPL */
1695 		selector = (selector & 0xfffc) | cpl;
1696 		break;
1697 	case VCPU_SREG_TR:
1698 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1699 			goto exception;
1700 		break;
1701 	case VCPU_SREG_LDTR:
1702 		if (seg_desc.s || seg_desc.type != 2)
1703 			goto exception;
1704 		break;
1705 	default: /*  DS, ES, FS, or GS */
1706 		/*
1707 		 * segment is not a data or readable code segment or
1708 		 * ((segment is a data or nonconforming code segment)
1709 		 * and ((RPL > DPL) or (CPL > DPL)))
1710 		 */
1711 		if ((seg_desc.type & 0xa) == 0x8 ||
1712 		    (((seg_desc.type & 0xc) != 0xc) &&
1713 		     (rpl > dpl || cpl > dpl)))
1714 			goto exception;
1715 		break;
1716 	}
1717 
1718 	if (!seg_desc.p) {
1719 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1720 		goto exception;
1721 	}
1722 
1723 	if (seg_desc.s) {
1724 		/* mark segment as accessed */
1725 		if (!(seg_desc.type & 1)) {
1726 			seg_desc.type |= 1;
1727 			ret = write_segment_descriptor(ctxt, selector,
1728 						       &seg_desc);
1729 			if (ret != X86EMUL_CONTINUE)
1730 				return ret;
1731 		}
1732 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1733 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1734 		if (ret != X86EMUL_CONTINUE)
1735 			return ret;
1736 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1737 						 ((u64)base3 << 32), ctxt,
1738 						 X86EMUL_F_DT_LOAD))
1739 			return emulate_gp(ctxt, err_code);
1740 	}
1741 
1742 	if (seg == VCPU_SREG_TR) {
1743 		old_desc = seg_desc;
1744 		seg_desc.type |= 2; /* busy */
1745 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1746 						  sizeof(seg_desc), &ctxt->exception);
1747 		if (ret != X86EMUL_CONTINUE)
1748 			return ret;
1749 	}
1750 load:
1751 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1752 	if (desc)
1753 		*desc = seg_desc;
1754 	return X86EMUL_CONTINUE;
1755 exception:
1756 	return emulate_exception(ctxt, err_vec, err_code, true);
1757 }
1758 
1759 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1760 				   u16 selector, int seg)
1761 {
1762 	u8 cpl = ctxt->ops->cpl(ctxt);
1763 
1764 	/*
1765 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1766 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1767 	 * but it's wrong).
1768 	 *
1769 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1770 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1771 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1772 	 * and only forbid it here.
1773 	 */
1774 	if (seg == VCPU_SREG_SS && selector == 3 &&
1775 	    ctxt->mode == X86EMUL_MODE_PROT64)
1776 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1777 
1778 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1779 					 X86_TRANSFER_NONE, NULL);
1780 }
1781 
1782 static void write_register_operand(struct operand *op)
1783 {
1784 	return assign_register(op->addr.reg, op->val, op->bytes);
1785 }
1786 
1787 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1788 {
1789 	switch (op->type) {
1790 	case OP_REG:
1791 		write_register_operand(op);
1792 		break;
1793 	case OP_MEM:
1794 		if (ctxt->lock_prefix)
1795 			return segmented_cmpxchg(ctxt,
1796 						 op->addr.mem,
1797 						 &op->orig_val,
1798 						 &op->val,
1799 						 op->bytes);
1800 		else
1801 			return segmented_write(ctxt,
1802 					       op->addr.mem,
1803 					       &op->val,
1804 					       op->bytes);
1805 	case OP_MEM_STR:
1806 		return segmented_write(ctxt,
1807 				       op->addr.mem,
1808 				       op->data,
1809 				       op->bytes * op->count);
1810 	case OP_XMM:
1811 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1812 		break;
1813 	case OP_MM:
1814 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1815 		break;
1816 	case OP_NONE:
1817 		/* no writeback */
1818 		break;
1819 	default:
1820 		break;
1821 	}
1822 	return X86EMUL_CONTINUE;
1823 }
1824 
1825 static int emulate_push(struct x86_emulate_ctxt *ctxt, const void *data, int len)
1826 {
1827 	struct segmented_address addr;
1828 
1829 	rsp_increment(ctxt, -len);
1830 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1831 	addr.seg = VCPU_SREG_SS;
1832 
1833 	return segmented_write(ctxt, addr, data, len);
1834 }
1835 
1836 static int em_push(struct x86_emulate_ctxt *ctxt)
1837 {
1838 	/* Disable writeback. */
1839 	ctxt->dst.type = OP_NONE;
1840 	return emulate_push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1841 }
1842 
1843 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1844 		       void *dest, int len)
1845 {
1846 	int rc;
1847 	struct segmented_address addr;
1848 
1849 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1850 	addr.seg = VCPU_SREG_SS;
1851 	rc = segmented_read(ctxt, addr, dest, len);
1852 	if (rc != X86EMUL_CONTINUE)
1853 		return rc;
1854 
1855 	rsp_increment(ctxt, len);
1856 	return rc;
1857 }
1858 
1859 static int em_pop(struct x86_emulate_ctxt *ctxt)
1860 {
1861 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1862 }
1863 
1864 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1865 			void *dest, int len)
1866 {
1867 	int rc;
1868 	unsigned long val = 0;
1869 	unsigned long change_mask;
1870 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1871 	int cpl = ctxt->ops->cpl(ctxt);
1872 
1873 	rc = emulate_pop(ctxt, &val, len);
1874 	if (rc != X86EMUL_CONTINUE)
1875 		return rc;
1876 
1877 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1878 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1879 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1880 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1881 
1882 	switch(ctxt->mode) {
1883 	case X86EMUL_MODE_PROT64:
1884 	case X86EMUL_MODE_PROT32:
1885 	case X86EMUL_MODE_PROT16:
1886 		if (cpl == 0)
1887 			change_mask |= X86_EFLAGS_IOPL;
1888 		if (cpl <= iopl)
1889 			change_mask |= X86_EFLAGS_IF;
1890 		break;
1891 	case X86EMUL_MODE_VM86:
1892 		if (iopl < 3)
1893 			return emulate_gp(ctxt, 0);
1894 		change_mask |= X86_EFLAGS_IF;
1895 		break;
1896 	default: /* real mode */
1897 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1898 		break;
1899 	}
1900 
1901 	*(unsigned long *)dest =
1902 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1903 
1904 	return rc;
1905 }
1906 
1907 static int em_popf(struct x86_emulate_ctxt *ctxt)
1908 {
1909 	ctxt->dst.type = OP_REG;
1910 	ctxt->dst.addr.reg = &ctxt->eflags;
1911 	ctxt->dst.bytes = ctxt->op_bytes;
1912 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1913 }
1914 
1915 static int em_enter(struct x86_emulate_ctxt *ctxt)
1916 {
1917 	int rc;
1918 	unsigned frame_size = ctxt->src.val;
1919 	unsigned nesting_level = ctxt->src2.val & 31;
1920 	ulong rbp;
1921 
1922 	if (nesting_level)
1923 		return X86EMUL_UNHANDLEABLE;
1924 
1925 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1926 	rc = emulate_push(ctxt, &rbp, stack_size(ctxt));
1927 	if (rc != X86EMUL_CONTINUE)
1928 		return rc;
1929 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1930 		      stack_mask(ctxt));
1931 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1932 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1933 		      stack_mask(ctxt));
1934 	return X86EMUL_CONTINUE;
1935 }
1936 
1937 static int em_leave(struct x86_emulate_ctxt *ctxt)
1938 {
1939 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1940 		      stack_mask(ctxt));
1941 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1942 }
1943 
1944 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1945 {
1946 	int seg = ctxt->src2.val;
1947 
1948 	ctxt->src.val = get_segment_selector(ctxt, seg);
1949 	if (ctxt->op_bytes == 4) {
1950 		rsp_increment(ctxt, -2);
1951 		ctxt->op_bytes = 2;
1952 	}
1953 
1954 	return em_push(ctxt);
1955 }
1956 
1957 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1958 {
1959 	int seg = ctxt->src2.val;
1960 	unsigned long selector = 0;
1961 	int rc;
1962 
1963 	rc = emulate_pop(ctxt, &selector, 2);
1964 	if (rc != X86EMUL_CONTINUE)
1965 		return rc;
1966 
1967 	if (seg == VCPU_SREG_SS)
1968 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1969 	if (ctxt->op_bytes > 2)
1970 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1971 
1972 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1973 	return rc;
1974 }
1975 
1976 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1977 {
1978 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1979 	int rc = X86EMUL_CONTINUE;
1980 	int reg = VCPU_REGS_RAX;
1981 
1982 	while (reg <= VCPU_REGS_RDI) {
1983 		(reg == VCPU_REGS_RSP) ?
1984 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1985 
1986 		rc = em_push(ctxt);
1987 		if (rc != X86EMUL_CONTINUE)
1988 			return rc;
1989 
1990 		++reg;
1991 	}
1992 
1993 	return rc;
1994 }
1995 
1996 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1997 {
1998 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1999 	return em_push(ctxt);
2000 }
2001 
2002 static int em_popa(struct x86_emulate_ctxt *ctxt)
2003 {
2004 	int rc = X86EMUL_CONTINUE;
2005 	int reg = VCPU_REGS_RDI;
2006 	u32 val = 0;
2007 
2008 	while (reg >= VCPU_REGS_RAX) {
2009 		if (reg == VCPU_REGS_RSP) {
2010 			rsp_increment(ctxt, ctxt->op_bytes);
2011 			--reg;
2012 		}
2013 
2014 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2015 		if (rc != X86EMUL_CONTINUE)
2016 			break;
2017 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2018 		--reg;
2019 	}
2020 	return rc;
2021 }
2022 
2023 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2024 {
2025 	const struct x86_emulate_ops *ops = ctxt->ops;
2026 	int rc;
2027 	struct desc_ptr dt;
2028 	gva_t cs_addr;
2029 	gva_t eip_addr;
2030 	u16 cs, eip;
2031 
2032 	/* TODO: Add limit checks */
2033 	ctxt->src.val = ctxt->eflags;
2034 	rc = em_push(ctxt);
2035 	if (rc != X86EMUL_CONTINUE)
2036 		return rc;
2037 
2038 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2039 
2040 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2041 	rc = em_push(ctxt);
2042 	if (rc != X86EMUL_CONTINUE)
2043 		return rc;
2044 
2045 	ctxt->src.val = ctxt->_eip;
2046 	rc = em_push(ctxt);
2047 	if (rc != X86EMUL_CONTINUE)
2048 		return rc;
2049 
2050 	ops->get_idt(ctxt, &dt);
2051 
2052 	eip_addr = dt.address + (irq << 2);
2053 	cs_addr = dt.address + (irq << 2) + 2;
2054 
2055 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2056 	if (rc != X86EMUL_CONTINUE)
2057 		return rc;
2058 
2059 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2060 	if (rc != X86EMUL_CONTINUE)
2061 		return rc;
2062 
2063 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2064 	if (rc != X86EMUL_CONTINUE)
2065 		return rc;
2066 
2067 	ctxt->_eip = eip;
2068 
2069 	return rc;
2070 }
2071 
2072 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2073 {
2074 	int rc;
2075 
2076 	invalidate_registers(ctxt);
2077 	rc = __emulate_int_real(ctxt, irq);
2078 	if (rc == X86EMUL_CONTINUE)
2079 		writeback_registers(ctxt);
2080 	return rc;
2081 }
2082 
2083 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2084 {
2085 	switch(ctxt->mode) {
2086 	case X86EMUL_MODE_REAL:
2087 		return __emulate_int_real(ctxt, irq);
2088 	case X86EMUL_MODE_VM86:
2089 	case X86EMUL_MODE_PROT16:
2090 	case X86EMUL_MODE_PROT32:
2091 	case X86EMUL_MODE_PROT64:
2092 	default:
2093 		/* Protected mode interrupts unimplemented yet */
2094 		return X86EMUL_UNHANDLEABLE;
2095 	}
2096 }
2097 
2098 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2099 {
2100 	int rc = X86EMUL_CONTINUE;
2101 	unsigned long temp_eip = 0;
2102 	unsigned long temp_eflags = 0;
2103 	unsigned long cs = 0;
2104 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2105 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2106 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2107 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2108 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2109 			     X86_EFLAGS_FIXED;
2110 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2111 				  X86_EFLAGS_VIP;
2112 
2113 	/* TODO: Add stack limit check */
2114 
2115 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2116 
2117 	if (rc != X86EMUL_CONTINUE)
2118 		return rc;
2119 
2120 	if (temp_eip & ~0xffff)
2121 		return emulate_gp(ctxt, 0);
2122 
2123 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2124 
2125 	if (rc != X86EMUL_CONTINUE)
2126 		return rc;
2127 
2128 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2129 
2130 	if (rc != X86EMUL_CONTINUE)
2131 		return rc;
2132 
2133 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2134 
2135 	if (rc != X86EMUL_CONTINUE)
2136 		return rc;
2137 
2138 	ctxt->_eip = temp_eip;
2139 
2140 	if (ctxt->op_bytes == 4)
2141 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2142 	else if (ctxt->op_bytes == 2) {
2143 		ctxt->eflags &= ~0xffff;
2144 		ctxt->eflags |= temp_eflags;
2145 	}
2146 
2147 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2148 	ctxt->eflags |= X86_EFLAGS_FIXED;
2149 	ctxt->ops->set_nmi_mask(ctxt, false);
2150 
2151 	return rc;
2152 }
2153 
2154 static int em_iret(struct x86_emulate_ctxt *ctxt)
2155 {
2156 	switch(ctxt->mode) {
2157 	case X86EMUL_MODE_REAL:
2158 		return emulate_iret_real(ctxt);
2159 	case X86EMUL_MODE_VM86:
2160 	case X86EMUL_MODE_PROT16:
2161 	case X86EMUL_MODE_PROT32:
2162 	case X86EMUL_MODE_PROT64:
2163 	default:
2164 		/* iret from protected mode unimplemented yet */
2165 		return X86EMUL_UNHANDLEABLE;
2166 	}
2167 }
2168 
2169 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2170 {
2171 	int rc;
2172 	unsigned short sel;
2173 	struct desc_struct new_desc;
2174 	u8 cpl = ctxt->ops->cpl(ctxt);
2175 
2176 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2177 
2178 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2179 				       X86_TRANSFER_CALL_JMP,
2180 				       &new_desc);
2181 	if (rc != X86EMUL_CONTINUE)
2182 		return rc;
2183 
2184 	rc = assign_eip_far(ctxt, ctxt->src.val);
2185 	/* Error handling is not implemented. */
2186 	if (rc != X86EMUL_CONTINUE)
2187 		return X86EMUL_UNHANDLEABLE;
2188 
2189 	return rc;
2190 }
2191 
2192 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2193 {
2194 	return assign_eip_near(ctxt, ctxt->src.val);
2195 }
2196 
2197 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2198 {
2199 	int rc;
2200 	long int old_eip;
2201 
2202 	old_eip = ctxt->_eip;
2203 	rc = assign_eip_near(ctxt, ctxt->src.val);
2204 	if (rc != X86EMUL_CONTINUE)
2205 		return rc;
2206 	ctxt->src.val = old_eip;
2207 	rc = em_push(ctxt);
2208 	return rc;
2209 }
2210 
2211 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2212 {
2213 	u64 old = ctxt->dst.orig_val64;
2214 
2215 	if (ctxt->dst.bytes == 16)
2216 		return X86EMUL_UNHANDLEABLE;
2217 
2218 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2219 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2220 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2221 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2222 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2223 	} else {
2224 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2225 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2226 
2227 		ctxt->eflags |= X86_EFLAGS_ZF;
2228 	}
2229 	return X86EMUL_CONTINUE;
2230 }
2231 
2232 static int em_ret(struct x86_emulate_ctxt *ctxt)
2233 {
2234 	int rc;
2235 	unsigned long eip = 0;
2236 
2237 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2238 	if (rc != X86EMUL_CONTINUE)
2239 		return rc;
2240 
2241 	return assign_eip_near(ctxt, eip);
2242 }
2243 
2244 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2245 {
2246 	int rc;
2247 	unsigned long eip = 0;
2248 	unsigned long cs = 0;
2249 	int cpl = ctxt->ops->cpl(ctxt);
2250 	struct desc_struct new_desc;
2251 
2252 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2253 	if (rc != X86EMUL_CONTINUE)
2254 		return rc;
2255 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2256 	if (rc != X86EMUL_CONTINUE)
2257 		return rc;
2258 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2259 				       X86_TRANSFER_RET,
2260 				       &new_desc);
2261 	if (rc != X86EMUL_CONTINUE)
2262 		return rc;
2263 	rc = assign_eip_far(ctxt, eip);
2264 	/* Error handling is not implemented. */
2265 	if (rc != X86EMUL_CONTINUE)
2266 		return X86EMUL_UNHANDLEABLE;
2267 
2268 	return rc;
2269 }
2270 
2271 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2272 {
2273         int rc;
2274 
2275         rc = em_ret_far(ctxt);
2276         if (rc != X86EMUL_CONTINUE)
2277                 return rc;
2278         rsp_increment(ctxt, ctxt->src.val);
2279         return X86EMUL_CONTINUE;
2280 }
2281 
2282 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2283 {
2284 	/* Save real source value, then compare EAX against destination. */
2285 	ctxt->dst.orig_val = ctxt->dst.val;
2286 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2287 	ctxt->src.orig_val = ctxt->src.val;
2288 	ctxt->src.val = ctxt->dst.orig_val;
2289 	fastop(ctxt, em_cmp);
2290 
2291 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2292 		/* Success: write back to memory; no update of EAX */
2293 		ctxt->src.type = OP_NONE;
2294 		ctxt->dst.val = ctxt->src.orig_val;
2295 	} else {
2296 		/* Failure: write the value we saw to EAX. */
2297 		ctxt->src.type = OP_REG;
2298 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2299 		ctxt->src.val = ctxt->dst.orig_val;
2300 		/* Create write-cycle to dest by writing the same value */
2301 		ctxt->dst.val = ctxt->dst.orig_val;
2302 	}
2303 	return X86EMUL_CONTINUE;
2304 }
2305 
2306 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2307 {
2308 	int seg = ctxt->src2.val;
2309 	unsigned short sel;
2310 	int rc;
2311 
2312 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2313 
2314 	rc = load_segment_descriptor(ctxt, sel, seg);
2315 	if (rc != X86EMUL_CONTINUE)
2316 		return rc;
2317 
2318 	ctxt->dst.val = ctxt->src.val;
2319 	return rc;
2320 }
2321 
2322 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2323 {
2324 	if (!ctxt->ops->is_smm(ctxt))
2325 		return emulate_ud(ctxt);
2326 
2327 	if (ctxt->ops->leave_smm(ctxt))
2328 		ctxt->ops->triple_fault(ctxt);
2329 
2330 	return emulator_recalc_and_set_mode(ctxt);
2331 }
2332 
2333 static void
2334 setup_syscalls_segments(struct desc_struct *cs, struct desc_struct *ss)
2335 {
2336 	cs->l = 0;		/* will be adjusted later */
2337 	set_desc_base(cs, 0);	/* flat segment */
2338 	cs->g = 1;		/* 4kb granularity */
2339 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2340 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2341 	cs->s = 1;
2342 	cs->dpl = 0;		/* will be adjusted later */
2343 	cs->p = 1;
2344 	cs->d = 1;
2345 	cs->avl = 0;
2346 
2347 	set_desc_base(ss, 0);	/* flat segment */
2348 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2349 	ss->g = 1;		/* 4kb granularity */
2350 	ss->s = 1;
2351 	ss->type = 0x03;	/* Read/Write, Accessed */
2352 	ss->d = 1;		/* 32bit stack segment */
2353 	ss->dpl = 0;
2354 	ss->p = 1;
2355 	ss->l = 0;
2356 	ss->avl = 0;
2357 }
2358 
2359 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2360 {
2361 	const struct x86_emulate_ops *ops = ctxt->ops;
2362 	struct desc_struct cs, ss;
2363 	u64 msr_data;
2364 	u16 cs_sel, ss_sel;
2365 	u64 efer = 0;
2366 
2367 	/* syscall is not available in real mode */
2368 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2369 	    ctxt->mode == X86EMUL_MODE_VM86)
2370 		return emulate_ud(ctxt);
2371 
2372 	/*
2373 	 * Intel compatible CPUs only support SYSCALL in 64-bit mode, whereas
2374 	 * AMD allows SYSCALL in any flavor of protected mode.  Note, it's
2375 	 * infeasible to emulate Intel behavior when running on AMD hardware,
2376 	 * as SYSCALL won't fault in the "wrong" mode, i.e. there is no #UD
2377 	 * for KVM to trap-and-emulate, unlike emulating AMD on Intel.
2378 	 */
2379 	if (ctxt->mode != X86EMUL_MODE_PROT64 &&
2380 	    ctxt->ops->guest_cpuid_is_intel_compatible(ctxt))
2381 		return emulate_ud(ctxt);
2382 
2383 	ops->get_msr(ctxt, MSR_EFER, &efer);
2384 	if (!(efer & EFER_SCE))
2385 		return emulate_ud(ctxt);
2386 
2387 	setup_syscalls_segments(&cs, &ss);
2388 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2389 	msr_data >>= 32;
2390 	cs_sel = (u16)(msr_data & 0xfffc);
2391 	ss_sel = (u16)(msr_data + 8);
2392 
2393 	if (efer & EFER_LMA) {
2394 		cs.d = 0;
2395 		cs.l = 1;
2396 	}
2397 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2398 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2399 
2400 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2401 	if (efer & EFER_LMA) {
2402 #ifdef CONFIG_X86_64
2403 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2404 
2405 		ops->get_msr(ctxt,
2406 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2407 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2408 		ctxt->_eip = msr_data;
2409 
2410 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2411 		ctxt->eflags &= ~msr_data;
2412 		ctxt->eflags |= X86_EFLAGS_FIXED;
2413 #endif
2414 	} else {
2415 		/* legacy mode */
2416 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2417 		ctxt->_eip = (u32)msr_data;
2418 
2419 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2420 	}
2421 
2422 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2423 	return X86EMUL_CONTINUE;
2424 }
2425 
2426 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2427 {
2428 	const struct x86_emulate_ops *ops = ctxt->ops;
2429 	struct desc_struct cs, ss;
2430 	u64 msr_data;
2431 	u16 cs_sel, ss_sel;
2432 	u64 efer = 0;
2433 
2434 	ops->get_msr(ctxt, MSR_EFER, &efer);
2435 	/* inject #GP if in real mode */
2436 	if (ctxt->mode == X86EMUL_MODE_REAL)
2437 		return emulate_gp(ctxt, 0);
2438 
2439 	/*
2440 	 * Intel's architecture allows SYSENTER in compatibility mode, but AMD
2441 	 * does not.  Note, AMD does allow SYSENTER in legacy protected mode.
2442 	 */
2443 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA) &&
2444 	    !ctxt->ops->guest_cpuid_is_intel_compatible(ctxt))
2445 		return emulate_ud(ctxt);
2446 
2447 	/* sysenter/sysexit have not been tested in 64bit mode. */
2448 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2449 		return X86EMUL_UNHANDLEABLE;
2450 
2451 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2452 	if ((msr_data & 0xfffc) == 0x0)
2453 		return emulate_gp(ctxt, 0);
2454 
2455 	setup_syscalls_segments(&cs, &ss);
2456 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2457 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2458 	ss_sel = cs_sel + 8;
2459 	if (efer & EFER_LMA) {
2460 		cs.d = 0;
2461 		cs.l = 1;
2462 	}
2463 
2464 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2465 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2466 
2467 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2468 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2469 
2470 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2471 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2472 							      (u32)msr_data;
2473 	if (efer & EFER_LMA)
2474 		ctxt->mode = X86EMUL_MODE_PROT64;
2475 
2476 	return X86EMUL_CONTINUE;
2477 }
2478 
2479 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2480 {
2481 	const struct x86_emulate_ops *ops = ctxt->ops;
2482 	struct desc_struct cs, ss;
2483 	u64 msr_data, rcx, rdx;
2484 	int usermode;
2485 	u16 cs_sel = 0, ss_sel = 0;
2486 
2487 	/* inject #GP if in real mode or Virtual 8086 mode */
2488 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2489 	    ctxt->mode == X86EMUL_MODE_VM86)
2490 		return emulate_gp(ctxt, 0);
2491 
2492 	setup_syscalls_segments(&cs, &ss);
2493 
2494 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2495 		usermode = X86EMUL_MODE_PROT64;
2496 	else
2497 		usermode = X86EMUL_MODE_PROT32;
2498 
2499 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2500 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2501 
2502 	cs.dpl = 3;
2503 	ss.dpl = 3;
2504 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2505 	switch (usermode) {
2506 	case X86EMUL_MODE_PROT32:
2507 		cs_sel = (u16)(msr_data + 16);
2508 		if ((msr_data & 0xfffc) == 0x0)
2509 			return emulate_gp(ctxt, 0);
2510 		ss_sel = (u16)(msr_data + 24);
2511 		rcx = (u32)rcx;
2512 		rdx = (u32)rdx;
2513 		break;
2514 	case X86EMUL_MODE_PROT64:
2515 		cs_sel = (u16)(msr_data + 32);
2516 		if (msr_data == 0x0)
2517 			return emulate_gp(ctxt, 0);
2518 		ss_sel = cs_sel + 8;
2519 		cs.d = 0;
2520 		cs.l = 1;
2521 		if (emul_is_noncanonical_address(rcx, ctxt, 0) ||
2522 		    emul_is_noncanonical_address(rdx, ctxt, 0))
2523 			return emulate_gp(ctxt, 0);
2524 		break;
2525 	}
2526 	cs_sel |= SEGMENT_RPL_MASK;
2527 	ss_sel |= SEGMENT_RPL_MASK;
2528 
2529 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2530 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2531 
2532 	ctxt->_eip = rdx;
2533 	ctxt->mode = usermode;
2534 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2535 
2536 	return X86EMUL_CONTINUE;
2537 }
2538 
2539 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2540 {
2541 	int iopl;
2542 	if (ctxt->mode == X86EMUL_MODE_REAL)
2543 		return false;
2544 	if (ctxt->mode == X86EMUL_MODE_VM86)
2545 		return true;
2546 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2547 	return ctxt->ops->cpl(ctxt) > iopl;
2548 }
2549 
2550 #define VMWARE_PORT_VMPORT	(0x5658)
2551 #define VMWARE_PORT_VMRPC	(0x5659)
2552 
2553 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2554 					    u16 port, u16 len)
2555 {
2556 	const struct x86_emulate_ops *ops = ctxt->ops;
2557 	struct desc_struct tr_seg;
2558 	u32 base3;
2559 	int r;
2560 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2561 	unsigned mask = (1 << len) - 1;
2562 	unsigned long base;
2563 
2564 	/*
2565 	 * VMware allows access to these ports even if denied
2566 	 * by TSS I/O permission bitmap. Mimic behavior.
2567 	 */
2568 	if (enable_vmware_backdoor &&
2569 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2570 		return true;
2571 
2572 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2573 	if (!tr_seg.p)
2574 		return false;
2575 	if (desc_limit_scaled(&tr_seg) < 103)
2576 		return false;
2577 	base = get_desc_base(&tr_seg);
2578 #ifdef CONFIG_X86_64
2579 	base |= ((u64)base3) << 32;
2580 #endif
2581 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2582 	if (r != X86EMUL_CONTINUE)
2583 		return false;
2584 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2585 		return false;
2586 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2587 	if (r != X86EMUL_CONTINUE)
2588 		return false;
2589 	if ((perm >> bit_idx) & mask)
2590 		return false;
2591 	return true;
2592 }
2593 
2594 static bool emulator_io_permitted(struct x86_emulate_ctxt *ctxt,
2595 				  u16 port, u16 len)
2596 {
2597 	if (ctxt->perm_ok)
2598 		return true;
2599 
2600 	if (emulator_bad_iopl(ctxt))
2601 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2602 			return false;
2603 
2604 	ctxt->perm_ok = true;
2605 
2606 	return true;
2607 }
2608 
2609 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2610 {
2611 	/*
2612 	 * Intel CPUs mask the counter and pointers in quite strange
2613 	 * manner when ECX is zero due to REP-string optimizations.
2614 	 */
2615 #ifdef CONFIG_X86_64
2616 	u32 eax, ebx, ecx, edx;
2617 
2618 	if (ctxt->ad_bytes != 4)
2619 		return;
2620 
2621 	eax = ecx = 0;
2622 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2623 	if (!is_guest_vendor_intel(ebx, ecx, edx))
2624 		return;
2625 
2626 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2627 
2628 	switch (ctxt->b) {
2629 	case 0xa4:	/* movsb */
2630 	case 0xa5:	/* movsd/w */
2631 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2632 		fallthrough;
2633 	case 0xaa:	/* stosb */
2634 	case 0xab:	/* stosd/w */
2635 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2636 	}
2637 #endif
2638 }
2639 
2640 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2641 				struct tss_segment_16 *tss)
2642 {
2643 	tss->ip = ctxt->_eip;
2644 	tss->flag = ctxt->eflags;
2645 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2646 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2647 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2648 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2649 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2650 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2651 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2652 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2653 
2654 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2655 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2656 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2657 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2658 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2659 }
2660 
2661 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2662 				 struct tss_segment_16 *tss)
2663 {
2664 	int ret;
2665 	u8 cpl;
2666 
2667 	ctxt->_eip = tss->ip;
2668 	ctxt->eflags = tss->flag | 2;
2669 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2670 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2671 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2672 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2673 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2674 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2675 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2676 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2677 
2678 	/*
2679 	 * SDM says that segment selectors are loaded before segment
2680 	 * descriptors
2681 	 */
2682 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
2683 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2684 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2685 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2686 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2687 
2688 	cpl = tss->cs & 3;
2689 
2690 	/*
2691 	 * Now load segment descriptors. If fault happens at this stage
2692 	 * it is handled in a context of new task
2693 	 */
2694 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
2695 					X86_TRANSFER_TASK_SWITCH, NULL);
2696 	if (ret != X86EMUL_CONTINUE)
2697 		return ret;
2698 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
2699 					X86_TRANSFER_TASK_SWITCH, NULL);
2700 	if (ret != X86EMUL_CONTINUE)
2701 		return ret;
2702 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
2703 					X86_TRANSFER_TASK_SWITCH, NULL);
2704 	if (ret != X86EMUL_CONTINUE)
2705 		return ret;
2706 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
2707 					X86_TRANSFER_TASK_SWITCH, NULL);
2708 	if (ret != X86EMUL_CONTINUE)
2709 		return ret;
2710 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
2711 					X86_TRANSFER_TASK_SWITCH, NULL);
2712 	if (ret != X86EMUL_CONTINUE)
2713 		return ret;
2714 
2715 	return X86EMUL_CONTINUE;
2716 }
2717 
2718 static int task_switch_16(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
2719 			  ulong old_tss_base, struct desc_struct *new_desc)
2720 {
2721 	struct tss_segment_16 tss_seg;
2722 	int ret;
2723 	u32 new_tss_base = get_desc_base(new_desc);
2724 
2725 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2726 	if (ret != X86EMUL_CONTINUE)
2727 		return ret;
2728 
2729 	save_state_to_tss16(ctxt, &tss_seg);
2730 
2731 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2732 	if (ret != X86EMUL_CONTINUE)
2733 		return ret;
2734 
2735 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
2736 	if (ret != X86EMUL_CONTINUE)
2737 		return ret;
2738 
2739 	if (old_tss_sel != 0xffff) {
2740 		tss_seg.prev_task_link = old_tss_sel;
2741 
2742 		ret = linear_write_system(ctxt, new_tss_base,
2743 					  &tss_seg.prev_task_link,
2744 					  sizeof(tss_seg.prev_task_link));
2745 		if (ret != X86EMUL_CONTINUE)
2746 			return ret;
2747 	}
2748 
2749 	return load_state_from_tss16(ctxt, &tss_seg);
2750 }
2751 
2752 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
2753 				struct tss_segment_32 *tss)
2754 {
2755 	/* CR3 and ldt selector are not saved intentionally */
2756 	tss->eip = ctxt->_eip;
2757 	tss->eflags = ctxt->eflags;
2758 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
2759 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
2760 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
2761 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
2762 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
2763 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
2764 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
2765 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
2766 
2767 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2768 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2769 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2770 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2771 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
2772 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
2773 }
2774 
2775 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
2776 				 struct tss_segment_32 *tss)
2777 {
2778 	int ret;
2779 	u8 cpl;
2780 
2781 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
2782 		return emulate_gp(ctxt, 0);
2783 	ctxt->_eip = tss->eip;
2784 	ctxt->eflags = tss->eflags | 2;
2785 
2786 	/* General purpose registers */
2787 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
2788 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
2789 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
2790 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
2791 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
2792 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
2793 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
2794 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
2795 
2796 	/*
2797 	 * SDM says that segment selectors are loaded before segment
2798 	 * descriptors.  This is important because CPL checks will
2799 	 * use CS.RPL.
2800 	 */
2801 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
2802 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2803 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2804 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2805 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2806 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
2807 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
2808 
2809 	/*
2810 	 * If we're switching between Protected Mode and VM86, we need to make
2811 	 * sure to update the mode before loading the segment descriptors so
2812 	 * that the selectors are interpreted correctly.
2813 	 */
2814 	if (ctxt->eflags & X86_EFLAGS_VM) {
2815 		ctxt->mode = X86EMUL_MODE_VM86;
2816 		cpl = 3;
2817 	} else {
2818 		ctxt->mode = X86EMUL_MODE_PROT32;
2819 		cpl = tss->cs & 3;
2820 	}
2821 
2822 	/*
2823 	 * Now load segment descriptors. If fault happens at this stage
2824 	 * it is handled in a context of new task
2825 	 */
2826 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
2827 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
2828 	if (ret != X86EMUL_CONTINUE)
2829 		return ret;
2830 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
2831 					X86_TRANSFER_TASK_SWITCH, NULL);
2832 	if (ret != X86EMUL_CONTINUE)
2833 		return ret;
2834 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
2835 					X86_TRANSFER_TASK_SWITCH, NULL);
2836 	if (ret != X86EMUL_CONTINUE)
2837 		return ret;
2838 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
2839 					X86_TRANSFER_TASK_SWITCH, NULL);
2840 	if (ret != X86EMUL_CONTINUE)
2841 		return ret;
2842 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
2843 					X86_TRANSFER_TASK_SWITCH, NULL);
2844 	if (ret != X86EMUL_CONTINUE)
2845 		return ret;
2846 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
2847 					X86_TRANSFER_TASK_SWITCH, NULL);
2848 	if (ret != X86EMUL_CONTINUE)
2849 		return ret;
2850 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
2851 					X86_TRANSFER_TASK_SWITCH, NULL);
2852 
2853 	return ret;
2854 }
2855 
2856 static int task_switch_32(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
2857 			  ulong old_tss_base, struct desc_struct *new_desc)
2858 {
2859 	struct tss_segment_32 tss_seg;
2860 	int ret;
2861 	u32 new_tss_base = get_desc_base(new_desc);
2862 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
2863 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
2864 
2865 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2866 	if (ret != X86EMUL_CONTINUE)
2867 		return ret;
2868 
2869 	save_state_to_tss32(ctxt, &tss_seg);
2870 
2871 	/* Only GP registers and segment selectors are saved */
2872 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
2873 				  ldt_sel_offset - eip_offset);
2874 	if (ret != X86EMUL_CONTINUE)
2875 		return ret;
2876 
2877 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
2878 	if (ret != X86EMUL_CONTINUE)
2879 		return ret;
2880 
2881 	if (old_tss_sel != 0xffff) {
2882 		tss_seg.prev_task_link = old_tss_sel;
2883 
2884 		ret = linear_write_system(ctxt, new_tss_base,
2885 					  &tss_seg.prev_task_link,
2886 					  sizeof(tss_seg.prev_task_link));
2887 		if (ret != X86EMUL_CONTINUE)
2888 			return ret;
2889 	}
2890 
2891 	return load_state_from_tss32(ctxt, &tss_seg);
2892 }
2893 
2894 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
2895 				   u16 tss_selector, int idt_index, int reason,
2896 				   bool has_error_code, u32 error_code)
2897 {
2898 	const struct x86_emulate_ops *ops = ctxt->ops;
2899 	struct desc_struct curr_tss_desc, next_tss_desc;
2900 	int ret;
2901 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
2902 	ulong old_tss_base =
2903 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
2904 	u32 desc_limit;
2905 	ulong desc_addr, dr7;
2906 
2907 	/* FIXME: old_tss_base == ~0 ? */
2908 
2909 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
2910 	if (ret != X86EMUL_CONTINUE)
2911 		return ret;
2912 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
2913 	if (ret != X86EMUL_CONTINUE)
2914 		return ret;
2915 
2916 	/* FIXME: check that next_tss_desc is tss */
2917 
2918 	/*
2919 	 * Check privileges. The three cases are task switch caused by...
2920 	 *
2921 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
2922 	 * 2. Exception/IRQ/iret: No check is performed
2923 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
2924 	 *    hardware checks it before exiting.
2925 	 */
2926 	if (reason == TASK_SWITCH_GATE) {
2927 		if (idt_index != -1) {
2928 			/* Software interrupts */
2929 			struct desc_struct task_gate_desc;
2930 			int dpl;
2931 
2932 			ret = read_interrupt_descriptor(ctxt, idt_index,
2933 							&task_gate_desc);
2934 			if (ret != X86EMUL_CONTINUE)
2935 				return ret;
2936 
2937 			dpl = task_gate_desc.dpl;
2938 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
2939 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
2940 		}
2941 	}
2942 
2943 	desc_limit = desc_limit_scaled(&next_tss_desc);
2944 	if (!next_tss_desc.p ||
2945 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
2946 	     desc_limit < 0x2b)) {
2947 		return emulate_ts(ctxt, tss_selector & 0xfffc);
2948 	}
2949 
2950 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
2951 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
2952 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
2953 	}
2954 
2955 	if (reason == TASK_SWITCH_IRET)
2956 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
2957 
2958 	/* set back link to prev task only if NT bit is set in eflags
2959 	   note that old_tss_sel is not used after this point */
2960 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
2961 		old_tss_sel = 0xffff;
2962 
2963 	if (next_tss_desc.type & 8)
2964 		ret = task_switch_32(ctxt, old_tss_sel, old_tss_base, &next_tss_desc);
2965 	else
2966 		ret = task_switch_16(ctxt, old_tss_sel,
2967 				     old_tss_base, &next_tss_desc);
2968 	if (ret != X86EMUL_CONTINUE)
2969 		return ret;
2970 
2971 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
2972 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
2973 
2974 	if (reason != TASK_SWITCH_IRET) {
2975 		next_tss_desc.type |= (1 << 1); /* set busy flag */
2976 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
2977 	}
2978 
2979 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
2980 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
2981 
2982 	if (has_error_code) {
2983 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
2984 		ctxt->lock_prefix = 0;
2985 		ctxt->src.val = (unsigned long) error_code;
2986 		ret = em_push(ctxt);
2987 	}
2988 
2989 	dr7 = ops->get_dr(ctxt, 7);
2990 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
2991 
2992 	return ret;
2993 }
2994 
2995 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
2996 			 u16 tss_selector, int idt_index, int reason,
2997 			 bool has_error_code, u32 error_code)
2998 {
2999 	int rc;
3000 
3001 	invalidate_registers(ctxt);
3002 	ctxt->_eip = ctxt->eip;
3003 	ctxt->dst.type = OP_NONE;
3004 
3005 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3006 				     has_error_code, error_code);
3007 
3008 	if (rc == X86EMUL_CONTINUE) {
3009 		ctxt->eip = ctxt->_eip;
3010 		writeback_registers(ctxt);
3011 	}
3012 
3013 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3014 }
3015 
3016 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3017 		struct operand *op)
3018 {
3019 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3020 
3021 	register_address_increment(ctxt, reg, df * op->bytes);
3022 	op->addr.mem.ea = register_address(ctxt, reg);
3023 }
3024 
3025 static int em_das(struct x86_emulate_ctxt *ctxt)
3026 {
3027 	u8 al, old_al;
3028 	bool af, cf, old_cf;
3029 
3030 	cf = ctxt->eflags & X86_EFLAGS_CF;
3031 	al = ctxt->dst.val;
3032 
3033 	old_al = al;
3034 	old_cf = cf;
3035 	cf = false;
3036 	af = ctxt->eflags & X86_EFLAGS_AF;
3037 	if ((al & 0x0f) > 9 || af) {
3038 		al -= 6;
3039 		cf = old_cf | (al >= 250);
3040 		af = true;
3041 	} else {
3042 		af = false;
3043 	}
3044 	if (old_al > 0x99 || old_cf) {
3045 		al -= 0x60;
3046 		cf = true;
3047 	}
3048 
3049 	ctxt->dst.val = al;
3050 	/* Set PF, ZF, SF */
3051 	ctxt->src.type = OP_IMM;
3052 	ctxt->src.val = 0;
3053 	ctxt->src.bytes = 1;
3054 	fastop(ctxt, em_or);
3055 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3056 	if (cf)
3057 		ctxt->eflags |= X86_EFLAGS_CF;
3058 	if (af)
3059 		ctxt->eflags |= X86_EFLAGS_AF;
3060 	return X86EMUL_CONTINUE;
3061 }
3062 
3063 static int em_aam(struct x86_emulate_ctxt *ctxt)
3064 {
3065 	u8 al, ah;
3066 
3067 	if (ctxt->src.val == 0)
3068 		return emulate_de(ctxt);
3069 
3070 	al = ctxt->dst.val & 0xff;
3071 	ah = al / ctxt->src.val;
3072 	al %= ctxt->src.val;
3073 
3074 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3075 
3076 	/* Set PF, ZF, SF */
3077 	ctxt->src.type = OP_IMM;
3078 	ctxt->src.val = 0;
3079 	ctxt->src.bytes = 1;
3080 	fastop(ctxt, em_or);
3081 
3082 	return X86EMUL_CONTINUE;
3083 }
3084 
3085 static int em_aad(struct x86_emulate_ctxt *ctxt)
3086 {
3087 	u8 al = ctxt->dst.val & 0xff;
3088 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3089 
3090 	al = (al + (ah * ctxt->src.val)) & 0xff;
3091 
3092 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3093 
3094 	/* Set PF, ZF, SF */
3095 	ctxt->src.type = OP_IMM;
3096 	ctxt->src.val = 0;
3097 	ctxt->src.bytes = 1;
3098 	fastop(ctxt, em_or);
3099 
3100 	return X86EMUL_CONTINUE;
3101 }
3102 
3103 static int em_call(struct x86_emulate_ctxt *ctxt)
3104 {
3105 	int rc;
3106 	long rel = ctxt->src.val;
3107 
3108 	ctxt->src.val = (unsigned long)ctxt->_eip;
3109 	rc = jmp_rel(ctxt, rel);
3110 	if (rc != X86EMUL_CONTINUE)
3111 		return rc;
3112 	return em_push(ctxt);
3113 }
3114 
3115 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3116 {
3117 	u16 sel, old_cs;
3118 	ulong old_eip;
3119 	int rc;
3120 	struct desc_struct old_desc, new_desc;
3121 	const struct x86_emulate_ops *ops = ctxt->ops;
3122 	int cpl = ctxt->ops->cpl(ctxt);
3123 	enum x86emul_mode prev_mode = ctxt->mode;
3124 
3125 	old_eip = ctxt->_eip;
3126 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3127 
3128 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3129 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3130 				       X86_TRANSFER_CALL_JMP, &new_desc);
3131 	if (rc != X86EMUL_CONTINUE)
3132 		return rc;
3133 
3134 	rc = assign_eip_far(ctxt, ctxt->src.val);
3135 	if (rc != X86EMUL_CONTINUE)
3136 		goto fail;
3137 
3138 	ctxt->src.val = old_cs;
3139 	rc = em_push(ctxt);
3140 	if (rc != X86EMUL_CONTINUE)
3141 		goto fail;
3142 
3143 	ctxt->src.val = old_eip;
3144 	rc = em_push(ctxt);
3145 	/* If we failed, we tainted the memory, but the very least we should
3146 	   restore cs */
3147 	if (rc != X86EMUL_CONTINUE) {
3148 		pr_warn_once("faulting far call emulation tainted memory\n");
3149 		goto fail;
3150 	}
3151 	return rc;
3152 fail:
3153 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3154 	ctxt->mode = prev_mode;
3155 	return rc;
3156 
3157 }
3158 
3159 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3160 {
3161 	int rc;
3162 	unsigned long eip = 0;
3163 
3164 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3165 	if (rc != X86EMUL_CONTINUE)
3166 		return rc;
3167 	rc = assign_eip_near(ctxt, eip);
3168 	if (rc != X86EMUL_CONTINUE)
3169 		return rc;
3170 	rsp_increment(ctxt, ctxt->src.val);
3171 	return X86EMUL_CONTINUE;
3172 }
3173 
3174 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3175 {
3176 	/* Write back the register source. */
3177 	ctxt->src.val = ctxt->dst.val;
3178 	write_register_operand(&ctxt->src);
3179 
3180 	/* Write back the memory destination with implicit LOCK prefix. */
3181 	ctxt->dst.val = ctxt->src.orig_val;
3182 	ctxt->lock_prefix = 1;
3183 	return X86EMUL_CONTINUE;
3184 }
3185 
3186 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3187 {
3188 	ctxt->dst.val = ctxt->src2.val;
3189 	return fastop(ctxt, em_imul);
3190 }
3191 
3192 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3193 {
3194 	ctxt->dst.type = OP_REG;
3195 	ctxt->dst.bytes = ctxt->src.bytes;
3196 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3197 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3198 
3199 	return X86EMUL_CONTINUE;
3200 }
3201 
3202 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3203 {
3204 	u64 tsc_aux = 0;
3205 
3206 	if (!ctxt->ops->guest_has_rdpid(ctxt))
3207 		return emulate_ud(ctxt);
3208 
3209 	ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux);
3210 	ctxt->dst.val = tsc_aux;
3211 	return X86EMUL_CONTINUE;
3212 }
3213 
3214 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3215 {
3216 	u64 tsc = 0;
3217 
3218 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3219 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3220 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3221 	return X86EMUL_CONTINUE;
3222 }
3223 
3224 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3225 {
3226 	u64 pmc;
3227 
3228 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3229 		return emulate_gp(ctxt, 0);
3230 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3231 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3232 	return X86EMUL_CONTINUE;
3233 }
3234 
3235 static int em_mov(struct x86_emulate_ctxt *ctxt)
3236 {
3237 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3238 	return X86EMUL_CONTINUE;
3239 }
3240 
3241 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3242 {
3243 	u16 tmp;
3244 
3245 	if (!ctxt->ops->guest_has_movbe(ctxt))
3246 		return emulate_ud(ctxt);
3247 
3248 	switch (ctxt->op_bytes) {
3249 	case 2:
3250 		/*
3251 		 * From MOVBE definition: "...When the operand size is 16 bits,
3252 		 * the upper word of the destination register remains unchanged
3253 		 * ..."
3254 		 *
3255 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3256 		 * rules so we have to do the operation almost per hand.
3257 		 */
3258 		tmp = (u16)ctxt->src.val;
3259 		ctxt->dst.val &= ~0xffffUL;
3260 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3261 		break;
3262 	case 4:
3263 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3264 		break;
3265 	case 8:
3266 		ctxt->dst.val = swab64(ctxt->src.val);
3267 		break;
3268 	default:
3269 		BUG();
3270 	}
3271 	return X86EMUL_CONTINUE;
3272 }
3273 
3274 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3275 {
3276 	int cr_num = ctxt->modrm_reg;
3277 	int r;
3278 
3279 	if (ctxt->ops->set_cr(ctxt, cr_num, ctxt->src.val))
3280 		return emulate_gp(ctxt, 0);
3281 
3282 	/* Disable writeback. */
3283 	ctxt->dst.type = OP_NONE;
3284 
3285 	if (cr_num == 0) {
3286 		/*
3287 		 * CR0 write might have updated CR0.PE and/or CR0.PG
3288 		 * which can affect the cpu's execution mode.
3289 		 */
3290 		r = emulator_recalc_and_set_mode(ctxt);
3291 		if (r != X86EMUL_CONTINUE)
3292 			return r;
3293 	}
3294 
3295 	return X86EMUL_CONTINUE;
3296 }
3297 
3298 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3299 {
3300 	unsigned long val;
3301 
3302 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3303 		val = ctxt->src.val & ~0ULL;
3304 	else
3305 		val = ctxt->src.val & ~0U;
3306 
3307 	/* #UD condition is already handled. */
3308 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3309 		return emulate_gp(ctxt, 0);
3310 
3311 	/* Disable writeback. */
3312 	ctxt->dst.type = OP_NONE;
3313 	return X86EMUL_CONTINUE;
3314 }
3315 
3316 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3317 {
3318 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3319 	u64 msr_data;
3320 	int r;
3321 
3322 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3323 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3324 	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);
3325 
3326 	if (r == X86EMUL_PROPAGATE_FAULT)
3327 		return emulate_gp(ctxt, 0);
3328 
3329 	return r;
3330 }
3331 
3332 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3333 {
3334 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3335 	u64 msr_data;
3336 	int r;
3337 
3338 	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);
3339 
3340 	if (r == X86EMUL_PROPAGATE_FAULT)
3341 		return emulate_gp(ctxt, 0);
3342 
3343 	if (r == X86EMUL_CONTINUE) {
3344 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3345 		*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3346 	}
3347 	return r;
3348 }
3349 
3350 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3351 {
3352 	if (segment > VCPU_SREG_GS &&
3353 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3354 	    ctxt->ops->cpl(ctxt) > 0)
3355 		return emulate_gp(ctxt, 0);
3356 
3357 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3358 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3359 		ctxt->dst.bytes = 2;
3360 	return X86EMUL_CONTINUE;
3361 }
3362 
3363 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3364 {
3365 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3366 		return emulate_ud(ctxt);
3367 
3368 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3369 }
3370 
3371 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3372 {
3373 	u16 sel = ctxt->src.val;
3374 
3375 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3376 		return emulate_ud(ctxt);
3377 
3378 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3379 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3380 
3381 	/* Disable writeback. */
3382 	ctxt->dst.type = OP_NONE;
3383 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3384 }
3385 
3386 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3387 {
3388 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3389 }
3390 
3391 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3392 {
3393 	u16 sel = ctxt->src.val;
3394 
3395 	/* Disable writeback. */
3396 	ctxt->dst.type = OP_NONE;
3397 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3398 }
3399 
3400 static int em_str(struct x86_emulate_ctxt *ctxt)
3401 {
3402 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3403 }
3404 
3405 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3406 {
3407 	u16 sel = ctxt->src.val;
3408 
3409 	/* Disable writeback. */
3410 	ctxt->dst.type = OP_NONE;
3411 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3412 }
3413 
3414 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3415 {
3416 	int rc;
3417 	ulong linear;
3418 	unsigned int max_size;
3419 
3420 	rc = __linearize(ctxt, ctxt->src.addr.mem, &max_size, 1, ctxt->mode,
3421 			 &linear, X86EMUL_F_INVLPG);
3422 	if (rc == X86EMUL_CONTINUE)
3423 		ctxt->ops->invlpg(ctxt, linear);
3424 	/* Disable writeback. */
3425 	ctxt->dst.type = OP_NONE;
3426 	return X86EMUL_CONTINUE;
3427 }
3428 
3429 static int em_clts(struct x86_emulate_ctxt *ctxt)
3430 {
3431 	ulong cr0;
3432 
3433 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3434 	cr0 &= ~X86_CR0_TS;
3435 	ctxt->ops->set_cr(ctxt, 0, cr0);
3436 	return X86EMUL_CONTINUE;
3437 }
3438 
3439 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3440 {
3441 	int rc = ctxt->ops->fix_hypercall(ctxt);
3442 
3443 	if (rc != X86EMUL_CONTINUE)
3444 		return rc;
3445 
3446 	/* Let the processor re-execute the fixed hypercall */
3447 	ctxt->_eip = ctxt->eip;
3448 	/* Disable writeback. */
3449 	ctxt->dst.type = OP_NONE;
3450 	return X86EMUL_CONTINUE;
3451 }
3452 
3453 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3454 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3455 					      struct desc_ptr *ptr))
3456 {
3457 	struct desc_ptr desc_ptr;
3458 
3459 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3460 	    ctxt->ops->cpl(ctxt) > 0)
3461 		return emulate_gp(ctxt, 0);
3462 
3463 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3464 		ctxt->op_bytes = 8;
3465 	get(ctxt, &desc_ptr);
3466 	if (ctxt->op_bytes == 2) {
3467 		ctxt->op_bytes = 4;
3468 		desc_ptr.address &= 0x00ffffff;
3469 	}
3470 	/* Disable writeback. */
3471 	ctxt->dst.type = OP_NONE;
3472 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3473 				   &desc_ptr, 2 + ctxt->op_bytes);
3474 }
3475 
3476 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3477 {
3478 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3479 }
3480 
3481 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3482 {
3483 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3484 }
3485 
3486 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3487 {
3488 	struct desc_ptr desc_ptr;
3489 	int rc;
3490 
3491 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3492 		ctxt->op_bytes = 8;
3493 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3494 			     &desc_ptr.size, &desc_ptr.address,
3495 			     ctxt->op_bytes);
3496 	if (rc != X86EMUL_CONTINUE)
3497 		return rc;
3498 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3499 	    emul_is_noncanonical_address(desc_ptr.address, ctxt,
3500 					 X86EMUL_F_DT_LOAD))
3501 		return emulate_gp(ctxt, 0);
3502 	if (lgdt)
3503 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3504 	else
3505 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3506 	/* Disable writeback. */
3507 	ctxt->dst.type = OP_NONE;
3508 	return X86EMUL_CONTINUE;
3509 }
3510 
3511 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3512 {
3513 	return em_lgdt_lidt(ctxt, true);
3514 }
3515 
3516 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3517 {
3518 	return em_lgdt_lidt(ctxt, false);
3519 }
3520 
3521 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3522 {
3523 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3524 	    ctxt->ops->cpl(ctxt) > 0)
3525 		return emulate_gp(ctxt, 0);
3526 
3527 	if (ctxt->dst.type == OP_MEM)
3528 		ctxt->dst.bytes = 2;
3529 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3530 	return X86EMUL_CONTINUE;
3531 }
3532 
3533 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3534 {
3535 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3536 			  | (ctxt->src.val & 0x0f));
3537 	ctxt->dst.type = OP_NONE;
3538 	return X86EMUL_CONTINUE;
3539 }
3540 
3541 static int em_loop(struct x86_emulate_ctxt *ctxt)
3542 {
3543 	int rc = X86EMUL_CONTINUE;
3544 
3545 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3546 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3547 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3548 		rc = jmp_rel(ctxt, ctxt->src.val);
3549 
3550 	return rc;
3551 }
3552 
3553 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3554 {
3555 	int rc = X86EMUL_CONTINUE;
3556 
3557 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3558 		rc = jmp_rel(ctxt, ctxt->src.val);
3559 
3560 	return rc;
3561 }
3562 
3563 static int em_in(struct x86_emulate_ctxt *ctxt)
3564 {
3565 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3566 			     &ctxt->dst.val))
3567 		return X86EMUL_IO_NEEDED;
3568 
3569 	return X86EMUL_CONTINUE;
3570 }
3571 
3572 static int em_out(struct x86_emulate_ctxt *ctxt)
3573 {
3574 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3575 				    &ctxt->src.val, 1);
3576 	/* Disable writeback. */
3577 	ctxt->dst.type = OP_NONE;
3578 	return X86EMUL_CONTINUE;
3579 }
3580 
3581 static int em_cli(struct x86_emulate_ctxt *ctxt)
3582 {
3583 	if (emulator_bad_iopl(ctxt))
3584 		return emulate_gp(ctxt, 0);
3585 
3586 	ctxt->eflags &= ~X86_EFLAGS_IF;
3587 	return X86EMUL_CONTINUE;
3588 }
3589 
3590 static int em_sti(struct x86_emulate_ctxt *ctxt)
3591 {
3592 	if (emulator_bad_iopl(ctxt))
3593 		return emulate_gp(ctxt, 0);
3594 
3595 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3596 	ctxt->eflags |= X86_EFLAGS_IF;
3597 	return X86EMUL_CONTINUE;
3598 }
3599 
3600 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3601 {
3602 	u32 eax, ebx, ecx, edx;
3603 	u64 msr = 0;
3604 
3605 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3606 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3607 	    ctxt->ops->cpl(ctxt)) {
3608 		return emulate_gp(ctxt, 0);
3609 	}
3610 
3611 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3612 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3613 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3614 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3615 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3616 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3617 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3618 	return X86EMUL_CONTINUE;
3619 }
3620 
3621 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3622 {
3623 	u32 flags;
3624 
3625 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3626 		X86_EFLAGS_SF;
3627 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3628 
3629 	ctxt->eflags &= ~0xffUL;
3630 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3631 	return X86EMUL_CONTINUE;
3632 }
3633 
3634 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3635 {
3636 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3637 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3638 	return X86EMUL_CONTINUE;
3639 }
3640 
3641 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3642 {
3643 	switch (ctxt->op_bytes) {
3644 #ifdef CONFIG_X86_64
3645 	case 8:
3646 		asm("bswap %0" : "+r"(ctxt->dst.val));
3647 		break;
3648 #endif
3649 	default:
3650 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3651 		break;
3652 	}
3653 	return X86EMUL_CONTINUE;
3654 }
3655 
3656 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3657 {
3658 	/* emulating clflush regardless of cpuid */
3659 	return X86EMUL_CONTINUE;
3660 }
3661 
3662 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
3663 {
3664 	/* emulating clflushopt regardless of cpuid */
3665 	return X86EMUL_CONTINUE;
3666 }
3667 
3668 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3669 {
3670 	ctxt->dst.val = (s32) ctxt->src.val;
3671 	return X86EMUL_CONTINUE;
3672 }
3673 
3674 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3675 {
3676 	if (!ctxt->ops->guest_has_fxsr(ctxt))
3677 		return emulate_ud(ctxt);
3678 
3679 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3680 		return emulate_nm(ctxt);
3681 
3682 	/*
3683 	 * Don't emulate a case that should never be hit, instead of working
3684 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3685 	 */
3686 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3687 		return X86EMUL_UNHANDLEABLE;
3688 
3689 	return X86EMUL_CONTINUE;
3690 }
3691 
3692 /*
3693  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
3694  * and restore MXCSR.
3695  */
3696 static size_t __fxstate_size(int nregs)
3697 {
3698 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
3699 }
3700 
3701 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
3702 {
3703 	bool cr4_osfxsr;
3704 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3705 		return __fxstate_size(16);
3706 
3707 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
3708 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
3709 }
3710 
3711 /*
3712  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
3713  *  1) 16 bit mode
3714  *  2) 32 bit mode
3715  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
3716  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
3717  *       save and restore
3718  *  3) 64-bit mode with REX.W prefix
3719  *     - like (2), but XMM 8-15 are being saved and restored
3720  *  4) 64-bit mode without REX.W prefix
3721  *     - like (3), but FIP and FDP are 64 bit
3722  *
3723  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
3724  * desired result.  (4) is not emulated.
3725  *
3726  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
3727  * and FPU DS) should match.
3728  */
3729 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
3730 {
3731 	struct fxregs_state fx_state;
3732 	int rc;
3733 
3734 	rc = check_fxsr(ctxt);
3735 	if (rc != X86EMUL_CONTINUE)
3736 		return rc;
3737 
3738 	kvm_fpu_get();
3739 
3740 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
3741 
3742 	kvm_fpu_put();
3743 
3744 	if (rc != X86EMUL_CONTINUE)
3745 		return rc;
3746 
3747 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
3748 		                   fxstate_size(ctxt));
3749 }
3750 
3751 /*
3752  * FXRSTOR might restore XMM registers not provided by the guest. Fill
3753  * in the host registers (via FXSAVE) instead, so they won't be modified.
3754  * (preemption has to stay disabled until FXRSTOR).
3755  *
3756  * Use noinline to keep the stack for other functions called by callers small.
3757  */
3758 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
3759 				 const size_t used_size)
3760 {
3761 	struct fxregs_state fx_tmp;
3762 	int rc;
3763 
3764 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
3765 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
3766 	       __fxstate_size(16) - used_size);
3767 
3768 	return rc;
3769 }
3770 
3771 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
3772 {
3773 	struct fxregs_state fx_state;
3774 	int rc;
3775 	size_t size;
3776 
3777 	rc = check_fxsr(ctxt);
3778 	if (rc != X86EMUL_CONTINUE)
3779 		return rc;
3780 
3781 	size = fxstate_size(ctxt);
3782 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
3783 	if (rc != X86EMUL_CONTINUE)
3784 		return rc;
3785 
3786 	kvm_fpu_get();
3787 
3788 	if (size < __fxstate_size(16)) {
3789 		rc = fxregs_fixup(&fx_state, size);
3790 		if (rc != X86EMUL_CONTINUE)
3791 			goto out;
3792 	}
3793 
3794 	if (fx_state.mxcsr >> 16) {
3795 		rc = emulate_gp(ctxt, 0);
3796 		goto out;
3797 	}
3798 
3799 	if (rc == X86EMUL_CONTINUE)
3800 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
3801 
3802 out:
3803 	kvm_fpu_put();
3804 
3805 	return rc;
3806 }
3807 
3808 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
3809 {
3810 	u32 eax, ecx, edx;
3811 
3812 	if (!(ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSXSAVE))
3813 		return emulate_ud(ctxt);
3814 
3815 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3816 	edx = reg_read(ctxt, VCPU_REGS_RDX);
3817 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3818 
3819 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
3820 		return emulate_gp(ctxt, 0);
3821 
3822 	return X86EMUL_CONTINUE;
3823 }
3824 
3825 static bool valid_cr(int nr)
3826 {
3827 	switch (nr) {
3828 	case 0:
3829 	case 2 ... 4:
3830 	case 8:
3831 		return true;
3832 	default:
3833 		return false;
3834 	}
3835 }
3836 
3837 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
3838 {
3839 	if (!valid_cr(ctxt->modrm_reg))
3840 		return emulate_ud(ctxt);
3841 
3842 	return X86EMUL_CONTINUE;
3843 }
3844 
3845 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
3846 {
3847 	int dr = ctxt->modrm_reg;
3848 	u64 cr4;
3849 
3850 	if (dr > 7)
3851 		return emulate_ud(ctxt);
3852 
3853 	cr4 = ctxt->ops->get_cr(ctxt, 4);
3854 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
3855 		return emulate_ud(ctxt);
3856 
3857 	if (ctxt->ops->get_dr(ctxt, 7) & DR7_GD) {
3858 		ulong dr6;
3859 
3860 		dr6 = ctxt->ops->get_dr(ctxt, 6);
3861 		dr6 &= ~DR_TRAP_BITS;
3862 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
3863 		ctxt->ops->set_dr(ctxt, 6, dr6);
3864 		return emulate_db(ctxt);
3865 	}
3866 
3867 	return X86EMUL_CONTINUE;
3868 }
3869 
3870 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
3871 {
3872 	u64 new_val = ctxt->src.val64;
3873 	int dr = ctxt->modrm_reg;
3874 
3875 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
3876 		return emulate_gp(ctxt, 0);
3877 
3878 	return check_dr_read(ctxt);
3879 }
3880 
3881 static int check_svme(struct x86_emulate_ctxt *ctxt)
3882 {
3883 	u64 efer = 0;
3884 
3885 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
3886 
3887 	if (!(efer & EFER_SVME))
3888 		return emulate_ud(ctxt);
3889 
3890 	return X86EMUL_CONTINUE;
3891 }
3892 
3893 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
3894 {
3895 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
3896 
3897 	/* Valid physical address? */
3898 	if (rax & 0xffff000000000000ULL)
3899 		return emulate_gp(ctxt, 0);
3900 
3901 	return check_svme(ctxt);
3902 }
3903 
3904 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
3905 {
3906 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
3907 
3908 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
3909 		return emulate_gp(ctxt, 0);
3910 
3911 	return X86EMUL_CONTINUE;
3912 }
3913 
3914 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
3915 {
3916 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
3917 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
3918 
3919 	/*
3920 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
3921 	 * in Ring3 when CR4.PCE=0.
3922 	 */
3923 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
3924 		return X86EMUL_CONTINUE;
3925 
3926 	/*
3927 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
3928 	 * check however is unnecessary because CPL is always 0 outside
3929 	 * protected mode.
3930 	 */
3931 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
3932 	    ctxt->ops->check_rdpmc_early(ctxt, rcx))
3933 		return emulate_gp(ctxt, 0);
3934 
3935 	return X86EMUL_CONTINUE;
3936 }
3937 
3938 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
3939 {
3940 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
3941 	if (!emulator_io_permitted(ctxt, ctxt->src.val, ctxt->dst.bytes))
3942 		return emulate_gp(ctxt, 0);
3943 
3944 	return X86EMUL_CONTINUE;
3945 }
3946 
3947 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
3948 {
3949 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
3950 	if (!emulator_io_permitted(ctxt, ctxt->dst.val, ctxt->src.bytes))
3951 		return emulate_gp(ctxt, 0);
3952 
3953 	return X86EMUL_CONTINUE;
3954 }
3955 
3956 #define D(_y) { .flags = (_y) }
3957 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
3958 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
3959 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
3960 #define N    D(NotImpl)
3961 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
3962 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
3963 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
3964 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
3965 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
3966 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
3967 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
3968 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
3969 #define II(_f, _e, _i) \
3970 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
3971 #define IIP(_f, _e, _i, _p) \
3972 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
3973 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
3974 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
3975 
3976 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
3977 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
3978 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
3979 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
3980 #define I2bvIP(_f, _e, _i, _p) \
3981 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
3982 
3983 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
3984 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
3985 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
3986 
3987 static const struct opcode group7_rm0[] = {
3988 	N,
3989 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
3990 	N, N, N, N, N, N,
3991 };
3992 
3993 static const struct opcode group7_rm1[] = {
3994 	DI(SrcNone | Priv, monitor),
3995 	DI(SrcNone | Priv, mwait),
3996 	N, N, N, N, N, N,
3997 };
3998 
3999 static const struct opcode group7_rm2[] = {
4000 	N,
4001 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4002 	N, N, N, N, N, N,
4003 };
4004 
4005 static const struct opcode group7_rm3[] = {
4006 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4007 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4008 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4009 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4010 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4011 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4012 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4013 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4014 };
4015 
4016 static const struct opcode group7_rm7[] = {
4017 	N,
4018 	DIP(SrcNone, rdtscp, check_rdtsc),
4019 	N, N, N, N, N, N,
4020 };
4021 
4022 static const struct opcode group1[] = {
4023 	F(Lock, em_add),
4024 	F(Lock | PageTable, em_or),
4025 	F(Lock, em_adc),
4026 	F(Lock, em_sbb),
4027 	F(Lock | PageTable, em_and),
4028 	F(Lock, em_sub),
4029 	F(Lock, em_xor),
4030 	F(NoWrite, em_cmp),
4031 };
4032 
4033 static const struct opcode group1A[] = {
4034 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4035 };
4036 
4037 static const struct opcode group2[] = {
4038 	F(DstMem | ModRM, em_rol),
4039 	F(DstMem | ModRM, em_ror),
4040 	F(DstMem | ModRM, em_rcl),
4041 	F(DstMem | ModRM, em_rcr),
4042 	F(DstMem | ModRM, em_shl),
4043 	F(DstMem | ModRM, em_shr),
4044 	F(DstMem | ModRM, em_shl),
4045 	F(DstMem | ModRM, em_sar),
4046 };
4047 
4048 static const struct opcode group3[] = {
4049 	F(DstMem | SrcImm | NoWrite, em_test),
4050 	F(DstMem | SrcImm | NoWrite, em_test),
4051 	F(DstMem | SrcNone | Lock, em_not),
4052 	F(DstMem | SrcNone | Lock, em_neg),
4053 	F(DstXacc | Src2Mem, em_mul_ex),
4054 	F(DstXacc | Src2Mem, em_imul_ex),
4055 	F(DstXacc | Src2Mem, em_div_ex),
4056 	F(DstXacc | Src2Mem, em_idiv_ex),
4057 };
4058 
4059 static const struct opcode group4[] = {
4060 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4061 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4062 	N, N, N, N, N, N,
4063 };
4064 
4065 static const struct opcode group5[] = {
4066 	F(DstMem | SrcNone | Lock,		em_inc),
4067 	F(DstMem | SrcNone | Lock,		em_dec),
4068 	I(SrcMem | NearBranch | IsBranch,       em_call_near_abs),
4069 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_call_far),
4070 	I(SrcMem | NearBranch | IsBranch,       em_jmp_abs),
4071 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_jmp_far),
4072 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4073 };
4074 
4075 static const struct opcode group6[] = {
4076 	II(Prot | DstMem,	   em_sldt, sldt),
4077 	II(Prot | DstMem,	   em_str, str),
4078 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4079 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4080 	N, N, N, N,
4081 };
4082 
4083 static const struct group_dual group7 = { {
4084 	II(Mov | DstMem,			em_sgdt, sgdt),
4085 	II(Mov | DstMem,			em_sidt, sidt),
4086 	II(SrcMem | Priv,			em_lgdt, lgdt),
4087 	II(SrcMem | Priv,			em_lidt, lidt),
4088 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4089 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4090 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4091 }, {
4092 	EXT(0, group7_rm0),
4093 	EXT(0, group7_rm1),
4094 	EXT(0, group7_rm2),
4095 	EXT(0, group7_rm3),
4096 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4097 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4098 	EXT(0, group7_rm7),
4099 } };
4100 
4101 static const struct opcode group8[] = {
4102 	N, N, N, N,
4103 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4104 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4105 	F(DstMem | SrcImmByte | Lock,			em_btr),
4106 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4107 };
4108 
4109 /*
4110  * The "memory" destination is actually always a register, since we come
4111  * from the register case of group9.
4112  */
4113 static const struct gprefix pfx_0f_c7_7 = {
4114 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4115 };
4116 
4117 
4118 static const struct group_dual group9 = { {
4119 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4120 }, {
4121 	N, N, N, N, N, N, N,
4122 	GP(0, &pfx_0f_c7_7),
4123 } };
4124 
4125 static const struct opcode group11[] = {
4126 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4127 	X7(D(Undefined)),
4128 };
4129 
4130 static const struct gprefix pfx_0f_ae_7 = {
4131 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4132 };
4133 
4134 static const struct group_dual group15 = { {
4135 	I(ModRM | Aligned16, em_fxsave),
4136 	I(ModRM | Aligned16, em_fxrstor),
4137 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4138 }, {
4139 	N, N, N, N, N, N, N, N,
4140 } };
4141 
4142 static const struct gprefix pfx_0f_6f_0f_7f = {
4143 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4144 };
4145 
4146 static const struct instr_dual instr_dual_0f_2b = {
4147 	I(0, em_mov), N
4148 };
4149 
4150 static const struct gprefix pfx_0f_2b = {
4151 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4152 };
4153 
4154 static const struct gprefix pfx_0f_10_0f_11 = {
4155 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4156 };
4157 
4158 static const struct gprefix pfx_0f_28_0f_29 = {
4159 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4160 };
4161 
4162 static const struct gprefix pfx_0f_e7 = {
4163 	N, I(Sse, em_mov), N, N,
4164 };
4165 
4166 static const struct escape escape_d9 = { {
4167 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4168 }, {
4169 	/* 0xC0 - 0xC7 */
4170 	N, N, N, N, N, N, N, N,
4171 	/* 0xC8 - 0xCF */
4172 	N, N, N, N, N, N, N, N,
4173 	/* 0xD0 - 0xC7 */
4174 	N, N, N, N, N, N, N, N,
4175 	/* 0xD8 - 0xDF */
4176 	N, N, N, N, N, N, N, N,
4177 	/* 0xE0 - 0xE7 */
4178 	N, N, N, N, N, N, N, N,
4179 	/* 0xE8 - 0xEF */
4180 	N, N, N, N, N, N, N, N,
4181 	/* 0xF0 - 0xF7 */
4182 	N, N, N, N, N, N, N, N,
4183 	/* 0xF8 - 0xFF */
4184 	N, N, N, N, N, N, N, N,
4185 } };
4186 
4187 static const struct escape escape_db = { {
4188 	N, N, N, N, N, N, N, N,
4189 }, {
4190 	/* 0xC0 - 0xC7 */
4191 	N, N, N, N, N, N, N, N,
4192 	/* 0xC8 - 0xCF */
4193 	N, N, N, N, N, N, N, N,
4194 	/* 0xD0 - 0xC7 */
4195 	N, N, N, N, N, N, N, N,
4196 	/* 0xD8 - 0xDF */
4197 	N, N, N, N, N, N, N, N,
4198 	/* 0xE0 - 0xE7 */
4199 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4200 	/* 0xE8 - 0xEF */
4201 	N, N, N, N, N, N, N, N,
4202 	/* 0xF0 - 0xF7 */
4203 	N, N, N, N, N, N, N, N,
4204 	/* 0xF8 - 0xFF */
4205 	N, N, N, N, N, N, N, N,
4206 } };
4207 
4208 static const struct escape escape_dd = { {
4209 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4210 }, {
4211 	/* 0xC0 - 0xC7 */
4212 	N, N, N, N, N, N, N, N,
4213 	/* 0xC8 - 0xCF */
4214 	N, N, N, N, N, N, N, N,
4215 	/* 0xD0 - 0xC7 */
4216 	N, N, N, N, N, N, N, N,
4217 	/* 0xD8 - 0xDF */
4218 	N, N, N, N, N, N, N, N,
4219 	/* 0xE0 - 0xE7 */
4220 	N, N, N, N, N, N, N, N,
4221 	/* 0xE8 - 0xEF */
4222 	N, N, N, N, N, N, N, N,
4223 	/* 0xF0 - 0xF7 */
4224 	N, N, N, N, N, N, N, N,
4225 	/* 0xF8 - 0xFF */
4226 	N, N, N, N, N, N, N, N,
4227 } };
4228 
4229 static const struct instr_dual instr_dual_0f_c3 = {
4230 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4231 };
4232 
4233 static const struct mode_dual mode_dual_63 = {
4234 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4235 };
4236 
4237 static const struct instr_dual instr_dual_8d = {
4238 	D(DstReg | SrcMem | ModRM | NoAccess), N
4239 };
4240 
4241 static const struct opcode opcode_table[256] = {
4242 	/* 0x00 - 0x07 */
4243 	F6ALU(Lock, em_add),
4244 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4245 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4246 	/* 0x08 - 0x0F */
4247 	F6ALU(Lock | PageTable, em_or),
4248 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4249 	N,
4250 	/* 0x10 - 0x17 */
4251 	F6ALU(Lock, em_adc),
4252 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4253 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4254 	/* 0x18 - 0x1F */
4255 	F6ALU(Lock, em_sbb),
4256 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4257 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4258 	/* 0x20 - 0x27 */
4259 	F6ALU(Lock | PageTable, em_and), N, N,
4260 	/* 0x28 - 0x2F */
4261 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4262 	/* 0x30 - 0x37 */
4263 	F6ALU(Lock, em_xor), N, N,
4264 	/* 0x38 - 0x3F */
4265 	F6ALU(NoWrite, em_cmp), N, N,
4266 	/* 0x40 - 0x4F */
4267 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4268 	/* 0x50 - 0x57 */
4269 	X8(I(SrcReg | Stack, em_push)),
4270 	/* 0x58 - 0x5F */
4271 	X8(I(DstReg | Stack, em_pop)),
4272 	/* 0x60 - 0x67 */
4273 	I(ImplicitOps | Stack | No64, em_pusha),
4274 	I(ImplicitOps | Stack | No64, em_popa),
4275 	N, MD(ModRM, &mode_dual_63),
4276 	N, N, N, N,
4277 	/* 0x68 - 0x6F */
4278 	I(SrcImm | Mov | Stack, em_push),
4279 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4280 	I(SrcImmByte | Mov | Stack, em_push),
4281 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4282 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4283 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4284 	/* 0x70 - 0x7F */
4285 	X16(D(SrcImmByte | NearBranch | IsBranch)),
4286 	/* 0x80 - 0x87 */
4287 	G(ByteOp | DstMem | SrcImm, group1),
4288 	G(DstMem | SrcImm, group1),
4289 	G(ByteOp | DstMem | SrcImm | No64, group1),
4290 	G(DstMem | SrcImmByte, group1),
4291 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4292 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4293 	/* 0x88 - 0x8F */
4294 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4295 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4296 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4297 	ID(0, &instr_dual_8d),
4298 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4299 	G(0, group1A),
4300 	/* 0x90 - 0x97 */
4301 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4302 	/* 0x98 - 0x9F */
4303 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4304 	I(SrcImmFAddr | No64 | IsBranch, em_call_far), N,
4305 	II(ImplicitOps | Stack, em_pushf, pushf),
4306 	II(ImplicitOps | Stack, em_popf, popf),
4307 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4308 	/* 0xA0 - 0xA7 */
4309 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4310 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4311 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4312 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4313 	/* 0xA8 - 0xAF */
4314 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4315 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4316 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4317 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4318 	/* 0xB0 - 0xB7 */
4319 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4320 	/* 0xB8 - 0xBF */
4321 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4322 	/* 0xC0 - 0xC7 */
4323 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4324 	I(ImplicitOps | NearBranch | SrcImmU16 | IsBranch, em_ret_near_imm),
4325 	I(ImplicitOps | NearBranch | IsBranch, em_ret),
4326 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4327 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4328 	G(ByteOp, group11), G(0, group11),
4329 	/* 0xC8 - 0xCF */
4330 	I(Stack | SrcImmU16 | Src2ImmByte | IsBranch, em_enter),
4331 	I(Stack | IsBranch, em_leave),
4332 	I(ImplicitOps | SrcImmU16 | IsBranch, em_ret_far_imm),
4333 	I(ImplicitOps | IsBranch, em_ret_far),
4334 	D(ImplicitOps | IsBranch), DI(SrcImmByte | IsBranch, intn),
4335 	D(ImplicitOps | No64 | IsBranch),
4336 	II(ImplicitOps | IsBranch, em_iret, iret),
4337 	/* 0xD0 - 0xD7 */
4338 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4339 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4340 	I(DstAcc | SrcImmUByte | No64, em_aam),
4341 	I(DstAcc | SrcImmUByte | No64, em_aad),
4342 	F(DstAcc | ByteOp | No64, em_salc),
4343 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4344 	/* 0xD8 - 0xDF */
4345 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4346 	/* 0xE0 - 0xE7 */
4347 	X3(I(SrcImmByte | NearBranch | IsBranch, em_loop)),
4348 	I(SrcImmByte | NearBranch | IsBranch, em_jcxz),
4349 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4350 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4351 	/* 0xE8 - 0xEF */
4352 	I(SrcImm | NearBranch | IsBranch, em_call),
4353 	D(SrcImm | ImplicitOps | NearBranch | IsBranch),
4354 	I(SrcImmFAddr | No64 | IsBranch, em_jmp_far),
4355 	D(SrcImmByte | ImplicitOps | NearBranch | IsBranch),
4356 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4357 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4358 	/* 0xF0 - 0xF7 */
4359 	N, DI(ImplicitOps, icebp), N, N,
4360 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4361 	G(ByteOp, group3), G(0, group3),
4362 	/* 0xF8 - 0xFF */
4363 	D(ImplicitOps), D(ImplicitOps),
4364 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4365 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4366 };
4367 
4368 static const struct opcode twobyte_table[256] = {
4369 	/* 0x00 - 0x0F */
4370 	G(0, group6), GD(0, &group7), N, N,
4371 	N, I(ImplicitOps | EmulateOnUD | IsBranch, em_syscall),
4372 	II(ImplicitOps | Priv, em_clts, clts), N,
4373 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4374 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4375 	/* 0x10 - 0x1F */
4376 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4377 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4378 	N, N, N, N, N, N,
4379 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4380 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4381 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4382 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4383 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4384 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4385 	/* 0x20 - 0x2F */
4386 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4387 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4388 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4389 						check_cr_access),
4390 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4391 						check_dr_write),
4392 	N, N, N, N,
4393 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4394 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4395 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4396 	N, N, N, N,
4397 	/* 0x30 - 0x3F */
4398 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4399 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4400 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4401 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4402 	I(ImplicitOps | EmulateOnUD | IsBranch, em_sysenter),
4403 	I(ImplicitOps | Priv | EmulateOnUD | IsBranch, em_sysexit),
4404 	N, N,
4405 	N, N, N, N, N, N, N, N,
4406 	/* 0x40 - 0x4F */
4407 	X16(D(DstReg | SrcMem | ModRM)),
4408 	/* 0x50 - 0x5F */
4409 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4410 	/* 0x60 - 0x6F */
4411 	N, N, N, N,
4412 	N, N, N, N,
4413 	N, N, N, N,
4414 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4415 	/* 0x70 - 0x7F */
4416 	N, N, N, N,
4417 	N, N, N, N,
4418 	N, N, N, N,
4419 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4420 	/* 0x80 - 0x8F */
4421 	X16(D(SrcImm | NearBranch | IsBranch)),
4422 	/* 0x90 - 0x9F */
4423 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4424 	/* 0xA0 - 0xA7 */
4425 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4426 	II(ImplicitOps, em_cpuid, cpuid),
4427 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4428 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4429 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4430 	/* 0xA8 - 0xAF */
4431 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4432 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4433 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4434 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4435 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4436 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4437 	/* 0xB0 - 0xB7 */
4438 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4439 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4440 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4441 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4442 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4443 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4444 	/* 0xB8 - 0xBF */
4445 	N, N,
4446 	G(BitOp, group8),
4447 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4448 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4449 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4450 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4451 	/* 0xC0 - 0xC7 */
4452 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4453 	N, ID(0, &instr_dual_0f_c3),
4454 	N, N, N, GD(0, &group9),
4455 	/* 0xC8 - 0xCF */
4456 	X8(I(DstReg, em_bswap)),
4457 	/* 0xD0 - 0xDF */
4458 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4459 	/* 0xE0 - 0xEF */
4460 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4461 	N, N, N, N, N, N, N, N,
4462 	/* 0xF0 - 0xFF */
4463 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4464 };
4465 
4466 static const struct instr_dual instr_dual_0f_38_f0 = {
4467 	I(DstReg | SrcMem | Mov, em_movbe), N
4468 };
4469 
4470 static const struct instr_dual instr_dual_0f_38_f1 = {
4471 	I(DstMem | SrcReg | Mov, em_movbe), N
4472 };
4473 
4474 static const struct gprefix three_byte_0f_38_f0 = {
4475 	ID(0, &instr_dual_0f_38_f0), ID(0, &instr_dual_0f_38_f0), N, N
4476 };
4477 
4478 static const struct gprefix three_byte_0f_38_f1 = {
4479 	ID(0, &instr_dual_0f_38_f1), ID(0, &instr_dual_0f_38_f1), N, N
4480 };
4481 
4482 /*
4483  * Insns below are selected by the prefix which indexed by the third opcode
4484  * byte.
4485  */
4486 static const struct opcode opcode_map_0f_38[256] = {
4487 	/* 0x00 - 0x7f */
4488 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4489 	/* 0x80 - 0xef */
4490 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4491 	/* 0xf0 - 0xf1 */
4492 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4493 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4494 	/* 0xf2 - 0xff */
4495 	N, N, X4(N), X8(N)
4496 };
4497 
4498 #undef D
4499 #undef N
4500 #undef G
4501 #undef GD
4502 #undef I
4503 #undef GP
4504 #undef EXT
4505 #undef MD
4506 #undef ID
4507 
4508 #undef D2bv
4509 #undef D2bvIP
4510 #undef I2bv
4511 #undef I2bvIP
4512 #undef I6ALU
4513 
4514 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4515 {
4516 	unsigned size;
4517 
4518 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4519 	if (size == 8)
4520 		size = 4;
4521 	return size;
4522 }
4523 
4524 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4525 		      unsigned size, bool sign_extension)
4526 {
4527 	int rc = X86EMUL_CONTINUE;
4528 
4529 	op->type = OP_IMM;
4530 	op->bytes = size;
4531 	op->addr.mem.ea = ctxt->_eip;
4532 	/* NB. Immediates are sign-extended as necessary. */
4533 	switch (op->bytes) {
4534 	case 1:
4535 		op->val = insn_fetch(s8, ctxt);
4536 		break;
4537 	case 2:
4538 		op->val = insn_fetch(s16, ctxt);
4539 		break;
4540 	case 4:
4541 		op->val = insn_fetch(s32, ctxt);
4542 		break;
4543 	case 8:
4544 		op->val = insn_fetch(s64, ctxt);
4545 		break;
4546 	}
4547 	if (!sign_extension) {
4548 		switch (op->bytes) {
4549 		case 1:
4550 			op->val &= 0xff;
4551 			break;
4552 		case 2:
4553 			op->val &= 0xffff;
4554 			break;
4555 		case 4:
4556 			op->val &= 0xffffffff;
4557 			break;
4558 		}
4559 	}
4560 done:
4561 	return rc;
4562 }
4563 
4564 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4565 			  unsigned d)
4566 {
4567 	int rc = X86EMUL_CONTINUE;
4568 
4569 	switch (d) {
4570 	case OpReg:
4571 		decode_register_operand(ctxt, op);
4572 		break;
4573 	case OpImmUByte:
4574 		rc = decode_imm(ctxt, op, 1, false);
4575 		break;
4576 	case OpMem:
4577 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4578 	mem_common:
4579 		*op = ctxt->memop;
4580 		ctxt->memopp = op;
4581 		if (ctxt->d & BitOp)
4582 			fetch_bit_operand(ctxt);
4583 		op->orig_val = op->val;
4584 		break;
4585 	case OpMem64:
4586 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4587 		goto mem_common;
4588 	case OpAcc:
4589 		op->type = OP_REG;
4590 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4591 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4592 		fetch_register_operand(op);
4593 		op->orig_val = op->val;
4594 		break;
4595 	case OpAccLo:
4596 		op->type = OP_REG;
4597 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4598 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4599 		fetch_register_operand(op);
4600 		op->orig_val = op->val;
4601 		break;
4602 	case OpAccHi:
4603 		if (ctxt->d & ByteOp) {
4604 			op->type = OP_NONE;
4605 			break;
4606 		}
4607 		op->type = OP_REG;
4608 		op->bytes = ctxt->op_bytes;
4609 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4610 		fetch_register_operand(op);
4611 		op->orig_val = op->val;
4612 		break;
4613 	case OpDI:
4614 		op->type = OP_MEM;
4615 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4616 		op->addr.mem.ea =
4617 			register_address(ctxt, VCPU_REGS_RDI);
4618 		op->addr.mem.seg = VCPU_SREG_ES;
4619 		op->val = 0;
4620 		op->count = 1;
4621 		break;
4622 	case OpDX:
4623 		op->type = OP_REG;
4624 		op->bytes = 2;
4625 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4626 		fetch_register_operand(op);
4627 		break;
4628 	case OpCL:
4629 		op->type = OP_IMM;
4630 		op->bytes = 1;
4631 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4632 		break;
4633 	case OpImmByte:
4634 		rc = decode_imm(ctxt, op, 1, true);
4635 		break;
4636 	case OpOne:
4637 		op->type = OP_IMM;
4638 		op->bytes = 1;
4639 		op->val = 1;
4640 		break;
4641 	case OpImm:
4642 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4643 		break;
4644 	case OpImm64:
4645 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4646 		break;
4647 	case OpMem8:
4648 		ctxt->memop.bytes = 1;
4649 		if (ctxt->memop.type == OP_REG) {
4650 			ctxt->memop.addr.reg = decode_register(ctxt,
4651 					ctxt->modrm_rm, true);
4652 			fetch_register_operand(&ctxt->memop);
4653 		}
4654 		goto mem_common;
4655 	case OpMem16:
4656 		ctxt->memop.bytes = 2;
4657 		goto mem_common;
4658 	case OpMem32:
4659 		ctxt->memop.bytes = 4;
4660 		goto mem_common;
4661 	case OpImmU16:
4662 		rc = decode_imm(ctxt, op, 2, false);
4663 		break;
4664 	case OpImmU:
4665 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4666 		break;
4667 	case OpSI:
4668 		op->type = OP_MEM;
4669 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4670 		op->addr.mem.ea =
4671 			register_address(ctxt, VCPU_REGS_RSI);
4672 		op->addr.mem.seg = ctxt->seg_override;
4673 		op->val = 0;
4674 		op->count = 1;
4675 		break;
4676 	case OpXLat:
4677 		op->type = OP_MEM;
4678 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4679 		op->addr.mem.ea =
4680 			address_mask(ctxt,
4681 				reg_read(ctxt, VCPU_REGS_RBX) +
4682 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
4683 		op->addr.mem.seg = ctxt->seg_override;
4684 		op->val = 0;
4685 		break;
4686 	case OpImmFAddr:
4687 		op->type = OP_IMM;
4688 		op->addr.mem.ea = ctxt->_eip;
4689 		op->bytes = ctxt->op_bytes + 2;
4690 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
4691 		break;
4692 	case OpMemFAddr:
4693 		ctxt->memop.bytes = ctxt->op_bytes + 2;
4694 		goto mem_common;
4695 	case OpES:
4696 		op->type = OP_IMM;
4697 		op->val = VCPU_SREG_ES;
4698 		break;
4699 	case OpCS:
4700 		op->type = OP_IMM;
4701 		op->val = VCPU_SREG_CS;
4702 		break;
4703 	case OpSS:
4704 		op->type = OP_IMM;
4705 		op->val = VCPU_SREG_SS;
4706 		break;
4707 	case OpDS:
4708 		op->type = OP_IMM;
4709 		op->val = VCPU_SREG_DS;
4710 		break;
4711 	case OpFS:
4712 		op->type = OP_IMM;
4713 		op->val = VCPU_SREG_FS;
4714 		break;
4715 	case OpGS:
4716 		op->type = OP_IMM;
4717 		op->val = VCPU_SREG_GS;
4718 		break;
4719 	case OpImplicit:
4720 		/* Special instructions do their own operand decoding. */
4721 	default:
4722 		op->type = OP_NONE; /* Disable writeback. */
4723 		break;
4724 	}
4725 
4726 done:
4727 	return rc;
4728 }
4729 
4730 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
4731 {
4732 	int rc = X86EMUL_CONTINUE;
4733 	int mode = ctxt->mode;
4734 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
4735 	bool op_prefix = false;
4736 	bool has_seg_override = false;
4737 	struct opcode opcode;
4738 	u16 dummy;
4739 	struct desc_struct desc;
4740 
4741 	ctxt->memop.type = OP_NONE;
4742 	ctxt->memopp = NULL;
4743 	ctxt->_eip = ctxt->eip;
4744 	ctxt->fetch.ptr = ctxt->fetch.data;
4745 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
4746 	ctxt->opcode_len = 1;
4747 	ctxt->intercept = x86_intercept_none;
4748 	if (insn_len > 0)
4749 		memcpy(ctxt->fetch.data, insn, insn_len);
4750 	else {
4751 		rc = __do_insn_fetch_bytes(ctxt, 1);
4752 		if (rc != X86EMUL_CONTINUE)
4753 			goto done;
4754 	}
4755 
4756 	switch (mode) {
4757 	case X86EMUL_MODE_REAL:
4758 	case X86EMUL_MODE_VM86:
4759 		def_op_bytes = def_ad_bytes = 2;
4760 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
4761 		if (desc.d)
4762 			def_op_bytes = def_ad_bytes = 4;
4763 		break;
4764 	case X86EMUL_MODE_PROT16:
4765 		def_op_bytes = def_ad_bytes = 2;
4766 		break;
4767 	case X86EMUL_MODE_PROT32:
4768 		def_op_bytes = def_ad_bytes = 4;
4769 		break;
4770 #ifdef CONFIG_X86_64
4771 	case X86EMUL_MODE_PROT64:
4772 		def_op_bytes = 4;
4773 		def_ad_bytes = 8;
4774 		break;
4775 #endif
4776 	default:
4777 		return EMULATION_FAILED;
4778 	}
4779 
4780 	ctxt->op_bytes = def_op_bytes;
4781 	ctxt->ad_bytes = def_ad_bytes;
4782 
4783 	/* Legacy prefixes. */
4784 	for (;;) {
4785 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
4786 		case 0x66:	/* operand-size override */
4787 			op_prefix = true;
4788 			/* switch between 2/4 bytes */
4789 			ctxt->op_bytes = def_op_bytes ^ 6;
4790 			break;
4791 		case 0x67:	/* address-size override */
4792 			if (mode == X86EMUL_MODE_PROT64)
4793 				/* switch between 4/8 bytes */
4794 				ctxt->ad_bytes = def_ad_bytes ^ 12;
4795 			else
4796 				/* switch between 2/4 bytes */
4797 				ctxt->ad_bytes = def_ad_bytes ^ 6;
4798 			break;
4799 		case 0x26:	/* ES override */
4800 			has_seg_override = true;
4801 			ctxt->seg_override = VCPU_SREG_ES;
4802 			break;
4803 		case 0x2e:	/* CS override */
4804 			has_seg_override = true;
4805 			ctxt->seg_override = VCPU_SREG_CS;
4806 			break;
4807 		case 0x36:	/* SS override */
4808 			has_seg_override = true;
4809 			ctxt->seg_override = VCPU_SREG_SS;
4810 			break;
4811 		case 0x3e:	/* DS override */
4812 			has_seg_override = true;
4813 			ctxt->seg_override = VCPU_SREG_DS;
4814 			break;
4815 		case 0x64:	/* FS override */
4816 			has_seg_override = true;
4817 			ctxt->seg_override = VCPU_SREG_FS;
4818 			break;
4819 		case 0x65:	/* GS override */
4820 			has_seg_override = true;
4821 			ctxt->seg_override = VCPU_SREG_GS;
4822 			break;
4823 		case 0x40 ... 0x4f: /* REX */
4824 			if (mode != X86EMUL_MODE_PROT64)
4825 				goto done_prefixes;
4826 			ctxt->rex_prefix = ctxt->b;
4827 			continue;
4828 		case 0xf0:	/* LOCK */
4829 			ctxt->lock_prefix = 1;
4830 			break;
4831 		case 0xf2:	/* REPNE/REPNZ */
4832 		case 0xf3:	/* REP/REPE/REPZ */
4833 			ctxt->rep_prefix = ctxt->b;
4834 			break;
4835 		default:
4836 			goto done_prefixes;
4837 		}
4838 
4839 		/* Any legacy prefix after a REX prefix nullifies its effect. */
4840 
4841 		ctxt->rex_prefix = 0;
4842 	}
4843 
4844 done_prefixes:
4845 
4846 	/* REX prefix. */
4847 	if (ctxt->rex_prefix & 8)
4848 		ctxt->op_bytes = 8;	/* REX.W */
4849 
4850 	/* Opcode byte(s). */
4851 	opcode = opcode_table[ctxt->b];
4852 	/* Two-byte opcode? */
4853 	if (ctxt->b == 0x0f) {
4854 		ctxt->opcode_len = 2;
4855 		ctxt->b = insn_fetch(u8, ctxt);
4856 		opcode = twobyte_table[ctxt->b];
4857 
4858 		/* 0F_38 opcode map */
4859 		if (ctxt->b == 0x38) {
4860 			ctxt->opcode_len = 3;
4861 			ctxt->b = insn_fetch(u8, ctxt);
4862 			opcode = opcode_map_0f_38[ctxt->b];
4863 		}
4864 	}
4865 	ctxt->d = opcode.flags;
4866 
4867 	if (ctxt->d & ModRM)
4868 		ctxt->modrm = insn_fetch(u8, ctxt);
4869 
4870 	/* vex-prefix instructions are not implemented */
4871 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
4872 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
4873 		ctxt->d = NotImpl;
4874 	}
4875 
4876 	while (ctxt->d & GroupMask) {
4877 		switch (ctxt->d & GroupMask) {
4878 		case Group:
4879 			goffset = (ctxt->modrm >> 3) & 7;
4880 			opcode = opcode.u.group[goffset];
4881 			break;
4882 		case GroupDual:
4883 			goffset = (ctxt->modrm >> 3) & 7;
4884 			if ((ctxt->modrm >> 6) == 3)
4885 				opcode = opcode.u.gdual->mod3[goffset];
4886 			else
4887 				opcode = opcode.u.gdual->mod012[goffset];
4888 			break;
4889 		case RMExt:
4890 			goffset = ctxt->modrm & 7;
4891 			opcode = opcode.u.group[goffset];
4892 			break;
4893 		case Prefix:
4894 			if (ctxt->rep_prefix && op_prefix)
4895 				return EMULATION_FAILED;
4896 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
4897 			switch (simd_prefix) {
4898 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
4899 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
4900 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
4901 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
4902 			}
4903 			break;
4904 		case Escape:
4905 			if (ctxt->modrm > 0xbf) {
4906 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
4907 				u32 index = array_index_nospec(
4908 					ctxt->modrm - 0xc0, size);
4909 
4910 				opcode = opcode.u.esc->high[index];
4911 			} else {
4912 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
4913 			}
4914 			break;
4915 		case InstrDual:
4916 			if ((ctxt->modrm >> 6) == 3)
4917 				opcode = opcode.u.idual->mod3;
4918 			else
4919 				opcode = opcode.u.idual->mod012;
4920 			break;
4921 		case ModeDual:
4922 			if (ctxt->mode == X86EMUL_MODE_PROT64)
4923 				opcode = opcode.u.mdual->mode64;
4924 			else
4925 				opcode = opcode.u.mdual->mode32;
4926 			break;
4927 		default:
4928 			return EMULATION_FAILED;
4929 		}
4930 
4931 		ctxt->d &= ~(u64)GroupMask;
4932 		ctxt->d |= opcode.flags;
4933 	}
4934 
4935 	ctxt->is_branch = opcode.flags & IsBranch;
4936 
4937 	/* Unrecognised? */
4938 	if (ctxt->d == 0)
4939 		return EMULATION_FAILED;
4940 
4941 	ctxt->execute = opcode.u.execute;
4942 
4943 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
4944 	    likely(!(ctxt->d & EmulateOnUD)))
4945 		return EMULATION_FAILED;
4946 
4947 	if (unlikely(ctxt->d &
4948 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
4949 	     No16))) {
4950 		/*
4951 		 * These are copied unconditionally here, and checked unconditionally
4952 		 * in x86_emulate_insn.
4953 		 */
4954 		ctxt->check_perm = opcode.check_perm;
4955 		ctxt->intercept = opcode.intercept;
4956 
4957 		if (ctxt->d & NotImpl)
4958 			return EMULATION_FAILED;
4959 
4960 		if (mode == X86EMUL_MODE_PROT64) {
4961 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
4962 				ctxt->op_bytes = 8;
4963 			else if (ctxt->d & NearBranch)
4964 				ctxt->op_bytes = 8;
4965 		}
4966 
4967 		if (ctxt->d & Op3264) {
4968 			if (mode == X86EMUL_MODE_PROT64)
4969 				ctxt->op_bytes = 8;
4970 			else
4971 				ctxt->op_bytes = 4;
4972 		}
4973 
4974 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
4975 			ctxt->op_bytes = 4;
4976 
4977 		if (ctxt->d & Sse)
4978 			ctxt->op_bytes = 16;
4979 		else if (ctxt->d & Mmx)
4980 			ctxt->op_bytes = 8;
4981 	}
4982 
4983 	/* ModRM and SIB bytes. */
4984 	if (ctxt->d & ModRM) {
4985 		rc = decode_modrm(ctxt, &ctxt->memop);
4986 		if (!has_seg_override) {
4987 			has_seg_override = true;
4988 			ctxt->seg_override = ctxt->modrm_seg;
4989 		}
4990 	} else if (ctxt->d & MemAbs)
4991 		rc = decode_abs(ctxt, &ctxt->memop);
4992 	if (rc != X86EMUL_CONTINUE)
4993 		goto done;
4994 
4995 	if (!has_seg_override)
4996 		ctxt->seg_override = VCPU_SREG_DS;
4997 
4998 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
4999 
5000 	/*
5001 	 * Decode and fetch the source operand: register, memory
5002 	 * or immediate.
5003 	 */
5004 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5005 	if (rc != X86EMUL_CONTINUE)
5006 		goto done;
5007 
5008 	/*
5009 	 * Decode and fetch the second source operand: register, memory
5010 	 * or immediate.
5011 	 */
5012 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5013 	if (rc != X86EMUL_CONTINUE)
5014 		goto done;
5015 
5016 	/* Decode and fetch the destination operand: register or memory. */
5017 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5018 
5019 	if (ctxt->rip_relative && likely(ctxt->memopp))
5020 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5021 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5022 
5023 done:
5024 	if (rc == X86EMUL_PROPAGATE_FAULT)
5025 		ctxt->have_exception = true;
5026 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5027 }
5028 
5029 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5030 {
5031 	return ctxt->d & PageTable;
5032 }
5033 
5034 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5035 {
5036 	/* The second termination condition only applies for REPE
5037 	 * and REPNE. Test if the repeat string operation prefix is
5038 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5039 	 * corresponding termination condition according to:
5040 	 * 	- if REPE/REPZ and ZF = 0 then done
5041 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5042 	 */
5043 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5044 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5045 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5046 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5047 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5048 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5049 		return true;
5050 
5051 	return false;
5052 }
5053 
5054 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5055 {
5056 	int rc;
5057 
5058 	kvm_fpu_get();
5059 	rc = asm_safe("fwait");
5060 	kvm_fpu_put();
5061 
5062 	if (unlikely(rc != X86EMUL_CONTINUE))
5063 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5064 
5065 	return X86EMUL_CONTINUE;
5066 }
5067 
5068 static void fetch_possible_mmx_operand(struct operand *op)
5069 {
5070 	if (op->type == OP_MM)
5071 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5072 }
5073 
5074 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5075 {
5076 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5077 
5078 	if (!(ctxt->d & ByteOp))
5079 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5080 
5081 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5082 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5083 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5084 	    : "c"(ctxt->src2.val));
5085 
5086 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5087 	if (!fop) /* exception is returned in fop variable */
5088 		return emulate_de(ctxt);
5089 	return X86EMUL_CONTINUE;
5090 }
5091 
5092 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5093 {
5094 	/* Clear fields that are set conditionally but read without a guard. */
5095 	ctxt->rip_relative = false;
5096 	ctxt->rex_prefix = 0;
5097 	ctxt->lock_prefix = 0;
5098 	ctxt->rep_prefix = 0;
5099 	ctxt->regs_valid = 0;
5100 	ctxt->regs_dirty = 0;
5101 
5102 	ctxt->io_read.pos = 0;
5103 	ctxt->io_read.end = 0;
5104 	ctxt->mem_read.end = 0;
5105 }
5106 
5107 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5108 {
5109 	const struct x86_emulate_ops *ops = ctxt->ops;
5110 	int rc = X86EMUL_CONTINUE;
5111 	int saved_dst_type = ctxt->dst.type;
5112 	bool is_guest_mode = ctxt->ops->is_guest_mode(ctxt);
5113 
5114 	ctxt->mem_read.pos = 0;
5115 
5116 	/* LOCK prefix is allowed only with some instructions */
5117 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5118 		rc = emulate_ud(ctxt);
5119 		goto done;
5120 	}
5121 
5122 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5123 		rc = emulate_ud(ctxt);
5124 		goto done;
5125 	}
5126 
5127 	if (unlikely(ctxt->d &
5128 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5129 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5130 				(ctxt->d & Undefined)) {
5131 			rc = emulate_ud(ctxt);
5132 			goto done;
5133 		}
5134 
5135 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5136 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5137 			rc = emulate_ud(ctxt);
5138 			goto done;
5139 		}
5140 
5141 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5142 			rc = emulate_nm(ctxt);
5143 			goto done;
5144 		}
5145 
5146 		if (ctxt->d & Mmx) {
5147 			rc = flush_pending_x87_faults(ctxt);
5148 			if (rc != X86EMUL_CONTINUE)
5149 				goto done;
5150 			/*
5151 			 * Now that we know the fpu is exception safe, we can fetch
5152 			 * operands from it.
5153 			 */
5154 			fetch_possible_mmx_operand(&ctxt->src);
5155 			fetch_possible_mmx_operand(&ctxt->src2);
5156 			if (!(ctxt->d & Mov))
5157 				fetch_possible_mmx_operand(&ctxt->dst);
5158 		}
5159 
5160 		if (unlikely(is_guest_mode) && ctxt->intercept) {
5161 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5162 						      X86_ICPT_PRE_EXCEPT);
5163 			if (rc != X86EMUL_CONTINUE)
5164 				goto done;
5165 		}
5166 
5167 		/* Instruction can only be executed in protected mode */
5168 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5169 			rc = emulate_ud(ctxt);
5170 			goto done;
5171 		}
5172 
5173 		/* Privileged instruction can be executed only in CPL=0 */
5174 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5175 			if (ctxt->d & PrivUD)
5176 				rc = emulate_ud(ctxt);
5177 			else
5178 				rc = emulate_gp(ctxt, 0);
5179 			goto done;
5180 		}
5181 
5182 		/* Do instruction specific permission checks */
5183 		if (ctxt->d & CheckPerm) {
5184 			rc = ctxt->check_perm(ctxt);
5185 			if (rc != X86EMUL_CONTINUE)
5186 				goto done;
5187 		}
5188 
5189 		if (unlikely(is_guest_mode) && (ctxt->d & Intercept)) {
5190 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5191 						      X86_ICPT_POST_EXCEPT);
5192 			if (rc != X86EMUL_CONTINUE)
5193 				goto done;
5194 		}
5195 
5196 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5197 			/* All REP prefixes have the same first termination condition */
5198 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5199 				string_registers_quirk(ctxt);
5200 				ctxt->eip = ctxt->_eip;
5201 				ctxt->eflags &= ~X86_EFLAGS_RF;
5202 				goto done;
5203 			}
5204 		}
5205 	}
5206 
5207 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5208 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5209 				    ctxt->src.valptr, ctxt->src.bytes);
5210 		if (rc != X86EMUL_CONTINUE)
5211 			goto done;
5212 		ctxt->src.orig_val64 = ctxt->src.val64;
5213 	}
5214 
5215 	if (ctxt->src2.type == OP_MEM) {
5216 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5217 				    &ctxt->src2.val, ctxt->src2.bytes);
5218 		if (rc != X86EMUL_CONTINUE)
5219 			goto done;
5220 	}
5221 
5222 	if ((ctxt->d & DstMask) == ImplicitOps)
5223 		goto special_insn;
5224 
5225 
5226 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5227 		/* optimisation - avoid slow emulated read if Mov */
5228 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5229 				   &ctxt->dst.val, ctxt->dst.bytes);
5230 		if (rc != X86EMUL_CONTINUE) {
5231 			if (!(ctxt->d & NoWrite) &&
5232 			    rc == X86EMUL_PROPAGATE_FAULT &&
5233 			    ctxt->exception.vector == PF_VECTOR)
5234 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5235 			goto done;
5236 		}
5237 	}
5238 	/* Copy full 64-bit value for CMPXCHG8B.  */
5239 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5240 
5241 special_insn:
5242 
5243 	if (unlikely(is_guest_mode) && (ctxt->d & Intercept)) {
5244 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5245 					      X86_ICPT_POST_MEMACCESS);
5246 		if (rc != X86EMUL_CONTINUE)
5247 			goto done;
5248 	}
5249 
5250 	if (ctxt->rep_prefix && (ctxt->d & String))
5251 		ctxt->eflags |= X86_EFLAGS_RF;
5252 	else
5253 		ctxt->eflags &= ~X86_EFLAGS_RF;
5254 
5255 	if (ctxt->execute) {
5256 		if (ctxt->d & Fastop)
5257 			rc = fastop(ctxt, ctxt->fop);
5258 		else
5259 			rc = ctxt->execute(ctxt);
5260 		if (rc != X86EMUL_CONTINUE)
5261 			goto done;
5262 		goto writeback;
5263 	}
5264 
5265 	if (ctxt->opcode_len == 2)
5266 		goto twobyte_insn;
5267 	else if (ctxt->opcode_len == 3)
5268 		goto threebyte_insn;
5269 
5270 	switch (ctxt->b) {
5271 	case 0x70 ... 0x7f: /* jcc (short) */
5272 		if (test_cc(ctxt->b, ctxt->eflags))
5273 			rc = jmp_rel(ctxt, ctxt->src.val);
5274 		break;
5275 	case 0x8d: /* lea r16/r32, m */
5276 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5277 		break;
5278 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5279 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5280 			ctxt->dst.type = OP_NONE;
5281 		else
5282 			rc = em_xchg(ctxt);
5283 		break;
5284 	case 0x98: /* cbw/cwde/cdqe */
5285 		switch (ctxt->op_bytes) {
5286 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5287 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5288 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5289 		}
5290 		break;
5291 	case 0xcc:		/* int3 */
5292 		rc = emulate_int(ctxt, 3);
5293 		break;
5294 	case 0xcd:		/* int n */
5295 		rc = emulate_int(ctxt, ctxt->src.val);
5296 		break;
5297 	case 0xce:		/* into */
5298 		if (ctxt->eflags & X86_EFLAGS_OF)
5299 			rc = emulate_int(ctxt, 4);
5300 		break;
5301 	case 0xe9: /* jmp rel */
5302 	case 0xeb: /* jmp rel short */
5303 		rc = jmp_rel(ctxt, ctxt->src.val);
5304 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5305 		break;
5306 	case 0xf4:              /* hlt */
5307 		ctxt->ops->halt(ctxt);
5308 		break;
5309 	case 0xf5:	/* cmc */
5310 		/* complement carry flag from eflags reg */
5311 		ctxt->eflags ^= X86_EFLAGS_CF;
5312 		break;
5313 	case 0xf8: /* clc */
5314 		ctxt->eflags &= ~X86_EFLAGS_CF;
5315 		break;
5316 	case 0xf9: /* stc */
5317 		ctxt->eflags |= X86_EFLAGS_CF;
5318 		break;
5319 	case 0xfc: /* cld */
5320 		ctxt->eflags &= ~X86_EFLAGS_DF;
5321 		break;
5322 	case 0xfd: /* std */
5323 		ctxt->eflags |= X86_EFLAGS_DF;
5324 		break;
5325 	default:
5326 		goto cannot_emulate;
5327 	}
5328 
5329 	if (rc != X86EMUL_CONTINUE)
5330 		goto done;
5331 
5332 writeback:
5333 	if (ctxt->d & SrcWrite) {
5334 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5335 		rc = writeback(ctxt, &ctxt->src);
5336 		if (rc != X86EMUL_CONTINUE)
5337 			goto done;
5338 	}
5339 	if (!(ctxt->d & NoWrite)) {
5340 		rc = writeback(ctxt, &ctxt->dst);
5341 		if (rc != X86EMUL_CONTINUE)
5342 			goto done;
5343 	}
5344 
5345 	/*
5346 	 * restore dst type in case the decoding will be reused
5347 	 * (happens for string instruction )
5348 	 */
5349 	ctxt->dst.type = saved_dst_type;
5350 
5351 	if ((ctxt->d & SrcMask) == SrcSI)
5352 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5353 
5354 	if ((ctxt->d & DstMask) == DstDI)
5355 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5356 
5357 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5358 		unsigned int count;
5359 		struct read_cache *r = &ctxt->io_read;
5360 		if ((ctxt->d & SrcMask) == SrcSI)
5361 			count = ctxt->src.count;
5362 		else
5363 			count = ctxt->dst.count;
5364 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5365 
5366 		if (!string_insn_completed(ctxt)) {
5367 			/*
5368 			 * Re-enter guest when pio read ahead buffer is empty
5369 			 * or, if it is not used, after each 1024 iteration.
5370 			 */
5371 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5372 			    (r->end == 0 || r->end != r->pos)) {
5373 				/*
5374 				 * Reset read cache. Usually happens before
5375 				 * decode, but since instruction is restarted
5376 				 * we have to do it here.
5377 				 */
5378 				ctxt->mem_read.end = 0;
5379 				writeback_registers(ctxt);
5380 				return EMULATION_RESTART;
5381 			}
5382 			goto done; /* skip rip writeback */
5383 		}
5384 		ctxt->eflags &= ~X86_EFLAGS_RF;
5385 	}
5386 
5387 	ctxt->eip = ctxt->_eip;
5388 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5389 		ctxt->eip = (u32)ctxt->_eip;
5390 
5391 done:
5392 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5393 		if (KVM_EMULATOR_BUG_ON(ctxt->exception.vector > 0x1f, ctxt))
5394 			return EMULATION_FAILED;
5395 		ctxt->have_exception = true;
5396 	}
5397 	if (rc == X86EMUL_INTERCEPTED)
5398 		return EMULATION_INTERCEPTED;
5399 
5400 	if (rc == X86EMUL_CONTINUE)
5401 		writeback_registers(ctxt);
5402 
5403 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5404 
5405 twobyte_insn:
5406 	switch (ctxt->b) {
5407 	case 0x09:		/* wbinvd */
5408 		(ctxt->ops->wbinvd)(ctxt);
5409 		break;
5410 	case 0x08:		/* invd */
5411 	case 0x0d:		/* GrpP (prefetch) */
5412 	case 0x18:		/* Grp16 (prefetch/nop) */
5413 	case 0x1f:		/* nop */
5414 		break;
5415 	case 0x20: /* mov cr, reg */
5416 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5417 		break;
5418 	case 0x21: /* mov from dr to reg */
5419 		ctxt->dst.val = ops->get_dr(ctxt, ctxt->modrm_reg);
5420 		break;
5421 	case 0x40 ... 0x4f:	/* cmov */
5422 		if (test_cc(ctxt->b, ctxt->eflags))
5423 			ctxt->dst.val = ctxt->src.val;
5424 		else if (ctxt->op_bytes != 4)
5425 			ctxt->dst.type = OP_NONE; /* no writeback */
5426 		break;
5427 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5428 		if (test_cc(ctxt->b, ctxt->eflags))
5429 			rc = jmp_rel(ctxt, ctxt->src.val);
5430 		break;
5431 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5432 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5433 		break;
5434 	case 0xb6 ... 0xb7:	/* movzx */
5435 		ctxt->dst.bytes = ctxt->op_bytes;
5436 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5437 						       : (u16) ctxt->src.val;
5438 		break;
5439 	case 0xbe ... 0xbf:	/* movsx */
5440 		ctxt->dst.bytes = ctxt->op_bytes;
5441 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5442 							(s16) ctxt->src.val;
5443 		break;
5444 	default:
5445 		goto cannot_emulate;
5446 	}
5447 
5448 threebyte_insn:
5449 
5450 	if (rc != X86EMUL_CONTINUE)
5451 		goto done;
5452 
5453 	goto writeback;
5454 
5455 cannot_emulate:
5456 	return EMULATION_FAILED;
5457 }
5458 
5459 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5460 {
5461 	invalidate_registers(ctxt);
5462 }
5463 
5464 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5465 {
5466 	writeback_registers(ctxt);
5467 }
5468 
5469 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5470 {
5471 	if (ctxt->rep_prefix && (ctxt->d & String))
5472 		return false;
5473 
5474 	if (ctxt->d & TwoMemOp)
5475 		return false;
5476 
5477 	return true;
5478 }
5479