xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision 9d5dbfe0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3  * emulate.c
4  *
5  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
6  *
7  * Copyright (c) 2005 Keir Fraser
8  *
9  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
10  * privileged instructions:
11  *
12  * Copyright (C) 2006 Qumranet
13  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
14  *
15  *   Avi Kivity <avi@qumranet.com>
16  *   Yaniv Kamay <yaniv@qumranet.com>
17  *
18  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
19  */
20 
21 #include <linux/kvm_host.h>
22 #include "kvm_cache_regs.h"
23 #include "kvm_emulate.h"
24 #include <linux/stringify.h>
25 #include <asm/debugreg.h>
26 #include <asm/nospec-branch.h>
27 #include <asm/ibt.h>
28 
29 #include "x86.h"
30 #include "tss.h"
31 #include "mmu.h"
32 #include "pmu.h"
33 
34 /*
35  * Operand types
36  */
37 #define OpNone             0ull
38 #define OpImplicit         1ull  /* No generic decode */
39 #define OpReg              2ull  /* Register */
40 #define OpMem              3ull  /* Memory */
41 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
42 #define OpDI               5ull  /* ES:DI/EDI/RDI */
43 #define OpMem64            6ull  /* Memory, 64-bit */
44 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
45 #define OpDX               8ull  /* DX register */
46 #define OpCL               9ull  /* CL register (for shifts) */
47 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
48 #define OpOne             11ull  /* Implied 1 */
49 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
50 #define OpMem16           13ull  /* Memory operand (16-bit). */
51 #define OpMem32           14ull  /* Memory operand (32-bit). */
52 #define OpImmU            15ull  /* Immediate operand, zero extended */
53 #define OpSI              16ull  /* SI/ESI/RSI */
54 #define OpImmFAddr        17ull  /* Immediate far address */
55 #define OpMemFAddr        18ull  /* Far address in memory */
56 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
57 #define OpES              20ull  /* ES */
58 #define OpCS              21ull  /* CS */
59 #define OpSS              22ull  /* SS */
60 #define OpDS              23ull  /* DS */
61 #define OpFS              24ull  /* FS */
62 #define OpGS              25ull  /* GS */
63 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
64 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
65 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
66 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
67 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
68 
69 #define OpBits             5  /* Width of operand field */
70 #define OpMask             ((1ull << OpBits) - 1)
71 
72 /*
73  * Opcode effective-address decode tables.
74  * Note that we only emulate instructions that have at least one memory
75  * operand (excluding implicit stack references). We assume that stack
76  * references and instruction fetches will never occur in special memory
77  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
78  * not be handled.
79  */
80 
81 /* Operand sizes: 8-bit operands or specified/overridden size. */
82 #define ByteOp      (1<<0)	/* 8-bit operands. */
83 /* Destination operand type. */
84 #define DstShift    1
85 #define ImplicitOps (OpImplicit << DstShift)
86 #define DstReg      (OpReg << DstShift)
87 #define DstMem      (OpMem << DstShift)
88 #define DstAcc      (OpAcc << DstShift)
89 #define DstDI       (OpDI << DstShift)
90 #define DstMem64    (OpMem64 << DstShift)
91 #define DstMem16    (OpMem16 << DstShift)
92 #define DstImmUByte (OpImmUByte << DstShift)
93 #define DstDX       (OpDX << DstShift)
94 #define DstAccLo    (OpAccLo << DstShift)
95 #define DstMask     (OpMask << DstShift)
96 /* Source operand type. */
97 #define SrcShift    6
98 #define SrcNone     (OpNone << SrcShift)
99 #define SrcReg      (OpReg << SrcShift)
100 #define SrcMem      (OpMem << SrcShift)
101 #define SrcMem16    (OpMem16 << SrcShift)
102 #define SrcMem32    (OpMem32 << SrcShift)
103 #define SrcImm      (OpImm << SrcShift)
104 #define SrcImmByte  (OpImmByte << SrcShift)
105 #define SrcOne      (OpOne << SrcShift)
106 #define SrcImmUByte (OpImmUByte << SrcShift)
107 #define SrcImmU     (OpImmU << SrcShift)
108 #define SrcSI       (OpSI << SrcShift)
109 #define SrcXLat     (OpXLat << SrcShift)
110 #define SrcImmFAddr (OpImmFAddr << SrcShift)
111 #define SrcMemFAddr (OpMemFAddr << SrcShift)
112 #define SrcAcc      (OpAcc << SrcShift)
113 #define SrcImmU16   (OpImmU16 << SrcShift)
114 #define SrcImm64    (OpImm64 << SrcShift)
115 #define SrcDX       (OpDX << SrcShift)
116 #define SrcMem8     (OpMem8 << SrcShift)
117 #define SrcAccHi    (OpAccHi << SrcShift)
118 #define SrcMask     (OpMask << SrcShift)
119 #define BitOp       (1<<11)
120 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
121 #define String      (1<<13)     /* String instruction (rep capable) */
122 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
123 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
124 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
125 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
126 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
127 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
128 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
129 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
130 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
131 #define Sse         (1<<18)     /* SSE Vector instruction */
132 /* Generic ModRM decode. */
133 #define ModRM       (1<<19)
134 /* Destination is only written; never read. */
135 #define Mov         (1<<20)
136 /* Misc flags */
137 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
138 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
139 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
140 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
141 #define Undefined   (1<<25) /* No Such Instruction */
142 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
143 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
144 #define No64	    (1<<28)
145 #define PageTable   (1 << 29)   /* instruction used to write page table */
146 #define NotImpl     (1 << 30)   /* instruction is not implemented */
147 /* Source 2 operand type */
148 #define Src2Shift   (31)
149 #define Src2None    (OpNone << Src2Shift)
150 #define Src2Mem     (OpMem << Src2Shift)
151 #define Src2CL      (OpCL << Src2Shift)
152 #define Src2ImmByte (OpImmByte << Src2Shift)
153 #define Src2One     (OpOne << Src2Shift)
154 #define Src2Imm     (OpImm << Src2Shift)
155 #define Src2ES      (OpES << Src2Shift)
156 #define Src2CS      (OpCS << Src2Shift)
157 #define Src2SS      (OpSS << Src2Shift)
158 #define Src2DS      (OpDS << Src2Shift)
159 #define Src2FS      (OpFS << Src2Shift)
160 #define Src2GS      (OpGS << Src2Shift)
161 #define Src2Mask    (OpMask << Src2Shift)
162 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
163 #define AlignMask   ((u64)7 << 41)
164 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
165 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
166 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
167 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
168 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
169 #define NoWrite     ((u64)1 << 45)  /* No writeback */
170 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
171 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
172 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
173 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
174 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
175 #define NearBranch  ((u64)1 << 52)  /* Near branches */
176 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
177 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
178 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
179 #define IsBranch    ((u64)1 << 56)  /* Instruction is considered a branch. */
180 
181 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
182 
183 #define X2(x...) x, x
184 #define X3(x...) X2(x), x
185 #define X4(x...) X2(x), X2(x)
186 #define X5(x...) X4(x), x
187 #define X6(x...) X4(x), X2(x)
188 #define X7(x...) X4(x), X3(x)
189 #define X8(x...) X4(x), X4(x)
190 #define X16(x...) X8(x), X8(x)
191 
192 struct opcode {
193 	u64 flags;
194 	u8 intercept;
195 	u8 pad[7];
196 	union {
197 		int (*execute)(struct x86_emulate_ctxt *ctxt);
198 		const struct opcode *group;
199 		const struct group_dual *gdual;
200 		const struct gprefix *gprefix;
201 		const struct escape *esc;
202 		const struct instr_dual *idual;
203 		const struct mode_dual *mdual;
204 		void (*fastop)(struct fastop *fake);
205 	} u;
206 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
207 };
208 
209 struct group_dual {
210 	struct opcode mod012[8];
211 	struct opcode mod3[8];
212 };
213 
214 struct gprefix {
215 	struct opcode pfx_no;
216 	struct opcode pfx_66;
217 	struct opcode pfx_f2;
218 	struct opcode pfx_f3;
219 };
220 
221 struct escape {
222 	struct opcode op[8];
223 	struct opcode high[64];
224 };
225 
226 struct instr_dual {
227 	struct opcode mod012;
228 	struct opcode mod3;
229 };
230 
231 struct mode_dual {
232 	struct opcode mode32;
233 	struct opcode mode64;
234 };
235 
236 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
237 
238 enum x86_transfer_type {
239 	X86_TRANSFER_NONE,
240 	X86_TRANSFER_CALL_JMP,
241 	X86_TRANSFER_RET,
242 	X86_TRANSFER_TASK_SWITCH,
243 };
244 
245 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
246 {
247 	unsigned long dirty = ctxt->regs_dirty;
248 	unsigned reg;
249 
250 	for_each_set_bit(reg, &dirty, NR_EMULATOR_GPRS)
251 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
252 }
253 
254 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
255 {
256 	ctxt->regs_dirty = 0;
257 	ctxt->regs_valid = 0;
258 }
259 
260 /*
261  * These EFLAGS bits are restored from saved value during emulation, and
262  * any changes are written back to the saved value after emulation.
263  */
264 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
265 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
266 
267 #ifdef CONFIG_X86_64
268 #define ON64(x) x
269 #else
270 #define ON64(x)
271 #endif
272 
273 /*
274  * fastop functions have a special calling convention:
275  *
276  * dst:    rax        (in/out)
277  * src:    rdx        (in/out)
278  * src2:   rcx        (in)
279  * flags:  rflags     (in/out)
280  * ex:     rsi        (in:fastop pointer, out:zero if exception)
281  *
282  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
283  * different operand sizes can be reached by calculation, rather than a jump
284  * table (which would be bigger than the code).
285  *
286  * The 16 byte alignment, considering 5 bytes for the RET thunk, 3 for ENDBR
287  * and 1 for the straight line speculation INT3, leaves 7 bytes for the
288  * body of the function.  Currently none is larger than 4.
289  */
290 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
291 
292 #define FASTOP_SIZE	16
293 
294 #define __FOP_FUNC(name) \
295 	".align " __stringify(FASTOP_SIZE) " \n\t" \
296 	".type " name ", @function \n\t" \
297 	name ":\n\t" \
298 	ASM_ENDBR \
299 	IBT_NOSEAL(name)
300 
301 #define FOP_FUNC(name) \
302 	__FOP_FUNC(#name)
303 
304 #define __FOP_RET(name) \
305 	"11: " ASM_RET \
306 	".size " name ", .-" name "\n\t"
307 
308 #define FOP_RET(name) \
309 	__FOP_RET(#name)
310 
311 #define __FOP_START(op, align) \
312 	extern void em_##op(struct fastop *fake); \
313 	asm(".pushsection .text, \"ax\" \n\t" \
314 	    ".global em_" #op " \n\t" \
315 	    ".align " __stringify(align) " \n\t" \
316 	    "em_" #op ":\n\t"
317 
318 #define FOP_START(op) __FOP_START(op, FASTOP_SIZE)
319 
320 #define FOP_END \
321 	    ".popsection")
322 
323 #define __FOPNOP(name) \
324 	__FOP_FUNC(name) \
325 	__FOP_RET(name)
326 
327 #define FOPNOP() \
328 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
329 
330 #define FOP1E(op,  dst) \
331 	__FOP_FUNC(#op "_" #dst) \
332 	"10: " #op " %" #dst " \n\t" \
333 	__FOP_RET(#op "_" #dst)
334 
335 #define FOP1EEX(op,  dst) \
336 	FOP1E(op, dst) _ASM_EXTABLE_TYPE_REG(10b, 11b, EX_TYPE_ZERO_REG, %%esi)
337 
338 #define FASTOP1(op) \
339 	FOP_START(op) \
340 	FOP1E(op##b, al) \
341 	FOP1E(op##w, ax) \
342 	FOP1E(op##l, eax) \
343 	ON64(FOP1E(op##q, rax))	\
344 	FOP_END
345 
346 /* 1-operand, using src2 (for MUL/DIV r/m) */
347 #define FASTOP1SRC2(op, name) \
348 	FOP_START(name) \
349 	FOP1E(op, cl) \
350 	FOP1E(op, cx) \
351 	FOP1E(op, ecx) \
352 	ON64(FOP1E(op, rcx)) \
353 	FOP_END
354 
355 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
356 #define FASTOP1SRC2EX(op, name) \
357 	FOP_START(name) \
358 	FOP1EEX(op, cl) \
359 	FOP1EEX(op, cx) \
360 	FOP1EEX(op, ecx) \
361 	ON64(FOP1EEX(op, rcx)) \
362 	FOP_END
363 
364 #define FOP2E(op,  dst, src)	   \
365 	__FOP_FUNC(#op "_" #dst "_" #src) \
366 	#op " %" #src ", %" #dst " \n\t" \
367 	__FOP_RET(#op "_" #dst "_" #src)
368 
369 #define FASTOP2(op) \
370 	FOP_START(op) \
371 	FOP2E(op##b, al, dl) \
372 	FOP2E(op##w, ax, dx) \
373 	FOP2E(op##l, eax, edx) \
374 	ON64(FOP2E(op##q, rax, rdx)) \
375 	FOP_END
376 
377 /* 2 operand, word only */
378 #define FASTOP2W(op) \
379 	FOP_START(op) \
380 	FOPNOP() \
381 	FOP2E(op##w, ax, dx) \
382 	FOP2E(op##l, eax, edx) \
383 	ON64(FOP2E(op##q, rax, rdx)) \
384 	FOP_END
385 
386 /* 2 operand, src is CL */
387 #define FASTOP2CL(op) \
388 	FOP_START(op) \
389 	FOP2E(op##b, al, cl) \
390 	FOP2E(op##w, ax, cl) \
391 	FOP2E(op##l, eax, cl) \
392 	ON64(FOP2E(op##q, rax, cl)) \
393 	FOP_END
394 
395 /* 2 operand, src and dest are reversed */
396 #define FASTOP2R(op, name) \
397 	FOP_START(name) \
398 	FOP2E(op##b, dl, al) \
399 	FOP2E(op##w, dx, ax) \
400 	FOP2E(op##l, edx, eax) \
401 	ON64(FOP2E(op##q, rdx, rax)) \
402 	FOP_END
403 
404 #define FOP3E(op,  dst, src, src2) \
405 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
406 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
407 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
408 
409 /* 3-operand, word-only, src2=cl */
410 #define FASTOP3WCL(op) \
411 	FOP_START(op) \
412 	FOPNOP() \
413 	FOP3E(op##w, ax, dx, cl) \
414 	FOP3E(op##l, eax, edx, cl) \
415 	ON64(FOP3E(op##q, rax, rdx, cl)) \
416 	FOP_END
417 
418 /* Special case for SETcc - 1 instruction per cc */
419 #define FOP_SETCC(op) \
420 	FOP_FUNC(op) \
421 	#op " %al \n\t" \
422 	FOP_RET(op)
423 
424 FOP_START(setcc)
425 FOP_SETCC(seto)
426 FOP_SETCC(setno)
427 FOP_SETCC(setc)
428 FOP_SETCC(setnc)
429 FOP_SETCC(setz)
430 FOP_SETCC(setnz)
431 FOP_SETCC(setbe)
432 FOP_SETCC(setnbe)
433 FOP_SETCC(sets)
434 FOP_SETCC(setns)
435 FOP_SETCC(setp)
436 FOP_SETCC(setnp)
437 FOP_SETCC(setl)
438 FOP_SETCC(setnl)
439 FOP_SETCC(setle)
440 FOP_SETCC(setnle)
441 FOP_END;
442 
443 FOP_START(salc)
444 FOP_FUNC(salc)
445 "pushf; sbb %al, %al; popf \n\t"
446 FOP_RET(salc)
447 FOP_END;
448 
449 /*
450  * XXX: inoutclob user must know where the argument is being expanded.
451  *      Using asm goto would allow us to remove _fault.
452  */
453 #define asm_safe(insn, inoutclob...) \
454 ({ \
455 	int _fault = 0; \
456  \
457 	asm volatile("1:" insn "\n" \
458 	             "2:\n" \
459 		     _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_ONE_REG, %[_fault]) \
460 	             : [_fault] "+r"(_fault) inoutclob ); \
461  \
462 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
463 })
464 
465 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
466 				    enum x86_intercept intercept,
467 				    enum x86_intercept_stage stage)
468 {
469 	struct x86_instruction_info info = {
470 		.intercept  = intercept,
471 		.rep_prefix = ctxt->rep_prefix,
472 		.modrm_mod  = ctxt->modrm_mod,
473 		.modrm_reg  = ctxt->modrm_reg,
474 		.modrm_rm   = ctxt->modrm_rm,
475 		.src_val    = ctxt->src.val64,
476 		.dst_val    = ctxt->dst.val64,
477 		.src_bytes  = ctxt->src.bytes,
478 		.dst_bytes  = ctxt->dst.bytes,
479 		.ad_bytes   = ctxt->ad_bytes,
480 		.next_rip   = ctxt->eip,
481 	};
482 
483 	return ctxt->ops->intercept(ctxt, &info, stage);
484 }
485 
486 static void assign_masked(ulong *dest, ulong src, ulong mask)
487 {
488 	*dest = (*dest & ~mask) | (src & mask);
489 }
490 
491 static void assign_register(unsigned long *reg, u64 val, int bytes)
492 {
493 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
494 	switch (bytes) {
495 	case 1:
496 		*(u8 *)reg = (u8)val;
497 		break;
498 	case 2:
499 		*(u16 *)reg = (u16)val;
500 		break;
501 	case 4:
502 		*reg = (u32)val;
503 		break;	/* 64b: zero-extend */
504 	case 8:
505 		*reg = val;
506 		break;
507 	}
508 }
509 
510 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
511 {
512 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
513 }
514 
515 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
516 {
517 	u16 sel;
518 	struct desc_struct ss;
519 
520 	if (ctxt->mode == X86EMUL_MODE_PROT64)
521 		return ~0UL;
522 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
523 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
524 }
525 
526 static int stack_size(struct x86_emulate_ctxt *ctxt)
527 {
528 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
529 }
530 
531 /* Access/update address held in a register, based on addressing mode. */
532 static inline unsigned long
533 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
534 {
535 	if (ctxt->ad_bytes == sizeof(unsigned long))
536 		return reg;
537 	else
538 		return reg & ad_mask(ctxt);
539 }
540 
541 static inline unsigned long
542 register_address(struct x86_emulate_ctxt *ctxt, int reg)
543 {
544 	return address_mask(ctxt, reg_read(ctxt, reg));
545 }
546 
547 static void masked_increment(ulong *reg, ulong mask, int inc)
548 {
549 	assign_masked(reg, *reg + inc, mask);
550 }
551 
552 static inline void
553 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
554 {
555 	ulong *preg = reg_rmw(ctxt, reg);
556 
557 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
558 }
559 
560 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
561 {
562 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
563 }
564 
565 static u32 desc_limit_scaled(struct desc_struct *desc)
566 {
567 	u32 limit = get_desc_limit(desc);
568 
569 	return desc->g ? (limit << 12) | 0xfff : limit;
570 }
571 
572 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
573 {
574 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
575 		return 0;
576 
577 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
578 }
579 
580 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
581 			     u32 error, bool valid)
582 {
583 	if (KVM_EMULATOR_BUG_ON(vec > 0x1f, ctxt))
584 		return X86EMUL_UNHANDLEABLE;
585 
586 	ctxt->exception.vector = vec;
587 	ctxt->exception.error_code = error;
588 	ctxt->exception.error_code_valid = valid;
589 	return X86EMUL_PROPAGATE_FAULT;
590 }
591 
592 static int emulate_db(struct x86_emulate_ctxt *ctxt)
593 {
594 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
595 }
596 
597 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
598 {
599 	return emulate_exception(ctxt, GP_VECTOR, err, true);
600 }
601 
602 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
603 {
604 	return emulate_exception(ctxt, SS_VECTOR, err, true);
605 }
606 
607 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
608 {
609 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
610 }
611 
612 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
613 {
614 	return emulate_exception(ctxt, TS_VECTOR, err, true);
615 }
616 
617 static int emulate_de(struct x86_emulate_ctxt *ctxt)
618 {
619 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
620 }
621 
622 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
623 {
624 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
625 }
626 
627 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
628 {
629 	u16 selector;
630 	struct desc_struct desc;
631 
632 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
633 	return selector;
634 }
635 
636 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
637 				 unsigned seg)
638 {
639 	u16 dummy;
640 	u32 base3;
641 	struct desc_struct desc;
642 
643 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
644 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
645 }
646 
647 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
648 {
649 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
650 }
651 
652 static inline bool emul_is_noncanonical_address(u64 la,
653 						struct x86_emulate_ctxt *ctxt)
654 {
655 	return !__is_canonical_address(la, ctxt_virt_addr_bits(ctxt));
656 }
657 
658 /*
659  * x86 defines three classes of vector instructions: explicitly
660  * aligned, explicitly unaligned, and the rest, which change behaviour
661  * depending on whether they're AVX encoded or not.
662  *
663  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
664  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
665  * 512 bytes of data must be aligned to a 16 byte boundary.
666  */
667 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
668 {
669 	u64 alignment = ctxt->d & AlignMask;
670 
671 	if (likely(size < 16))
672 		return 1;
673 
674 	switch (alignment) {
675 	case Unaligned:
676 	case Avx:
677 		return 1;
678 	case Aligned16:
679 		return 16;
680 	case Aligned:
681 	default:
682 		return size;
683 	}
684 }
685 
686 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
687 				       struct segmented_address addr,
688 				       unsigned *max_size, unsigned size,
689 				       bool write, bool fetch,
690 				       enum x86emul_mode mode, ulong *linear)
691 {
692 	struct desc_struct desc;
693 	bool usable;
694 	ulong la;
695 	u32 lim;
696 	u16 sel;
697 	u8  va_bits;
698 
699 	la = seg_base(ctxt, addr.seg) + addr.ea;
700 	*max_size = 0;
701 	switch (mode) {
702 	case X86EMUL_MODE_PROT64:
703 		*linear = la;
704 		va_bits = ctxt_virt_addr_bits(ctxt);
705 		if (!__is_canonical_address(la, va_bits))
706 			goto bad;
707 
708 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
709 		if (size > *max_size)
710 			goto bad;
711 		break;
712 	default:
713 		*linear = la = (u32)la;
714 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
715 						addr.seg);
716 		if (!usable)
717 			goto bad;
718 		/* code segment in protected mode or read-only data segment */
719 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
720 					|| !(desc.type & 2)) && write)
721 			goto bad;
722 		/* unreadable code segment */
723 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
724 			goto bad;
725 		lim = desc_limit_scaled(&desc);
726 		if (!(desc.type & 8) && (desc.type & 4)) {
727 			/* expand-down segment */
728 			if (addr.ea <= lim)
729 				goto bad;
730 			lim = desc.d ? 0xffffffff : 0xffff;
731 		}
732 		if (addr.ea > lim)
733 			goto bad;
734 		if (lim == 0xffffffff)
735 			*max_size = ~0u;
736 		else {
737 			*max_size = (u64)lim + 1 - addr.ea;
738 			if (size > *max_size)
739 				goto bad;
740 		}
741 		break;
742 	}
743 	if (la & (insn_alignment(ctxt, size) - 1))
744 		return emulate_gp(ctxt, 0);
745 	return X86EMUL_CONTINUE;
746 bad:
747 	if (addr.seg == VCPU_SREG_SS)
748 		return emulate_ss(ctxt, 0);
749 	else
750 		return emulate_gp(ctxt, 0);
751 }
752 
753 static int linearize(struct x86_emulate_ctxt *ctxt,
754 		     struct segmented_address addr,
755 		     unsigned size, bool write,
756 		     ulong *linear)
757 {
758 	unsigned max_size;
759 	return __linearize(ctxt, addr, &max_size, size, write, false,
760 			   ctxt->mode, linear);
761 }
762 
763 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst)
764 {
765 	ulong linear;
766 	int rc;
767 	unsigned max_size;
768 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
769 					   .ea = dst };
770 
771 	if (ctxt->op_bytes != sizeof(unsigned long))
772 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
773 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, ctxt->mode, &linear);
774 	if (rc == X86EMUL_CONTINUE)
775 		ctxt->_eip = addr.ea;
776 	return rc;
777 }
778 
779 static inline int emulator_recalc_and_set_mode(struct x86_emulate_ctxt *ctxt)
780 {
781 	u64 efer;
782 	struct desc_struct cs;
783 	u16 selector;
784 	u32 base3;
785 
786 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
787 
788 	if (!(ctxt->ops->get_cr(ctxt, 0) & X86_CR0_PE)) {
789 		/* Real mode. cpu must not have long mode active */
790 		if (efer & EFER_LMA)
791 			return X86EMUL_UNHANDLEABLE;
792 		ctxt->mode = X86EMUL_MODE_REAL;
793 		return X86EMUL_CONTINUE;
794 	}
795 
796 	if (ctxt->eflags & X86_EFLAGS_VM) {
797 		/* Protected/VM86 mode. cpu must not have long mode active */
798 		if (efer & EFER_LMA)
799 			return X86EMUL_UNHANDLEABLE;
800 		ctxt->mode = X86EMUL_MODE_VM86;
801 		return X86EMUL_CONTINUE;
802 	}
803 
804 	if (!ctxt->ops->get_segment(ctxt, &selector, &cs, &base3, VCPU_SREG_CS))
805 		return X86EMUL_UNHANDLEABLE;
806 
807 	if (efer & EFER_LMA) {
808 		if (cs.l) {
809 			/* Proper long mode */
810 			ctxt->mode = X86EMUL_MODE_PROT64;
811 		} else if (cs.d) {
812 			/* 32 bit compatibility mode*/
813 			ctxt->mode = X86EMUL_MODE_PROT32;
814 		} else {
815 			ctxt->mode = X86EMUL_MODE_PROT16;
816 		}
817 	} else {
818 		/* Legacy 32 bit / 16 bit mode */
819 		ctxt->mode = cs.d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
820 	}
821 
822 	return X86EMUL_CONTINUE;
823 }
824 
825 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
826 {
827 	return assign_eip(ctxt, dst);
828 }
829 
830 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst)
831 {
832 	int rc = emulator_recalc_and_set_mode(ctxt);
833 
834 	if (rc != X86EMUL_CONTINUE)
835 		return rc;
836 
837 	return assign_eip(ctxt, dst);
838 }
839 
840 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
841 {
842 	return assign_eip_near(ctxt, ctxt->_eip + rel);
843 }
844 
845 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear,
846 			      void *data, unsigned size)
847 {
848 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true);
849 }
850 
851 static int linear_write_system(struct x86_emulate_ctxt *ctxt,
852 			       ulong linear, void *data,
853 			       unsigned int size)
854 {
855 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true);
856 }
857 
858 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
859 			      struct segmented_address addr,
860 			      void *data,
861 			      unsigned size)
862 {
863 	int rc;
864 	ulong linear;
865 
866 	rc = linearize(ctxt, addr, size, false, &linear);
867 	if (rc != X86EMUL_CONTINUE)
868 		return rc;
869 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false);
870 }
871 
872 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
873 			       struct segmented_address addr,
874 			       void *data,
875 			       unsigned int size)
876 {
877 	int rc;
878 	ulong linear;
879 
880 	rc = linearize(ctxt, addr, size, true, &linear);
881 	if (rc != X86EMUL_CONTINUE)
882 		return rc;
883 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false);
884 }
885 
886 /*
887  * Prefetch the remaining bytes of the instruction without crossing page
888  * boundary if they are not in fetch_cache yet.
889  */
890 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
891 {
892 	int rc;
893 	unsigned size, max_size;
894 	unsigned long linear;
895 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
896 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
897 					   .ea = ctxt->eip + cur_size };
898 
899 	/*
900 	 * We do not know exactly how many bytes will be needed, and
901 	 * __linearize is expensive, so fetch as much as possible.  We
902 	 * just have to avoid going beyond the 15 byte limit, the end
903 	 * of the segment, or the end of the page.
904 	 *
905 	 * __linearize is called with size 0 so that it does not do any
906 	 * boundary check itself.  Instead, we use max_size to check
907 	 * against op_size.
908 	 */
909 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
910 			 &linear);
911 	if (unlikely(rc != X86EMUL_CONTINUE))
912 		return rc;
913 
914 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
915 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
916 
917 	/*
918 	 * One instruction can only straddle two pages,
919 	 * and one has been loaded at the beginning of
920 	 * x86_decode_insn.  So, if not enough bytes
921 	 * still, we must have hit the 15-byte boundary.
922 	 */
923 	if (unlikely(size < op_size))
924 		return emulate_gp(ctxt, 0);
925 
926 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
927 			      size, &ctxt->exception);
928 	if (unlikely(rc != X86EMUL_CONTINUE))
929 		return rc;
930 	ctxt->fetch.end += size;
931 	return X86EMUL_CONTINUE;
932 }
933 
934 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
935 					       unsigned size)
936 {
937 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
938 
939 	if (unlikely(done_size < size))
940 		return __do_insn_fetch_bytes(ctxt, size - done_size);
941 	else
942 		return X86EMUL_CONTINUE;
943 }
944 
945 /* Fetch next part of the instruction being emulated. */
946 #define insn_fetch(_type, _ctxt)					\
947 ({	_type _x;							\
948 									\
949 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
950 	if (rc != X86EMUL_CONTINUE)					\
951 		goto done;						\
952 	ctxt->_eip += sizeof(_type);					\
953 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
954 	ctxt->fetch.ptr += sizeof(_type);				\
955 	_x;								\
956 })
957 
958 #define insn_fetch_arr(_arr, _size, _ctxt)				\
959 ({									\
960 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
961 	if (rc != X86EMUL_CONTINUE)					\
962 		goto done;						\
963 	ctxt->_eip += (_size);						\
964 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
965 	ctxt->fetch.ptr += (_size);					\
966 })
967 
968 /*
969  * Given the 'reg' portion of a ModRM byte, and a register block, return a
970  * pointer into the block that addresses the relevant register.
971  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
972  */
973 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
974 			     int byteop)
975 {
976 	void *p;
977 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
978 
979 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
980 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
981 	else
982 		p = reg_rmw(ctxt, modrm_reg);
983 	return p;
984 }
985 
986 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
987 			   struct segmented_address addr,
988 			   u16 *size, unsigned long *address, int op_bytes)
989 {
990 	int rc;
991 
992 	if (op_bytes == 2)
993 		op_bytes = 3;
994 	*address = 0;
995 	rc = segmented_read_std(ctxt, addr, size, 2);
996 	if (rc != X86EMUL_CONTINUE)
997 		return rc;
998 	addr.ea += 2;
999 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
1000 	return rc;
1001 }
1002 
1003 FASTOP2(add);
1004 FASTOP2(or);
1005 FASTOP2(adc);
1006 FASTOP2(sbb);
1007 FASTOP2(and);
1008 FASTOP2(sub);
1009 FASTOP2(xor);
1010 FASTOP2(cmp);
1011 FASTOP2(test);
1012 
1013 FASTOP1SRC2(mul, mul_ex);
1014 FASTOP1SRC2(imul, imul_ex);
1015 FASTOP1SRC2EX(div, div_ex);
1016 FASTOP1SRC2EX(idiv, idiv_ex);
1017 
1018 FASTOP3WCL(shld);
1019 FASTOP3WCL(shrd);
1020 
1021 FASTOP2W(imul);
1022 
1023 FASTOP1(not);
1024 FASTOP1(neg);
1025 FASTOP1(inc);
1026 FASTOP1(dec);
1027 
1028 FASTOP2CL(rol);
1029 FASTOP2CL(ror);
1030 FASTOP2CL(rcl);
1031 FASTOP2CL(rcr);
1032 FASTOP2CL(shl);
1033 FASTOP2CL(shr);
1034 FASTOP2CL(sar);
1035 
1036 FASTOP2W(bsf);
1037 FASTOP2W(bsr);
1038 FASTOP2W(bt);
1039 FASTOP2W(bts);
1040 FASTOP2W(btr);
1041 FASTOP2W(btc);
1042 
1043 FASTOP2(xadd);
1044 
1045 FASTOP2R(cmp, cmp_r);
1046 
1047 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1048 {
1049 	/* If src is zero, do not writeback, but update flags */
1050 	if (ctxt->src.val == 0)
1051 		ctxt->dst.type = OP_NONE;
1052 	return fastop(ctxt, em_bsf);
1053 }
1054 
1055 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1056 {
1057 	/* If src is zero, do not writeback, but update flags */
1058 	if (ctxt->src.val == 0)
1059 		ctxt->dst.type = OP_NONE;
1060 	return fastop(ctxt, em_bsr);
1061 }
1062 
1063 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1064 {
1065 	u8 rc;
1066 	void (*fop)(void) = (void *)em_setcc + FASTOP_SIZE * (condition & 0xf);
1067 
1068 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1069 	asm("push %[flags]; popf; " CALL_NOSPEC
1070 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1071 	return rc;
1072 }
1073 
1074 static void fetch_register_operand(struct operand *op)
1075 {
1076 	switch (op->bytes) {
1077 	case 1:
1078 		op->val = *(u8 *)op->addr.reg;
1079 		break;
1080 	case 2:
1081 		op->val = *(u16 *)op->addr.reg;
1082 		break;
1083 	case 4:
1084 		op->val = *(u32 *)op->addr.reg;
1085 		break;
1086 	case 8:
1087 		op->val = *(u64 *)op->addr.reg;
1088 		break;
1089 	}
1090 }
1091 
1092 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1093 {
1094 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1095 		return emulate_nm(ctxt);
1096 
1097 	kvm_fpu_get();
1098 	asm volatile("fninit");
1099 	kvm_fpu_put();
1100 	return X86EMUL_CONTINUE;
1101 }
1102 
1103 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1104 {
1105 	u16 fcw;
1106 
1107 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1108 		return emulate_nm(ctxt);
1109 
1110 	kvm_fpu_get();
1111 	asm volatile("fnstcw %0": "+m"(fcw));
1112 	kvm_fpu_put();
1113 
1114 	ctxt->dst.val = fcw;
1115 
1116 	return X86EMUL_CONTINUE;
1117 }
1118 
1119 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1120 {
1121 	u16 fsw;
1122 
1123 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1124 		return emulate_nm(ctxt);
1125 
1126 	kvm_fpu_get();
1127 	asm volatile("fnstsw %0": "+m"(fsw));
1128 	kvm_fpu_put();
1129 
1130 	ctxt->dst.val = fsw;
1131 
1132 	return X86EMUL_CONTINUE;
1133 }
1134 
1135 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1136 				    struct operand *op)
1137 {
1138 	unsigned int reg;
1139 
1140 	if (ctxt->d & ModRM)
1141 		reg = ctxt->modrm_reg;
1142 	else
1143 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1144 
1145 	if (ctxt->d & Sse) {
1146 		op->type = OP_XMM;
1147 		op->bytes = 16;
1148 		op->addr.xmm = reg;
1149 		kvm_read_sse_reg(reg, &op->vec_val);
1150 		return;
1151 	}
1152 	if (ctxt->d & Mmx) {
1153 		reg &= 7;
1154 		op->type = OP_MM;
1155 		op->bytes = 8;
1156 		op->addr.mm = reg;
1157 		return;
1158 	}
1159 
1160 	op->type = OP_REG;
1161 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1162 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1163 
1164 	fetch_register_operand(op);
1165 	op->orig_val = op->val;
1166 }
1167 
1168 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1169 {
1170 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1171 		ctxt->modrm_seg = VCPU_SREG_SS;
1172 }
1173 
1174 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1175 			struct operand *op)
1176 {
1177 	u8 sib;
1178 	int index_reg, base_reg, scale;
1179 	int rc = X86EMUL_CONTINUE;
1180 	ulong modrm_ea = 0;
1181 
1182 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1183 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1184 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1185 
1186 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1187 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1188 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1189 	ctxt->modrm_seg = VCPU_SREG_DS;
1190 
1191 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1192 		op->type = OP_REG;
1193 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1194 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1195 				ctxt->d & ByteOp);
1196 		if (ctxt->d & Sse) {
1197 			op->type = OP_XMM;
1198 			op->bytes = 16;
1199 			op->addr.xmm = ctxt->modrm_rm;
1200 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1201 			return rc;
1202 		}
1203 		if (ctxt->d & Mmx) {
1204 			op->type = OP_MM;
1205 			op->bytes = 8;
1206 			op->addr.mm = ctxt->modrm_rm & 7;
1207 			return rc;
1208 		}
1209 		fetch_register_operand(op);
1210 		return rc;
1211 	}
1212 
1213 	op->type = OP_MEM;
1214 
1215 	if (ctxt->ad_bytes == 2) {
1216 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1217 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1218 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1219 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1220 
1221 		/* 16-bit ModR/M decode. */
1222 		switch (ctxt->modrm_mod) {
1223 		case 0:
1224 			if (ctxt->modrm_rm == 6)
1225 				modrm_ea += insn_fetch(u16, ctxt);
1226 			break;
1227 		case 1:
1228 			modrm_ea += insn_fetch(s8, ctxt);
1229 			break;
1230 		case 2:
1231 			modrm_ea += insn_fetch(u16, ctxt);
1232 			break;
1233 		}
1234 		switch (ctxt->modrm_rm) {
1235 		case 0:
1236 			modrm_ea += bx + si;
1237 			break;
1238 		case 1:
1239 			modrm_ea += bx + di;
1240 			break;
1241 		case 2:
1242 			modrm_ea += bp + si;
1243 			break;
1244 		case 3:
1245 			modrm_ea += bp + di;
1246 			break;
1247 		case 4:
1248 			modrm_ea += si;
1249 			break;
1250 		case 5:
1251 			modrm_ea += di;
1252 			break;
1253 		case 6:
1254 			if (ctxt->modrm_mod != 0)
1255 				modrm_ea += bp;
1256 			break;
1257 		case 7:
1258 			modrm_ea += bx;
1259 			break;
1260 		}
1261 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1262 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1263 			ctxt->modrm_seg = VCPU_SREG_SS;
1264 		modrm_ea = (u16)modrm_ea;
1265 	} else {
1266 		/* 32/64-bit ModR/M decode. */
1267 		if ((ctxt->modrm_rm & 7) == 4) {
1268 			sib = insn_fetch(u8, ctxt);
1269 			index_reg |= (sib >> 3) & 7;
1270 			base_reg |= sib & 7;
1271 			scale = sib >> 6;
1272 
1273 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1274 				modrm_ea += insn_fetch(s32, ctxt);
1275 			else {
1276 				modrm_ea += reg_read(ctxt, base_reg);
1277 				adjust_modrm_seg(ctxt, base_reg);
1278 				/* Increment ESP on POP [ESP] */
1279 				if ((ctxt->d & IncSP) &&
1280 				    base_reg == VCPU_REGS_RSP)
1281 					modrm_ea += ctxt->op_bytes;
1282 			}
1283 			if (index_reg != 4)
1284 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1285 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1286 			modrm_ea += insn_fetch(s32, ctxt);
1287 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1288 				ctxt->rip_relative = 1;
1289 		} else {
1290 			base_reg = ctxt->modrm_rm;
1291 			modrm_ea += reg_read(ctxt, base_reg);
1292 			adjust_modrm_seg(ctxt, base_reg);
1293 		}
1294 		switch (ctxt->modrm_mod) {
1295 		case 1:
1296 			modrm_ea += insn_fetch(s8, ctxt);
1297 			break;
1298 		case 2:
1299 			modrm_ea += insn_fetch(s32, ctxt);
1300 			break;
1301 		}
1302 	}
1303 	op->addr.mem.ea = modrm_ea;
1304 	if (ctxt->ad_bytes != 8)
1305 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1306 
1307 done:
1308 	return rc;
1309 }
1310 
1311 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1312 		      struct operand *op)
1313 {
1314 	int rc = X86EMUL_CONTINUE;
1315 
1316 	op->type = OP_MEM;
1317 	switch (ctxt->ad_bytes) {
1318 	case 2:
1319 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1320 		break;
1321 	case 4:
1322 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1323 		break;
1324 	case 8:
1325 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1326 		break;
1327 	}
1328 done:
1329 	return rc;
1330 }
1331 
1332 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1333 {
1334 	long sv = 0, mask;
1335 
1336 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1337 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1338 
1339 		if (ctxt->src.bytes == 2)
1340 			sv = (s16)ctxt->src.val & (s16)mask;
1341 		else if (ctxt->src.bytes == 4)
1342 			sv = (s32)ctxt->src.val & (s32)mask;
1343 		else
1344 			sv = (s64)ctxt->src.val & (s64)mask;
1345 
1346 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1347 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1348 	}
1349 
1350 	/* only subword offset */
1351 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1352 }
1353 
1354 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1355 			 unsigned long addr, void *dest, unsigned size)
1356 {
1357 	int rc;
1358 	struct read_cache *mc = &ctxt->mem_read;
1359 
1360 	if (mc->pos < mc->end)
1361 		goto read_cached;
1362 
1363 	if (KVM_EMULATOR_BUG_ON((mc->end + size) >= sizeof(mc->data), ctxt))
1364 		return X86EMUL_UNHANDLEABLE;
1365 
1366 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1367 				      &ctxt->exception);
1368 	if (rc != X86EMUL_CONTINUE)
1369 		return rc;
1370 
1371 	mc->end += size;
1372 
1373 read_cached:
1374 	memcpy(dest, mc->data + mc->pos, size);
1375 	mc->pos += size;
1376 	return X86EMUL_CONTINUE;
1377 }
1378 
1379 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1380 			  struct segmented_address addr,
1381 			  void *data,
1382 			  unsigned size)
1383 {
1384 	int rc;
1385 	ulong linear;
1386 
1387 	rc = linearize(ctxt, addr, size, false, &linear);
1388 	if (rc != X86EMUL_CONTINUE)
1389 		return rc;
1390 	return read_emulated(ctxt, linear, data, size);
1391 }
1392 
1393 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1394 			   struct segmented_address addr,
1395 			   const void *data,
1396 			   unsigned size)
1397 {
1398 	int rc;
1399 	ulong linear;
1400 
1401 	rc = linearize(ctxt, addr, size, true, &linear);
1402 	if (rc != X86EMUL_CONTINUE)
1403 		return rc;
1404 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1405 					 &ctxt->exception);
1406 }
1407 
1408 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1409 			     struct segmented_address addr,
1410 			     const void *orig_data, const void *data,
1411 			     unsigned size)
1412 {
1413 	int rc;
1414 	ulong linear;
1415 
1416 	rc = linearize(ctxt, addr, size, true, &linear);
1417 	if (rc != X86EMUL_CONTINUE)
1418 		return rc;
1419 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1420 					   size, &ctxt->exception);
1421 }
1422 
1423 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1424 			   unsigned int size, unsigned short port,
1425 			   void *dest)
1426 {
1427 	struct read_cache *rc = &ctxt->io_read;
1428 
1429 	if (rc->pos == rc->end) { /* refill pio read ahead */
1430 		unsigned int in_page, n;
1431 		unsigned int count = ctxt->rep_prefix ?
1432 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1433 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1434 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1435 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1436 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1437 		if (n == 0)
1438 			n = 1;
1439 		rc->pos = rc->end = 0;
1440 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1441 			return 0;
1442 		rc->end = n * size;
1443 	}
1444 
1445 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1446 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1447 		ctxt->dst.data = rc->data + rc->pos;
1448 		ctxt->dst.type = OP_MEM_STR;
1449 		ctxt->dst.count = (rc->end - rc->pos) / size;
1450 		rc->pos = rc->end;
1451 	} else {
1452 		memcpy(dest, rc->data + rc->pos, size);
1453 		rc->pos += size;
1454 	}
1455 	return 1;
1456 }
1457 
1458 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1459 				     u16 index, struct desc_struct *desc)
1460 {
1461 	struct desc_ptr dt;
1462 	ulong addr;
1463 
1464 	ctxt->ops->get_idt(ctxt, &dt);
1465 
1466 	if (dt.size < index * 8 + 7)
1467 		return emulate_gp(ctxt, index << 3 | 0x2);
1468 
1469 	addr = dt.address + index * 8;
1470 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1471 }
1472 
1473 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1474 				     u16 selector, struct desc_ptr *dt)
1475 {
1476 	const struct x86_emulate_ops *ops = ctxt->ops;
1477 	u32 base3 = 0;
1478 
1479 	if (selector & 1 << 2) {
1480 		struct desc_struct desc;
1481 		u16 sel;
1482 
1483 		memset(dt, 0, sizeof(*dt));
1484 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1485 				      VCPU_SREG_LDTR))
1486 			return;
1487 
1488 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1489 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1490 	} else
1491 		ops->get_gdt(ctxt, dt);
1492 }
1493 
1494 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1495 			      u16 selector, ulong *desc_addr_p)
1496 {
1497 	struct desc_ptr dt;
1498 	u16 index = selector >> 3;
1499 	ulong addr;
1500 
1501 	get_descriptor_table_ptr(ctxt, selector, &dt);
1502 
1503 	if (dt.size < index * 8 + 7)
1504 		return emulate_gp(ctxt, selector & 0xfffc);
1505 
1506 	addr = dt.address + index * 8;
1507 
1508 #ifdef CONFIG_X86_64
1509 	if (addr >> 32 != 0) {
1510 		u64 efer = 0;
1511 
1512 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1513 		if (!(efer & EFER_LMA))
1514 			addr &= (u32)-1;
1515 	}
1516 #endif
1517 
1518 	*desc_addr_p = addr;
1519 	return X86EMUL_CONTINUE;
1520 }
1521 
1522 /* allowed just for 8 bytes segments */
1523 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1524 				   u16 selector, struct desc_struct *desc,
1525 				   ulong *desc_addr_p)
1526 {
1527 	int rc;
1528 
1529 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1530 	if (rc != X86EMUL_CONTINUE)
1531 		return rc;
1532 
1533 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1534 }
1535 
1536 /* allowed just for 8 bytes segments */
1537 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1538 				    u16 selector, struct desc_struct *desc)
1539 {
1540 	int rc;
1541 	ulong addr;
1542 
1543 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1544 	if (rc != X86EMUL_CONTINUE)
1545 		return rc;
1546 
1547 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1548 }
1549 
1550 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1551 				     u16 selector, int seg, u8 cpl,
1552 				     enum x86_transfer_type transfer,
1553 				     struct desc_struct *desc)
1554 {
1555 	struct desc_struct seg_desc, old_desc;
1556 	u8 dpl, rpl;
1557 	unsigned err_vec = GP_VECTOR;
1558 	u32 err_code = 0;
1559 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1560 	ulong desc_addr;
1561 	int ret;
1562 	u16 dummy;
1563 	u32 base3 = 0;
1564 
1565 	memset(&seg_desc, 0, sizeof(seg_desc));
1566 
1567 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1568 		/* set real mode segment descriptor (keep limit etc. for
1569 		 * unreal mode) */
1570 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1571 		set_desc_base(&seg_desc, selector << 4);
1572 		goto load;
1573 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1574 		/* VM86 needs a clean new segment descriptor */
1575 		set_desc_base(&seg_desc, selector << 4);
1576 		set_desc_limit(&seg_desc, 0xffff);
1577 		seg_desc.type = 3;
1578 		seg_desc.p = 1;
1579 		seg_desc.s = 1;
1580 		seg_desc.dpl = 3;
1581 		goto load;
1582 	}
1583 
1584 	rpl = selector & 3;
1585 
1586 	/* TR should be in GDT only */
1587 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1588 		goto exception;
1589 
1590 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1591 	if (null_selector) {
1592 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1593 			goto exception;
1594 
1595 		if (seg == VCPU_SREG_SS) {
1596 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1597 				goto exception;
1598 
1599 			/*
1600 			 * ctxt->ops->set_segment expects the CPL to be in
1601 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1602 			 */
1603 			seg_desc.type = 3;
1604 			seg_desc.p = 1;
1605 			seg_desc.s = 1;
1606 			seg_desc.dpl = cpl;
1607 			seg_desc.d = 1;
1608 			seg_desc.g = 1;
1609 		}
1610 
1611 		/* Skip all following checks */
1612 		goto load;
1613 	}
1614 
1615 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1616 	if (ret != X86EMUL_CONTINUE)
1617 		return ret;
1618 
1619 	err_code = selector & 0xfffc;
1620 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1621 							   GP_VECTOR;
1622 
1623 	/* can't load system descriptor into segment selector */
1624 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1625 		if (transfer == X86_TRANSFER_CALL_JMP)
1626 			return X86EMUL_UNHANDLEABLE;
1627 		goto exception;
1628 	}
1629 
1630 	dpl = seg_desc.dpl;
1631 
1632 	switch (seg) {
1633 	case VCPU_SREG_SS:
1634 		/*
1635 		 * segment is not a writable data segment or segment
1636 		 * selector's RPL != CPL or segment selector's RPL != CPL
1637 		 */
1638 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1639 			goto exception;
1640 		break;
1641 	case VCPU_SREG_CS:
1642 		if (!(seg_desc.type & 8))
1643 			goto exception;
1644 
1645 		if (transfer == X86_TRANSFER_RET) {
1646 			/* RET can never return to an inner privilege level. */
1647 			if (rpl < cpl)
1648 				goto exception;
1649 			/* Outer-privilege level return is not implemented */
1650 			if (rpl > cpl)
1651 				return X86EMUL_UNHANDLEABLE;
1652 		}
1653 		if (transfer == X86_TRANSFER_RET || transfer == X86_TRANSFER_TASK_SWITCH) {
1654 			if (seg_desc.type & 4) {
1655 				/* conforming */
1656 				if (dpl > rpl)
1657 					goto exception;
1658 			} else {
1659 				/* nonconforming */
1660 				if (dpl != rpl)
1661 					goto exception;
1662 			}
1663 		} else { /* X86_TRANSFER_CALL_JMP */
1664 			if (seg_desc.type & 4) {
1665 				/* conforming */
1666 				if (dpl > cpl)
1667 					goto exception;
1668 			} else {
1669 				/* nonconforming */
1670 				if (rpl > cpl || dpl != cpl)
1671 					goto exception;
1672 			}
1673 		}
1674 		/* in long-mode d/b must be clear if l is set */
1675 		if (seg_desc.d && seg_desc.l) {
1676 			u64 efer = 0;
1677 
1678 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1679 			if (efer & EFER_LMA)
1680 				goto exception;
1681 		}
1682 
1683 		/* CS(RPL) <- CPL */
1684 		selector = (selector & 0xfffc) | cpl;
1685 		break;
1686 	case VCPU_SREG_TR:
1687 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1688 			goto exception;
1689 		break;
1690 	case VCPU_SREG_LDTR:
1691 		if (seg_desc.s || seg_desc.type != 2)
1692 			goto exception;
1693 		break;
1694 	default: /*  DS, ES, FS, or GS */
1695 		/*
1696 		 * segment is not a data or readable code segment or
1697 		 * ((segment is a data or nonconforming code segment)
1698 		 * and (both RPL and CPL > DPL))
1699 		 */
1700 		if ((seg_desc.type & 0xa) == 0x8 ||
1701 		    (((seg_desc.type & 0xc) != 0xc) &&
1702 		     (rpl > dpl && cpl > dpl)))
1703 			goto exception;
1704 		break;
1705 	}
1706 
1707 	if (!seg_desc.p) {
1708 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1709 		goto exception;
1710 	}
1711 
1712 	if (seg_desc.s) {
1713 		/* mark segment as accessed */
1714 		if (!(seg_desc.type & 1)) {
1715 			seg_desc.type |= 1;
1716 			ret = write_segment_descriptor(ctxt, selector,
1717 						       &seg_desc);
1718 			if (ret != X86EMUL_CONTINUE)
1719 				return ret;
1720 		}
1721 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1722 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1723 		if (ret != X86EMUL_CONTINUE)
1724 			return ret;
1725 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1726 						 ((u64)base3 << 32), ctxt))
1727 			return emulate_gp(ctxt, err_code);
1728 	}
1729 
1730 	if (seg == VCPU_SREG_TR) {
1731 		old_desc = seg_desc;
1732 		seg_desc.type |= 2; /* busy */
1733 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1734 						  sizeof(seg_desc), &ctxt->exception);
1735 		if (ret != X86EMUL_CONTINUE)
1736 			return ret;
1737 	}
1738 load:
1739 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1740 	if (desc)
1741 		*desc = seg_desc;
1742 	return X86EMUL_CONTINUE;
1743 exception:
1744 	return emulate_exception(ctxt, err_vec, err_code, true);
1745 }
1746 
1747 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1748 				   u16 selector, int seg)
1749 {
1750 	u8 cpl = ctxt->ops->cpl(ctxt);
1751 
1752 	/*
1753 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1754 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1755 	 * but it's wrong).
1756 	 *
1757 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1758 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1759 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1760 	 * and only forbid it here.
1761 	 */
1762 	if (seg == VCPU_SREG_SS && selector == 3 &&
1763 	    ctxt->mode == X86EMUL_MODE_PROT64)
1764 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1765 
1766 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1767 					 X86_TRANSFER_NONE, NULL);
1768 }
1769 
1770 static void write_register_operand(struct operand *op)
1771 {
1772 	return assign_register(op->addr.reg, op->val, op->bytes);
1773 }
1774 
1775 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1776 {
1777 	switch (op->type) {
1778 	case OP_REG:
1779 		write_register_operand(op);
1780 		break;
1781 	case OP_MEM:
1782 		if (ctxt->lock_prefix)
1783 			return segmented_cmpxchg(ctxt,
1784 						 op->addr.mem,
1785 						 &op->orig_val,
1786 						 &op->val,
1787 						 op->bytes);
1788 		else
1789 			return segmented_write(ctxt,
1790 					       op->addr.mem,
1791 					       &op->val,
1792 					       op->bytes);
1793 		break;
1794 	case OP_MEM_STR:
1795 		return segmented_write(ctxt,
1796 				       op->addr.mem,
1797 				       op->data,
1798 				       op->bytes * op->count);
1799 		break;
1800 	case OP_XMM:
1801 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1802 		break;
1803 	case OP_MM:
1804 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1805 		break;
1806 	case OP_NONE:
1807 		/* no writeback */
1808 		break;
1809 	default:
1810 		break;
1811 	}
1812 	return X86EMUL_CONTINUE;
1813 }
1814 
1815 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1816 {
1817 	struct segmented_address addr;
1818 
1819 	rsp_increment(ctxt, -bytes);
1820 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1821 	addr.seg = VCPU_SREG_SS;
1822 
1823 	return segmented_write(ctxt, addr, data, bytes);
1824 }
1825 
1826 static int em_push(struct x86_emulate_ctxt *ctxt)
1827 {
1828 	/* Disable writeback. */
1829 	ctxt->dst.type = OP_NONE;
1830 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1831 }
1832 
1833 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1834 		       void *dest, int len)
1835 {
1836 	int rc;
1837 	struct segmented_address addr;
1838 
1839 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1840 	addr.seg = VCPU_SREG_SS;
1841 	rc = segmented_read(ctxt, addr, dest, len);
1842 	if (rc != X86EMUL_CONTINUE)
1843 		return rc;
1844 
1845 	rsp_increment(ctxt, len);
1846 	return rc;
1847 }
1848 
1849 static int em_pop(struct x86_emulate_ctxt *ctxt)
1850 {
1851 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1852 }
1853 
1854 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1855 			void *dest, int len)
1856 {
1857 	int rc;
1858 	unsigned long val, change_mask;
1859 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1860 	int cpl = ctxt->ops->cpl(ctxt);
1861 
1862 	rc = emulate_pop(ctxt, &val, len);
1863 	if (rc != X86EMUL_CONTINUE)
1864 		return rc;
1865 
1866 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1867 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1868 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1869 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1870 
1871 	switch(ctxt->mode) {
1872 	case X86EMUL_MODE_PROT64:
1873 	case X86EMUL_MODE_PROT32:
1874 	case X86EMUL_MODE_PROT16:
1875 		if (cpl == 0)
1876 			change_mask |= X86_EFLAGS_IOPL;
1877 		if (cpl <= iopl)
1878 			change_mask |= X86_EFLAGS_IF;
1879 		break;
1880 	case X86EMUL_MODE_VM86:
1881 		if (iopl < 3)
1882 			return emulate_gp(ctxt, 0);
1883 		change_mask |= X86_EFLAGS_IF;
1884 		break;
1885 	default: /* real mode */
1886 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1887 		break;
1888 	}
1889 
1890 	*(unsigned long *)dest =
1891 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1892 
1893 	return rc;
1894 }
1895 
1896 static int em_popf(struct x86_emulate_ctxt *ctxt)
1897 {
1898 	ctxt->dst.type = OP_REG;
1899 	ctxt->dst.addr.reg = &ctxt->eflags;
1900 	ctxt->dst.bytes = ctxt->op_bytes;
1901 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1902 }
1903 
1904 static int em_enter(struct x86_emulate_ctxt *ctxt)
1905 {
1906 	int rc;
1907 	unsigned frame_size = ctxt->src.val;
1908 	unsigned nesting_level = ctxt->src2.val & 31;
1909 	ulong rbp;
1910 
1911 	if (nesting_level)
1912 		return X86EMUL_UNHANDLEABLE;
1913 
1914 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1915 	rc = push(ctxt, &rbp, stack_size(ctxt));
1916 	if (rc != X86EMUL_CONTINUE)
1917 		return rc;
1918 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1919 		      stack_mask(ctxt));
1920 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1921 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1922 		      stack_mask(ctxt));
1923 	return X86EMUL_CONTINUE;
1924 }
1925 
1926 static int em_leave(struct x86_emulate_ctxt *ctxt)
1927 {
1928 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1929 		      stack_mask(ctxt));
1930 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1931 }
1932 
1933 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1934 {
1935 	int seg = ctxt->src2.val;
1936 
1937 	ctxt->src.val = get_segment_selector(ctxt, seg);
1938 	if (ctxt->op_bytes == 4) {
1939 		rsp_increment(ctxt, -2);
1940 		ctxt->op_bytes = 2;
1941 	}
1942 
1943 	return em_push(ctxt);
1944 }
1945 
1946 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1947 {
1948 	int seg = ctxt->src2.val;
1949 	unsigned long selector;
1950 	int rc;
1951 
1952 	rc = emulate_pop(ctxt, &selector, 2);
1953 	if (rc != X86EMUL_CONTINUE)
1954 		return rc;
1955 
1956 	if (seg == VCPU_SREG_SS)
1957 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1958 	if (ctxt->op_bytes > 2)
1959 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1960 
1961 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1962 	return rc;
1963 }
1964 
1965 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1966 {
1967 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1968 	int rc = X86EMUL_CONTINUE;
1969 	int reg = VCPU_REGS_RAX;
1970 
1971 	while (reg <= VCPU_REGS_RDI) {
1972 		(reg == VCPU_REGS_RSP) ?
1973 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1974 
1975 		rc = em_push(ctxt);
1976 		if (rc != X86EMUL_CONTINUE)
1977 			return rc;
1978 
1979 		++reg;
1980 	}
1981 
1982 	return rc;
1983 }
1984 
1985 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1986 {
1987 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1988 	return em_push(ctxt);
1989 }
1990 
1991 static int em_popa(struct x86_emulate_ctxt *ctxt)
1992 {
1993 	int rc = X86EMUL_CONTINUE;
1994 	int reg = VCPU_REGS_RDI;
1995 	u32 val;
1996 
1997 	while (reg >= VCPU_REGS_RAX) {
1998 		if (reg == VCPU_REGS_RSP) {
1999 			rsp_increment(ctxt, ctxt->op_bytes);
2000 			--reg;
2001 		}
2002 
2003 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2004 		if (rc != X86EMUL_CONTINUE)
2005 			break;
2006 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2007 		--reg;
2008 	}
2009 	return rc;
2010 }
2011 
2012 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2013 {
2014 	const struct x86_emulate_ops *ops = ctxt->ops;
2015 	int rc;
2016 	struct desc_ptr dt;
2017 	gva_t cs_addr;
2018 	gva_t eip_addr;
2019 	u16 cs, eip;
2020 
2021 	/* TODO: Add limit checks */
2022 	ctxt->src.val = ctxt->eflags;
2023 	rc = em_push(ctxt);
2024 	if (rc != X86EMUL_CONTINUE)
2025 		return rc;
2026 
2027 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2028 
2029 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2030 	rc = em_push(ctxt);
2031 	if (rc != X86EMUL_CONTINUE)
2032 		return rc;
2033 
2034 	ctxt->src.val = ctxt->_eip;
2035 	rc = em_push(ctxt);
2036 	if (rc != X86EMUL_CONTINUE)
2037 		return rc;
2038 
2039 	ops->get_idt(ctxt, &dt);
2040 
2041 	eip_addr = dt.address + (irq << 2);
2042 	cs_addr = dt.address + (irq << 2) + 2;
2043 
2044 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2045 	if (rc != X86EMUL_CONTINUE)
2046 		return rc;
2047 
2048 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2049 	if (rc != X86EMUL_CONTINUE)
2050 		return rc;
2051 
2052 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2053 	if (rc != X86EMUL_CONTINUE)
2054 		return rc;
2055 
2056 	ctxt->_eip = eip;
2057 
2058 	return rc;
2059 }
2060 
2061 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2062 {
2063 	int rc;
2064 
2065 	invalidate_registers(ctxt);
2066 	rc = __emulate_int_real(ctxt, irq);
2067 	if (rc == X86EMUL_CONTINUE)
2068 		writeback_registers(ctxt);
2069 	return rc;
2070 }
2071 
2072 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2073 {
2074 	switch(ctxt->mode) {
2075 	case X86EMUL_MODE_REAL:
2076 		return __emulate_int_real(ctxt, irq);
2077 	case X86EMUL_MODE_VM86:
2078 	case X86EMUL_MODE_PROT16:
2079 	case X86EMUL_MODE_PROT32:
2080 	case X86EMUL_MODE_PROT64:
2081 	default:
2082 		/* Protected mode interrupts unimplemented yet */
2083 		return X86EMUL_UNHANDLEABLE;
2084 	}
2085 }
2086 
2087 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2088 {
2089 	int rc = X86EMUL_CONTINUE;
2090 	unsigned long temp_eip = 0;
2091 	unsigned long temp_eflags = 0;
2092 	unsigned long cs = 0;
2093 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2094 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2095 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2096 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2097 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2098 			     X86_EFLAGS_FIXED;
2099 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2100 				  X86_EFLAGS_VIP;
2101 
2102 	/* TODO: Add stack limit check */
2103 
2104 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2105 
2106 	if (rc != X86EMUL_CONTINUE)
2107 		return rc;
2108 
2109 	if (temp_eip & ~0xffff)
2110 		return emulate_gp(ctxt, 0);
2111 
2112 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2113 
2114 	if (rc != X86EMUL_CONTINUE)
2115 		return rc;
2116 
2117 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2118 
2119 	if (rc != X86EMUL_CONTINUE)
2120 		return rc;
2121 
2122 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2123 
2124 	if (rc != X86EMUL_CONTINUE)
2125 		return rc;
2126 
2127 	ctxt->_eip = temp_eip;
2128 
2129 	if (ctxt->op_bytes == 4)
2130 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2131 	else if (ctxt->op_bytes == 2) {
2132 		ctxt->eflags &= ~0xffff;
2133 		ctxt->eflags |= temp_eflags;
2134 	}
2135 
2136 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2137 	ctxt->eflags |= X86_EFLAGS_FIXED;
2138 	ctxt->ops->set_nmi_mask(ctxt, false);
2139 
2140 	return rc;
2141 }
2142 
2143 static int em_iret(struct x86_emulate_ctxt *ctxt)
2144 {
2145 	switch(ctxt->mode) {
2146 	case X86EMUL_MODE_REAL:
2147 		return emulate_iret_real(ctxt);
2148 	case X86EMUL_MODE_VM86:
2149 	case X86EMUL_MODE_PROT16:
2150 	case X86EMUL_MODE_PROT32:
2151 	case X86EMUL_MODE_PROT64:
2152 	default:
2153 		/* iret from protected mode unimplemented yet */
2154 		return X86EMUL_UNHANDLEABLE;
2155 	}
2156 }
2157 
2158 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2159 {
2160 	int rc;
2161 	unsigned short sel;
2162 	struct desc_struct new_desc;
2163 	u8 cpl = ctxt->ops->cpl(ctxt);
2164 
2165 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2166 
2167 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2168 				       X86_TRANSFER_CALL_JMP,
2169 				       &new_desc);
2170 	if (rc != X86EMUL_CONTINUE)
2171 		return rc;
2172 
2173 	rc = assign_eip_far(ctxt, ctxt->src.val);
2174 	/* Error handling is not implemented. */
2175 	if (rc != X86EMUL_CONTINUE)
2176 		return X86EMUL_UNHANDLEABLE;
2177 
2178 	return rc;
2179 }
2180 
2181 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2182 {
2183 	return assign_eip_near(ctxt, ctxt->src.val);
2184 }
2185 
2186 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2187 {
2188 	int rc;
2189 	long int old_eip;
2190 
2191 	old_eip = ctxt->_eip;
2192 	rc = assign_eip_near(ctxt, ctxt->src.val);
2193 	if (rc != X86EMUL_CONTINUE)
2194 		return rc;
2195 	ctxt->src.val = old_eip;
2196 	rc = em_push(ctxt);
2197 	return rc;
2198 }
2199 
2200 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2201 {
2202 	u64 old = ctxt->dst.orig_val64;
2203 
2204 	if (ctxt->dst.bytes == 16)
2205 		return X86EMUL_UNHANDLEABLE;
2206 
2207 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2208 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2209 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2210 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2211 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2212 	} else {
2213 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2214 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2215 
2216 		ctxt->eflags |= X86_EFLAGS_ZF;
2217 	}
2218 	return X86EMUL_CONTINUE;
2219 }
2220 
2221 static int em_ret(struct x86_emulate_ctxt *ctxt)
2222 {
2223 	int rc;
2224 	unsigned long eip;
2225 
2226 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2227 	if (rc != X86EMUL_CONTINUE)
2228 		return rc;
2229 
2230 	return assign_eip_near(ctxt, eip);
2231 }
2232 
2233 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2234 {
2235 	int rc;
2236 	unsigned long eip, cs;
2237 	int cpl = ctxt->ops->cpl(ctxt);
2238 	struct desc_struct new_desc;
2239 
2240 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2241 	if (rc != X86EMUL_CONTINUE)
2242 		return rc;
2243 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2244 	if (rc != X86EMUL_CONTINUE)
2245 		return rc;
2246 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2247 				       X86_TRANSFER_RET,
2248 				       &new_desc);
2249 	if (rc != X86EMUL_CONTINUE)
2250 		return rc;
2251 	rc = assign_eip_far(ctxt, eip);
2252 	/* Error handling is not implemented. */
2253 	if (rc != X86EMUL_CONTINUE)
2254 		return X86EMUL_UNHANDLEABLE;
2255 
2256 	return rc;
2257 }
2258 
2259 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2260 {
2261         int rc;
2262 
2263         rc = em_ret_far(ctxt);
2264         if (rc != X86EMUL_CONTINUE)
2265                 return rc;
2266         rsp_increment(ctxt, ctxt->src.val);
2267         return X86EMUL_CONTINUE;
2268 }
2269 
2270 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2271 {
2272 	/* Save real source value, then compare EAX against destination. */
2273 	ctxt->dst.orig_val = ctxt->dst.val;
2274 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2275 	ctxt->src.orig_val = ctxt->src.val;
2276 	ctxt->src.val = ctxt->dst.orig_val;
2277 	fastop(ctxt, em_cmp);
2278 
2279 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2280 		/* Success: write back to memory; no update of EAX */
2281 		ctxt->src.type = OP_NONE;
2282 		ctxt->dst.val = ctxt->src.orig_val;
2283 	} else {
2284 		/* Failure: write the value we saw to EAX. */
2285 		ctxt->src.type = OP_REG;
2286 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2287 		ctxt->src.val = ctxt->dst.orig_val;
2288 		/* Create write-cycle to dest by writing the same value */
2289 		ctxt->dst.val = ctxt->dst.orig_val;
2290 	}
2291 	return X86EMUL_CONTINUE;
2292 }
2293 
2294 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2295 {
2296 	int seg = ctxt->src2.val;
2297 	unsigned short sel;
2298 	int rc;
2299 
2300 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2301 
2302 	rc = load_segment_descriptor(ctxt, sel, seg);
2303 	if (rc != X86EMUL_CONTINUE)
2304 		return rc;
2305 
2306 	ctxt->dst.val = ctxt->src.val;
2307 	return rc;
2308 }
2309 
2310 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2311 {
2312 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2313 		return emulate_ud(ctxt);
2314 
2315 	if (ctxt->ops->leave_smm(ctxt))
2316 		ctxt->ops->triple_fault(ctxt);
2317 
2318 	return emulator_recalc_and_set_mode(ctxt);
2319 }
2320 
2321 static void
2322 setup_syscalls_segments(struct desc_struct *cs, struct desc_struct *ss)
2323 {
2324 	cs->l = 0;		/* will be adjusted later */
2325 	set_desc_base(cs, 0);	/* flat segment */
2326 	cs->g = 1;		/* 4kb granularity */
2327 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2328 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2329 	cs->s = 1;
2330 	cs->dpl = 0;		/* will be adjusted later */
2331 	cs->p = 1;
2332 	cs->d = 1;
2333 	cs->avl = 0;
2334 
2335 	set_desc_base(ss, 0);	/* flat segment */
2336 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2337 	ss->g = 1;		/* 4kb granularity */
2338 	ss->s = 1;
2339 	ss->type = 0x03;	/* Read/Write, Accessed */
2340 	ss->d = 1;		/* 32bit stack segment */
2341 	ss->dpl = 0;
2342 	ss->p = 1;
2343 	ss->l = 0;
2344 	ss->avl = 0;
2345 }
2346 
2347 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2348 {
2349 	u32 eax, ebx, ecx, edx;
2350 
2351 	eax = ecx = 0;
2352 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2353 	return is_guest_vendor_intel(ebx, ecx, edx);
2354 }
2355 
2356 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2357 {
2358 	const struct x86_emulate_ops *ops = ctxt->ops;
2359 	u32 eax, ebx, ecx, edx;
2360 
2361 	/*
2362 	 * syscall should always be enabled in longmode - so only become
2363 	 * vendor specific (cpuid) if other modes are active...
2364 	 */
2365 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2366 		return true;
2367 
2368 	eax = 0x00000000;
2369 	ecx = 0x00000000;
2370 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2371 	/*
2372 	 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a
2373 	 * 64bit guest with a 32bit compat-app running will #UD !! While this
2374 	 * behaviour can be fixed (by emulating) into AMD response - CPUs of
2375 	 * AMD can't behave like Intel.
2376 	 */
2377 	if (is_guest_vendor_intel(ebx, ecx, edx))
2378 		return false;
2379 
2380 	if (is_guest_vendor_amd(ebx, ecx, edx) ||
2381 	    is_guest_vendor_hygon(ebx, ecx, edx))
2382 		return true;
2383 
2384 	/*
2385 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2386 	 * stricter rules...
2387 	 */
2388 	return false;
2389 }
2390 
2391 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2392 {
2393 	const struct x86_emulate_ops *ops = ctxt->ops;
2394 	struct desc_struct cs, ss;
2395 	u64 msr_data;
2396 	u16 cs_sel, ss_sel;
2397 	u64 efer = 0;
2398 
2399 	/* syscall is not available in real mode */
2400 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2401 	    ctxt->mode == X86EMUL_MODE_VM86)
2402 		return emulate_ud(ctxt);
2403 
2404 	if (!(em_syscall_is_enabled(ctxt)))
2405 		return emulate_ud(ctxt);
2406 
2407 	ops->get_msr(ctxt, MSR_EFER, &efer);
2408 	if (!(efer & EFER_SCE))
2409 		return emulate_ud(ctxt);
2410 
2411 	setup_syscalls_segments(&cs, &ss);
2412 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2413 	msr_data >>= 32;
2414 	cs_sel = (u16)(msr_data & 0xfffc);
2415 	ss_sel = (u16)(msr_data + 8);
2416 
2417 	if (efer & EFER_LMA) {
2418 		cs.d = 0;
2419 		cs.l = 1;
2420 	}
2421 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2422 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2423 
2424 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2425 	if (efer & EFER_LMA) {
2426 #ifdef CONFIG_X86_64
2427 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2428 
2429 		ops->get_msr(ctxt,
2430 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2431 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2432 		ctxt->_eip = msr_data;
2433 
2434 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2435 		ctxt->eflags &= ~msr_data;
2436 		ctxt->eflags |= X86_EFLAGS_FIXED;
2437 #endif
2438 	} else {
2439 		/* legacy mode */
2440 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2441 		ctxt->_eip = (u32)msr_data;
2442 
2443 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2444 	}
2445 
2446 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2447 	return X86EMUL_CONTINUE;
2448 }
2449 
2450 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2451 {
2452 	const struct x86_emulate_ops *ops = ctxt->ops;
2453 	struct desc_struct cs, ss;
2454 	u64 msr_data;
2455 	u16 cs_sel, ss_sel;
2456 	u64 efer = 0;
2457 
2458 	ops->get_msr(ctxt, MSR_EFER, &efer);
2459 	/* inject #GP if in real mode */
2460 	if (ctxt->mode == X86EMUL_MODE_REAL)
2461 		return emulate_gp(ctxt, 0);
2462 
2463 	/*
2464 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2465 	 * mode).
2466 	 */
2467 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2468 	    && !vendor_intel(ctxt))
2469 		return emulate_ud(ctxt);
2470 
2471 	/* sysenter/sysexit have not been tested in 64bit mode. */
2472 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2473 		return X86EMUL_UNHANDLEABLE;
2474 
2475 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2476 	if ((msr_data & 0xfffc) == 0x0)
2477 		return emulate_gp(ctxt, 0);
2478 
2479 	setup_syscalls_segments(&cs, &ss);
2480 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2481 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2482 	ss_sel = cs_sel + 8;
2483 	if (efer & EFER_LMA) {
2484 		cs.d = 0;
2485 		cs.l = 1;
2486 	}
2487 
2488 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2489 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2490 
2491 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2492 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2493 
2494 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2495 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2496 							      (u32)msr_data;
2497 	if (efer & EFER_LMA)
2498 		ctxt->mode = X86EMUL_MODE_PROT64;
2499 
2500 	return X86EMUL_CONTINUE;
2501 }
2502 
2503 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2504 {
2505 	const struct x86_emulate_ops *ops = ctxt->ops;
2506 	struct desc_struct cs, ss;
2507 	u64 msr_data, rcx, rdx;
2508 	int usermode;
2509 	u16 cs_sel = 0, ss_sel = 0;
2510 
2511 	/* inject #GP if in real mode or Virtual 8086 mode */
2512 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2513 	    ctxt->mode == X86EMUL_MODE_VM86)
2514 		return emulate_gp(ctxt, 0);
2515 
2516 	setup_syscalls_segments(&cs, &ss);
2517 
2518 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2519 		usermode = X86EMUL_MODE_PROT64;
2520 	else
2521 		usermode = X86EMUL_MODE_PROT32;
2522 
2523 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2524 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2525 
2526 	cs.dpl = 3;
2527 	ss.dpl = 3;
2528 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2529 	switch (usermode) {
2530 	case X86EMUL_MODE_PROT32:
2531 		cs_sel = (u16)(msr_data + 16);
2532 		if ((msr_data & 0xfffc) == 0x0)
2533 			return emulate_gp(ctxt, 0);
2534 		ss_sel = (u16)(msr_data + 24);
2535 		rcx = (u32)rcx;
2536 		rdx = (u32)rdx;
2537 		break;
2538 	case X86EMUL_MODE_PROT64:
2539 		cs_sel = (u16)(msr_data + 32);
2540 		if (msr_data == 0x0)
2541 			return emulate_gp(ctxt, 0);
2542 		ss_sel = cs_sel + 8;
2543 		cs.d = 0;
2544 		cs.l = 1;
2545 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2546 		    emul_is_noncanonical_address(rdx, ctxt))
2547 			return emulate_gp(ctxt, 0);
2548 		break;
2549 	}
2550 	cs_sel |= SEGMENT_RPL_MASK;
2551 	ss_sel |= SEGMENT_RPL_MASK;
2552 
2553 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2554 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2555 
2556 	ctxt->_eip = rdx;
2557 	ctxt->mode = usermode;
2558 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2559 
2560 	return X86EMUL_CONTINUE;
2561 }
2562 
2563 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2564 {
2565 	int iopl;
2566 	if (ctxt->mode == X86EMUL_MODE_REAL)
2567 		return false;
2568 	if (ctxt->mode == X86EMUL_MODE_VM86)
2569 		return true;
2570 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2571 	return ctxt->ops->cpl(ctxt) > iopl;
2572 }
2573 
2574 #define VMWARE_PORT_VMPORT	(0x5658)
2575 #define VMWARE_PORT_VMRPC	(0x5659)
2576 
2577 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2578 					    u16 port, u16 len)
2579 {
2580 	const struct x86_emulate_ops *ops = ctxt->ops;
2581 	struct desc_struct tr_seg;
2582 	u32 base3;
2583 	int r;
2584 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2585 	unsigned mask = (1 << len) - 1;
2586 	unsigned long base;
2587 
2588 	/*
2589 	 * VMware allows access to these ports even if denied
2590 	 * by TSS I/O permission bitmap. Mimic behavior.
2591 	 */
2592 	if (enable_vmware_backdoor &&
2593 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2594 		return true;
2595 
2596 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2597 	if (!tr_seg.p)
2598 		return false;
2599 	if (desc_limit_scaled(&tr_seg) < 103)
2600 		return false;
2601 	base = get_desc_base(&tr_seg);
2602 #ifdef CONFIG_X86_64
2603 	base |= ((u64)base3) << 32;
2604 #endif
2605 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2606 	if (r != X86EMUL_CONTINUE)
2607 		return false;
2608 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2609 		return false;
2610 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2611 	if (r != X86EMUL_CONTINUE)
2612 		return false;
2613 	if ((perm >> bit_idx) & mask)
2614 		return false;
2615 	return true;
2616 }
2617 
2618 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2619 				 u16 port, u16 len)
2620 {
2621 	if (ctxt->perm_ok)
2622 		return true;
2623 
2624 	if (emulator_bad_iopl(ctxt))
2625 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2626 			return false;
2627 
2628 	ctxt->perm_ok = true;
2629 
2630 	return true;
2631 }
2632 
2633 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2634 {
2635 	/*
2636 	 * Intel CPUs mask the counter and pointers in quite strange
2637 	 * manner when ECX is zero due to REP-string optimizations.
2638 	 */
2639 #ifdef CONFIG_X86_64
2640 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2641 		return;
2642 
2643 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2644 
2645 	switch (ctxt->b) {
2646 	case 0xa4:	/* movsb */
2647 	case 0xa5:	/* movsd/w */
2648 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2649 		fallthrough;
2650 	case 0xaa:	/* stosb */
2651 	case 0xab:	/* stosd/w */
2652 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2653 	}
2654 #endif
2655 }
2656 
2657 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2658 				struct tss_segment_16 *tss)
2659 {
2660 	tss->ip = ctxt->_eip;
2661 	tss->flag = ctxt->eflags;
2662 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2663 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2664 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2665 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2666 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2667 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2668 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2669 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2670 
2671 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2672 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2673 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2674 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2675 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2676 }
2677 
2678 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2679 				 struct tss_segment_16 *tss)
2680 {
2681 	int ret;
2682 	u8 cpl;
2683 
2684 	ctxt->_eip = tss->ip;
2685 	ctxt->eflags = tss->flag | 2;
2686 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2687 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2688 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2689 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2690 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2691 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2692 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2693 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2694 
2695 	/*
2696 	 * SDM says that segment selectors are loaded before segment
2697 	 * descriptors
2698 	 */
2699 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
2700 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2701 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2702 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2703 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2704 
2705 	cpl = tss->cs & 3;
2706 
2707 	/*
2708 	 * Now load segment descriptors. If fault happens at this stage
2709 	 * it is handled in a context of new task
2710 	 */
2711 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
2712 					X86_TRANSFER_TASK_SWITCH, NULL);
2713 	if (ret != X86EMUL_CONTINUE)
2714 		return ret;
2715 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
2716 					X86_TRANSFER_TASK_SWITCH, NULL);
2717 	if (ret != X86EMUL_CONTINUE)
2718 		return ret;
2719 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
2720 					X86_TRANSFER_TASK_SWITCH, NULL);
2721 	if (ret != X86EMUL_CONTINUE)
2722 		return ret;
2723 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
2724 					X86_TRANSFER_TASK_SWITCH, NULL);
2725 	if (ret != X86EMUL_CONTINUE)
2726 		return ret;
2727 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
2728 					X86_TRANSFER_TASK_SWITCH, NULL);
2729 	if (ret != X86EMUL_CONTINUE)
2730 		return ret;
2731 
2732 	return X86EMUL_CONTINUE;
2733 }
2734 
2735 static int task_switch_16(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
2736 			  ulong old_tss_base, struct desc_struct *new_desc)
2737 {
2738 	struct tss_segment_16 tss_seg;
2739 	int ret;
2740 	u32 new_tss_base = get_desc_base(new_desc);
2741 
2742 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2743 	if (ret != X86EMUL_CONTINUE)
2744 		return ret;
2745 
2746 	save_state_to_tss16(ctxt, &tss_seg);
2747 
2748 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2749 	if (ret != X86EMUL_CONTINUE)
2750 		return ret;
2751 
2752 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
2753 	if (ret != X86EMUL_CONTINUE)
2754 		return ret;
2755 
2756 	if (old_tss_sel != 0xffff) {
2757 		tss_seg.prev_task_link = old_tss_sel;
2758 
2759 		ret = linear_write_system(ctxt, new_tss_base,
2760 					  &tss_seg.prev_task_link,
2761 					  sizeof(tss_seg.prev_task_link));
2762 		if (ret != X86EMUL_CONTINUE)
2763 			return ret;
2764 	}
2765 
2766 	return load_state_from_tss16(ctxt, &tss_seg);
2767 }
2768 
2769 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
2770 				struct tss_segment_32 *tss)
2771 {
2772 	/* CR3 and ldt selector are not saved intentionally */
2773 	tss->eip = ctxt->_eip;
2774 	tss->eflags = ctxt->eflags;
2775 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
2776 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
2777 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
2778 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
2779 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
2780 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
2781 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
2782 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
2783 
2784 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2785 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2786 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2787 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2788 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
2789 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
2790 }
2791 
2792 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
2793 				 struct tss_segment_32 *tss)
2794 {
2795 	int ret;
2796 	u8 cpl;
2797 
2798 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
2799 		return emulate_gp(ctxt, 0);
2800 	ctxt->_eip = tss->eip;
2801 	ctxt->eflags = tss->eflags | 2;
2802 
2803 	/* General purpose registers */
2804 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
2805 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
2806 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
2807 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
2808 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
2809 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
2810 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
2811 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
2812 
2813 	/*
2814 	 * SDM says that segment selectors are loaded before segment
2815 	 * descriptors.  This is important because CPL checks will
2816 	 * use CS.RPL.
2817 	 */
2818 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
2819 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2820 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2821 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2822 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2823 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
2824 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
2825 
2826 	/*
2827 	 * If we're switching between Protected Mode and VM86, we need to make
2828 	 * sure to update the mode before loading the segment descriptors so
2829 	 * that the selectors are interpreted correctly.
2830 	 */
2831 	if (ctxt->eflags & X86_EFLAGS_VM) {
2832 		ctxt->mode = X86EMUL_MODE_VM86;
2833 		cpl = 3;
2834 	} else {
2835 		ctxt->mode = X86EMUL_MODE_PROT32;
2836 		cpl = tss->cs & 3;
2837 	}
2838 
2839 	/*
2840 	 * Now load segment descriptors. If fault happens at this stage
2841 	 * it is handled in a context of new task
2842 	 */
2843 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
2844 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
2845 	if (ret != X86EMUL_CONTINUE)
2846 		return ret;
2847 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
2848 					X86_TRANSFER_TASK_SWITCH, NULL);
2849 	if (ret != X86EMUL_CONTINUE)
2850 		return ret;
2851 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
2852 					X86_TRANSFER_TASK_SWITCH, NULL);
2853 	if (ret != X86EMUL_CONTINUE)
2854 		return ret;
2855 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
2856 					X86_TRANSFER_TASK_SWITCH, NULL);
2857 	if (ret != X86EMUL_CONTINUE)
2858 		return ret;
2859 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
2860 					X86_TRANSFER_TASK_SWITCH, NULL);
2861 	if (ret != X86EMUL_CONTINUE)
2862 		return ret;
2863 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
2864 					X86_TRANSFER_TASK_SWITCH, NULL);
2865 	if (ret != X86EMUL_CONTINUE)
2866 		return ret;
2867 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
2868 					X86_TRANSFER_TASK_SWITCH, NULL);
2869 
2870 	return ret;
2871 }
2872 
2873 static int task_switch_32(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
2874 			  ulong old_tss_base, struct desc_struct *new_desc)
2875 {
2876 	struct tss_segment_32 tss_seg;
2877 	int ret;
2878 	u32 new_tss_base = get_desc_base(new_desc);
2879 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
2880 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
2881 
2882 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2883 	if (ret != X86EMUL_CONTINUE)
2884 		return ret;
2885 
2886 	save_state_to_tss32(ctxt, &tss_seg);
2887 
2888 	/* Only GP registers and segment selectors are saved */
2889 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
2890 				  ldt_sel_offset - eip_offset);
2891 	if (ret != X86EMUL_CONTINUE)
2892 		return ret;
2893 
2894 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
2895 	if (ret != X86EMUL_CONTINUE)
2896 		return ret;
2897 
2898 	if (old_tss_sel != 0xffff) {
2899 		tss_seg.prev_task_link = old_tss_sel;
2900 
2901 		ret = linear_write_system(ctxt, new_tss_base,
2902 					  &tss_seg.prev_task_link,
2903 					  sizeof(tss_seg.prev_task_link));
2904 		if (ret != X86EMUL_CONTINUE)
2905 			return ret;
2906 	}
2907 
2908 	return load_state_from_tss32(ctxt, &tss_seg);
2909 }
2910 
2911 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
2912 				   u16 tss_selector, int idt_index, int reason,
2913 				   bool has_error_code, u32 error_code)
2914 {
2915 	const struct x86_emulate_ops *ops = ctxt->ops;
2916 	struct desc_struct curr_tss_desc, next_tss_desc;
2917 	int ret;
2918 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
2919 	ulong old_tss_base =
2920 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
2921 	u32 desc_limit;
2922 	ulong desc_addr, dr7;
2923 
2924 	/* FIXME: old_tss_base == ~0 ? */
2925 
2926 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
2927 	if (ret != X86EMUL_CONTINUE)
2928 		return ret;
2929 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
2930 	if (ret != X86EMUL_CONTINUE)
2931 		return ret;
2932 
2933 	/* FIXME: check that next_tss_desc is tss */
2934 
2935 	/*
2936 	 * Check privileges. The three cases are task switch caused by...
2937 	 *
2938 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
2939 	 * 2. Exception/IRQ/iret: No check is performed
2940 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
2941 	 *    hardware checks it before exiting.
2942 	 */
2943 	if (reason == TASK_SWITCH_GATE) {
2944 		if (idt_index != -1) {
2945 			/* Software interrupts */
2946 			struct desc_struct task_gate_desc;
2947 			int dpl;
2948 
2949 			ret = read_interrupt_descriptor(ctxt, idt_index,
2950 							&task_gate_desc);
2951 			if (ret != X86EMUL_CONTINUE)
2952 				return ret;
2953 
2954 			dpl = task_gate_desc.dpl;
2955 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
2956 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
2957 		}
2958 	}
2959 
2960 	desc_limit = desc_limit_scaled(&next_tss_desc);
2961 	if (!next_tss_desc.p ||
2962 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
2963 	     desc_limit < 0x2b)) {
2964 		return emulate_ts(ctxt, tss_selector & 0xfffc);
2965 	}
2966 
2967 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
2968 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
2969 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
2970 	}
2971 
2972 	if (reason == TASK_SWITCH_IRET)
2973 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
2974 
2975 	/* set back link to prev task only if NT bit is set in eflags
2976 	   note that old_tss_sel is not used after this point */
2977 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
2978 		old_tss_sel = 0xffff;
2979 
2980 	if (next_tss_desc.type & 8)
2981 		ret = task_switch_32(ctxt, old_tss_sel, old_tss_base, &next_tss_desc);
2982 	else
2983 		ret = task_switch_16(ctxt, old_tss_sel,
2984 				     old_tss_base, &next_tss_desc);
2985 	if (ret != X86EMUL_CONTINUE)
2986 		return ret;
2987 
2988 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
2989 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
2990 
2991 	if (reason != TASK_SWITCH_IRET) {
2992 		next_tss_desc.type |= (1 << 1); /* set busy flag */
2993 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
2994 	}
2995 
2996 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
2997 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
2998 
2999 	if (has_error_code) {
3000 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3001 		ctxt->lock_prefix = 0;
3002 		ctxt->src.val = (unsigned long) error_code;
3003 		ret = em_push(ctxt);
3004 	}
3005 
3006 	ops->get_dr(ctxt, 7, &dr7);
3007 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3008 
3009 	return ret;
3010 }
3011 
3012 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3013 			 u16 tss_selector, int idt_index, int reason,
3014 			 bool has_error_code, u32 error_code)
3015 {
3016 	int rc;
3017 
3018 	invalidate_registers(ctxt);
3019 	ctxt->_eip = ctxt->eip;
3020 	ctxt->dst.type = OP_NONE;
3021 
3022 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3023 				     has_error_code, error_code);
3024 
3025 	if (rc == X86EMUL_CONTINUE) {
3026 		ctxt->eip = ctxt->_eip;
3027 		writeback_registers(ctxt);
3028 	}
3029 
3030 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3031 }
3032 
3033 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3034 		struct operand *op)
3035 {
3036 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3037 
3038 	register_address_increment(ctxt, reg, df * op->bytes);
3039 	op->addr.mem.ea = register_address(ctxt, reg);
3040 }
3041 
3042 static int em_das(struct x86_emulate_ctxt *ctxt)
3043 {
3044 	u8 al, old_al;
3045 	bool af, cf, old_cf;
3046 
3047 	cf = ctxt->eflags & X86_EFLAGS_CF;
3048 	al = ctxt->dst.val;
3049 
3050 	old_al = al;
3051 	old_cf = cf;
3052 	cf = false;
3053 	af = ctxt->eflags & X86_EFLAGS_AF;
3054 	if ((al & 0x0f) > 9 || af) {
3055 		al -= 6;
3056 		cf = old_cf | (al >= 250);
3057 		af = true;
3058 	} else {
3059 		af = false;
3060 	}
3061 	if (old_al > 0x99 || old_cf) {
3062 		al -= 0x60;
3063 		cf = true;
3064 	}
3065 
3066 	ctxt->dst.val = al;
3067 	/* Set PF, ZF, SF */
3068 	ctxt->src.type = OP_IMM;
3069 	ctxt->src.val = 0;
3070 	ctxt->src.bytes = 1;
3071 	fastop(ctxt, em_or);
3072 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3073 	if (cf)
3074 		ctxt->eflags |= X86_EFLAGS_CF;
3075 	if (af)
3076 		ctxt->eflags |= X86_EFLAGS_AF;
3077 	return X86EMUL_CONTINUE;
3078 }
3079 
3080 static int em_aam(struct x86_emulate_ctxt *ctxt)
3081 {
3082 	u8 al, ah;
3083 
3084 	if (ctxt->src.val == 0)
3085 		return emulate_de(ctxt);
3086 
3087 	al = ctxt->dst.val & 0xff;
3088 	ah = al / ctxt->src.val;
3089 	al %= ctxt->src.val;
3090 
3091 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3092 
3093 	/* Set PF, ZF, SF */
3094 	ctxt->src.type = OP_IMM;
3095 	ctxt->src.val = 0;
3096 	ctxt->src.bytes = 1;
3097 	fastop(ctxt, em_or);
3098 
3099 	return X86EMUL_CONTINUE;
3100 }
3101 
3102 static int em_aad(struct x86_emulate_ctxt *ctxt)
3103 {
3104 	u8 al = ctxt->dst.val & 0xff;
3105 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3106 
3107 	al = (al + (ah * ctxt->src.val)) & 0xff;
3108 
3109 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3110 
3111 	/* Set PF, ZF, SF */
3112 	ctxt->src.type = OP_IMM;
3113 	ctxt->src.val = 0;
3114 	ctxt->src.bytes = 1;
3115 	fastop(ctxt, em_or);
3116 
3117 	return X86EMUL_CONTINUE;
3118 }
3119 
3120 static int em_call(struct x86_emulate_ctxt *ctxt)
3121 {
3122 	int rc;
3123 	long rel = ctxt->src.val;
3124 
3125 	ctxt->src.val = (unsigned long)ctxt->_eip;
3126 	rc = jmp_rel(ctxt, rel);
3127 	if (rc != X86EMUL_CONTINUE)
3128 		return rc;
3129 	return em_push(ctxt);
3130 }
3131 
3132 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3133 {
3134 	u16 sel, old_cs;
3135 	ulong old_eip;
3136 	int rc;
3137 	struct desc_struct old_desc, new_desc;
3138 	const struct x86_emulate_ops *ops = ctxt->ops;
3139 	int cpl = ctxt->ops->cpl(ctxt);
3140 	enum x86emul_mode prev_mode = ctxt->mode;
3141 
3142 	old_eip = ctxt->_eip;
3143 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3144 
3145 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3146 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3147 				       X86_TRANSFER_CALL_JMP, &new_desc);
3148 	if (rc != X86EMUL_CONTINUE)
3149 		return rc;
3150 
3151 	rc = assign_eip_far(ctxt, ctxt->src.val);
3152 	if (rc != X86EMUL_CONTINUE)
3153 		goto fail;
3154 
3155 	ctxt->src.val = old_cs;
3156 	rc = em_push(ctxt);
3157 	if (rc != X86EMUL_CONTINUE)
3158 		goto fail;
3159 
3160 	ctxt->src.val = old_eip;
3161 	rc = em_push(ctxt);
3162 	/* If we failed, we tainted the memory, but the very least we should
3163 	   restore cs */
3164 	if (rc != X86EMUL_CONTINUE) {
3165 		pr_warn_once("faulting far call emulation tainted memory\n");
3166 		goto fail;
3167 	}
3168 	return rc;
3169 fail:
3170 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3171 	ctxt->mode = prev_mode;
3172 	return rc;
3173 
3174 }
3175 
3176 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3177 {
3178 	int rc;
3179 	unsigned long eip;
3180 
3181 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3182 	if (rc != X86EMUL_CONTINUE)
3183 		return rc;
3184 	rc = assign_eip_near(ctxt, eip);
3185 	if (rc != X86EMUL_CONTINUE)
3186 		return rc;
3187 	rsp_increment(ctxt, ctxt->src.val);
3188 	return X86EMUL_CONTINUE;
3189 }
3190 
3191 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3192 {
3193 	/* Write back the register source. */
3194 	ctxt->src.val = ctxt->dst.val;
3195 	write_register_operand(&ctxt->src);
3196 
3197 	/* Write back the memory destination with implicit LOCK prefix. */
3198 	ctxt->dst.val = ctxt->src.orig_val;
3199 	ctxt->lock_prefix = 1;
3200 	return X86EMUL_CONTINUE;
3201 }
3202 
3203 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3204 {
3205 	ctxt->dst.val = ctxt->src2.val;
3206 	return fastop(ctxt, em_imul);
3207 }
3208 
3209 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3210 {
3211 	ctxt->dst.type = OP_REG;
3212 	ctxt->dst.bytes = ctxt->src.bytes;
3213 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3214 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3215 
3216 	return X86EMUL_CONTINUE;
3217 }
3218 
3219 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3220 {
3221 	u64 tsc_aux = 0;
3222 
3223 	if (!ctxt->ops->guest_has_rdpid(ctxt))
3224 		return emulate_ud(ctxt);
3225 
3226 	ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux);
3227 	ctxt->dst.val = tsc_aux;
3228 	return X86EMUL_CONTINUE;
3229 }
3230 
3231 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3232 {
3233 	u64 tsc = 0;
3234 
3235 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3236 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3237 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3238 	return X86EMUL_CONTINUE;
3239 }
3240 
3241 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3242 {
3243 	u64 pmc;
3244 
3245 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3246 		return emulate_gp(ctxt, 0);
3247 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3248 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3249 	return X86EMUL_CONTINUE;
3250 }
3251 
3252 static int em_mov(struct x86_emulate_ctxt *ctxt)
3253 {
3254 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3255 	return X86EMUL_CONTINUE;
3256 }
3257 
3258 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3259 {
3260 	u16 tmp;
3261 
3262 	if (!ctxt->ops->guest_has_movbe(ctxt))
3263 		return emulate_ud(ctxt);
3264 
3265 	switch (ctxt->op_bytes) {
3266 	case 2:
3267 		/*
3268 		 * From MOVBE definition: "...When the operand size is 16 bits,
3269 		 * the upper word of the destination register remains unchanged
3270 		 * ..."
3271 		 *
3272 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3273 		 * rules so we have to do the operation almost per hand.
3274 		 */
3275 		tmp = (u16)ctxt->src.val;
3276 		ctxt->dst.val &= ~0xffffUL;
3277 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3278 		break;
3279 	case 4:
3280 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3281 		break;
3282 	case 8:
3283 		ctxt->dst.val = swab64(ctxt->src.val);
3284 		break;
3285 	default:
3286 		BUG();
3287 	}
3288 	return X86EMUL_CONTINUE;
3289 }
3290 
3291 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3292 {
3293 	int cr_num = ctxt->modrm_reg;
3294 	int r;
3295 
3296 	if (ctxt->ops->set_cr(ctxt, cr_num, ctxt->src.val))
3297 		return emulate_gp(ctxt, 0);
3298 
3299 	/* Disable writeback. */
3300 	ctxt->dst.type = OP_NONE;
3301 
3302 	if (cr_num == 0) {
3303 		/*
3304 		 * CR0 write might have updated CR0.PE and/or CR0.PG
3305 		 * which can affect the cpu's execution mode.
3306 		 */
3307 		r = emulator_recalc_and_set_mode(ctxt);
3308 		if (r != X86EMUL_CONTINUE)
3309 			return r;
3310 	}
3311 
3312 	return X86EMUL_CONTINUE;
3313 }
3314 
3315 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3316 {
3317 	unsigned long val;
3318 
3319 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3320 		val = ctxt->src.val & ~0ULL;
3321 	else
3322 		val = ctxt->src.val & ~0U;
3323 
3324 	/* #UD condition is already handled. */
3325 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3326 		return emulate_gp(ctxt, 0);
3327 
3328 	/* Disable writeback. */
3329 	ctxt->dst.type = OP_NONE;
3330 	return X86EMUL_CONTINUE;
3331 }
3332 
3333 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3334 {
3335 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3336 	u64 msr_data;
3337 	int r;
3338 
3339 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3340 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3341 	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);
3342 
3343 	if (r == X86EMUL_PROPAGATE_FAULT)
3344 		return emulate_gp(ctxt, 0);
3345 
3346 	return r;
3347 }
3348 
3349 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3350 {
3351 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3352 	u64 msr_data;
3353 	int r;
3354 
3355 	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);
3356 
3357 	if (r == X86EMUL_PROPAGATE_FAULT)
3358 		return emulate_gp(ctxt, 0);
3359 
3360 	if (r == X86EMUL_CONTINUE) {
3361 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3362 		*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3363 	}
3364 	return r;
3365 }
3366 
3367 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3368 {
3369 	if (segment > VCPU_SREG_GS &&
3370 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3371 	    ctxt->ops->cpl(ctxt) > 0)
3372 		return emulate_gp(ctxt, 0);
3373 
3374 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3375 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3376 		ctxt->dst.bytes = 2;
3377 	return X86EMUL_CONTINUE;
3378 }
3379 
3380 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3381 {
3382 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3383 		return emulate_ud(ctxt);
3384 
3385 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3386 }
3387 
3388 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3389 {
3390 	u16 sel = ctxt->src.val;
3391 
3392 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3393 		return emulate_ud(ctxt);
3394 
3395 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3396 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3397 
3398 	/* Disable writeback. */
3399 	ctxt->dst.type = OP_NONE;
3400 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3401 }
3402 
3403 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3404 {
3405 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3406 }
3407 
3408 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3409 {
3410 	u16 sel = ctxt->src.val;
3411 
3412 	/* Disable writeback. */
3413 	ctxt->dst.type = OP_NONE;
3414 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3415 }
3416 
3417 static int em_str(struct x86_emulate_ctxt *ctxt)
3418 {
3419 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3420 }
3421 
3422 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3423 {
3424 	u16 sel = ctxt->src.val;
3425 
3426 	/* Disable writeback. */
3427 	ctxt->dst.type = OP_NONE;
3428 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3429 }
3430 
3431 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3432 {
3433 	int rc;
3434 	ulong linear;
3435 
3436 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3437 	if (rc == X86EMUL_CONTINUE)
3438 		ctxt->ops->invlpg(ctxt, linear);
3439 	/* Disable writeback. */
3440 	ctxt->dst.type = OP_NONE;
3441 	return X86EMUL_CONTINUE;
3442 }
3443 
3444 static int em_clts(struct x86_emulate_ctxt *ctxt)
3445 {
3446 	ulong cr0;
3447 
3448 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3449 	cr0 &= ~X86_CR0_TS;
3450 	ctxt->ops->set_cr(ctxt, 0, cr0);
3451 	return X86EMUL_CONTINUE;
3452 }
3453 
3454 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3455 {
3456 	int rc = ctxt->ops->fix_hypercall(ctxt);
3457 
3458 	if (rc != X86EMUL_CONTINUE)
3459 		return rc;
3460 
3461 	/* Let the processor re-execute the fixed hypercall */
3462 	ctxt->_eip = ctxt->eip;
3463 	/* Disable writeback. */
3464 	ctxt->dst.type = OP_NONE;
3465 	return X86EMUL_CONTINUE;
3466 }
3467 
3468 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3469 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3470 					      struct desc_ptr *ptr))
3471 {
3472 	struct desc_ptr desc_ptr;
3473 
3474 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3475 	    ctxt->ops->cpl(ctxt) > 0)
3476 		return emulate_gp(ctxt, 0);
3477 
3478 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3479 		ctxt->op_bytes = 8;
3480 	get(ctxt, &desc_ptr);
3481 	if (ctxt->op_bytes == 2) {
3482 		ctxt->op_bytes = 4;
3483 		desc_ptr.address &= 0x00ffffff;
3484 	}
3485 	/* Disable writeback. */
3486 	ctxt->dst.type = OP_NONE;
3487 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3488 				   &desc_ptr, 2 + ctxt->op_bytes);
3489 }
3490 
3491 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3492 {
3493 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3494 }
3495 
3496 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3497 {
3498 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3499 }
3500 
3501 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3502 {
3503 	struct desc_ptr desc_ptr;
3504 	int rc;
3505 
3506 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3507 		ctxt->op_bytes = 8;
3508 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3509 			     &desc_ptr.size, &desc_ptr.address,
3510 			     ctxt->op_bytes);
3511 	if (rc != X86EMUL_CONTINUE)
3512 		return rc;
3513 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3514 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3515 		return emulate_gp(ctxt, 0);
3516 	if (lgdt)
3517 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3518 	else
3519 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3520 	/* Disable writeback. */
3521 	ctxt->dst.type = OP_NONE;
3522 	return X86EMUL_CONTINUE;
3523 }
3524 
3525 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3526 {
3527 	return em_lgdt_lidt(ctxt, true);
3528 }
3529 
3530 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3531 {
3532 	return em_lgdt_lidt(ctxt, false);
3533 }
3534 
3535 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3536 {
3537 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3538 	    ctxt->ops->cpl(ctxt) > 0)
3539 		return emulate_gp(ctxt, 0);
3540 
3541 	if (ctxt->dst.type == OP_MEM)
3542 		ctxt->dst.bytes = 2;
3543 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3544 	return X86EMUL_CONTINUE;
3545 }
3546 
3547 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3548 {
3549 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3550 			  | (ctxt->src.val & 0x0f));
3551 	ctxt->dst.type = OP_NONE;
3552 	return X86EMUL_CONTINUE;
3553 }
3554 
3555 static int em_loop(struct x86_emulate_ctxt *ctxt)
3556 {
3557 	int rc = X86EMUL_CONTINUE;
3558 
3559 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3560 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3561 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3562 		rc = jmp_rel(ctxt, ctxt->src.val);
3563 
3564 	return rc;
3565 }
3566 
3567 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3568 {
3569 	int rc = X86EMUL_CONTINUE;
3570 
3571 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3572 		rc = jmp_rel(ctxt, ctxt->src.val);
3573 
3574 	return rc;
3575 }
3576 
3577 static int em_in(struct x86_emulate_ctxt *ctxt)
3578 {
3579 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3580 			     &ctxt->dst.val))
3581 		return X86EMUL_IO_NEEDED;
3582 
3583 	return X86EMUL_CONTINUE;
3584 }
3585 
3586 static int em_out(struct x86_emulate_ctxt *ctxt)
3587 {
3588 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3589 				    &ctxt->src.val, 1);
3590 	/* Disable writeback. */
3591 	ctxt->dst.type = OP_NONE;
3592 	return X86EMUL_CONTINUE;
3593 }
3594 
3595 static int em_cli(struct x86_emulate_ctxt *ctxt)
3596 {
3597 	if (emulator_bad_iopl(ctxt))
3598 		return emulate_gp(ctxt, 0);
3599 
3600 	ctxt->eflags &= ~X86_EFLAGS_IF;
3601 	return X86EMUL_CONTINUE;
3602 }
3603 
3604 static int em_sti(struct x86_emulate_ctxt *ctxt)
3605 {
3606 	if (emulator_bad_iopl(ctxt))
3607 		return emulate_gp(ctxt, 0);
3608 
3609 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3610 	ctxt->eflags |= X86_EFLAGS_IF;
3611 	return X86EMUL_CONTINUE;
3612 }
3613 
3614 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3615 {
3616 	u32 eax, ebx, ecx, edx;
3617 	u64 msr = 0;
3618 
3619 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3620 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3621 	    ctxt->ops->cpl(ctxt)) {
3622 		return emulate_gp(ctxt, 0);
3623 	}
3624 
3625 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3626 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3627 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3628 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3629 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3630 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3631 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3632 	return X86EMUL_CONTINUE;
3633 }
3634 
3635 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3636 {
3637 	u32 flags;
3638 
3639 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3640 		X86_EFLAGS_SF;
3641 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3642 
3643 	ctxt->eflags &= ~0xffUL;
3644 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3645 	return X86EMUL_CONTINUE;
3646 }
3647 
3648 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3649 {
3650 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3651 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3652 	return X86EMUL_CONTINUE;
3653 }
3654 
3655 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3656 {
3657 	switch (ctxt->op_bytes) {
3658 #ifdef CONFIG_X86_64
3659 	case 8:
3660 		asm("bswap %0" : "+r"(ctxt->dst.val));
3661 		break;
3662 #endif
3663 	default:
3664 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3665 		break;
3666 	}
3667 	return X86EMUL_CONTINUE;
3668 }
3669 
3670 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3671 {
3672 	/* emulating clflush regardless of cpuid */
3673 	return X86EMUL_CONTINUE;
3674 }
3675 
3676 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
3677 {
3678 	/* emulating clflushopt regardless of cpuid */
3679 	return X86EMUL_CONTINUE;
3680 }
3681 
3682 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3683 {
3684 	ctxt->dst.val = (s32) ctxt->src.val;
3685 	return X86EMUL_CONTINUE;
3686 }
3687 
3688 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3689 {
3690 	if (!ctxt->ops->guest_has_fxsr(ctxt))
3691 		return emulate_ud(ctxt);
3692 
3693 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3694 		return emulate_nm(ctxt);
3695 
3696 	/*
3697 	 * Don't emulate a case that should never be hit, instead of working
3698 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3699 	 */
3700 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3701 		return X86EMUL_UNHANDLEABLE;
3702 
3703 	return X86EMUL_CONTINUE;
3704 }
3705 
3706 /*
3707  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
3708  * and restore MXCSR.
3709  */
3710 static size_t __fxstate_size(int nregs)
3711 {
3712 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
3713 }
3714 
3715 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
3716 {
3717 	bool cr4_osfxsr;
3718 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3719 		return __fxstate_size(16);
3720 
3721 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
3722 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
3723 }
3724 
3725 /*
3726  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
3727  *  1) 16 bit mode
3728  *  2) 32 bit mode
3729  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
3730  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
3731  *       save and restore
3732  *  3) 64-bit mode with REX.W prefix
3733  *     - like (2), but XMM 8-15 are being saved and restored
3734  *  4) 64-bit mode without REX.W prefix
3735  *     - like (3), but FIP and FDP are 64 bit
3736  *
3737  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
3738  * desired result.  (4) is not emulated.
3739  *
3740  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
3741  * and FPU DS) should match.
3742  */
3743 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
3744 {
3745 	struct fxregs_state fx_state;
3746 	int rc;
3747 
3748 	rc = check_fxsr(ctxt);
3749 	if (rc != X86EMUL_CONTINUE)
3750 		return rc;
3751 
3752 	kvm_fpu_get();
3753 
3754 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
3755 
3756 	kvm_fpu_put();
3757 
3758 	if (rc != X86EMUL_CONTINUE)
3759 		return rc;
3760 
3761 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
3762 		                   fxstate_size(ctxt));
3763 }
3764 
3765 /*
3766  * FXRSTOR might restore XMM registers not provided by the guest. Fill
3767  * in the host registers (via FXSAVE) instead, so they won't be modified.
3768  * (preemption has to stay disabled until FXRSTOR).
3769  *
3770  * Use noinline to keep the stack for other functions called by callers small.
3771  */
3772 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
3773 				 const size_t used_size)
3774 {
3775 	struct fxregs_state fx_tmp;
3776 	int rc;
3777 
3778 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
3779 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
3780 	       __fxstate_size(16) - used_size);
3781 
3782 	return rc;
3783 }
3784 
3785 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
3786 {
3787 	struct fxregs_state fx_state;
3788 	int rc;
3789 	size_t size;
3790 
3791 	rc = check_fxsr(ctxt);
3792 	if (rc != X86EMUL_CONTINUE)
3793 		return rc;
3794 
3795 	size = fxstate_size(ctxt);
3796 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
3797 	if (rc != X86EMUL_CONTINUE)
3798 		return rc;
3799 
3800 	kvm_fpu_get();
3801 
3802 	if (size < __fxstate_size(16)) {
3803 		rc = fxregs_fixup(&fx_state, size);
3804 		if (rc != X86EMUL_CONTINUE)
3805 			goto out;
3806 	}
3807 
3808 	if (fx_state.mxcsr >> 16) {
3809 		rc = emulate_gp(ctxt, 0);
3810 		goto out;
3811 	}
3812 
3813 	if (rc == X86EMUL_CONTINUE)
3814 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
3815 
3816 out:
3817 	kvm_fpu_put();
3818 
3819 	return rc;
3820 }
3821 
3822 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
3823 {
3824 	u32 eax, ecx, edx;
3825 
3826 	if (!(ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSXSAVE))
3827 		return emulate_ud(ctxt);
3828 
3829 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3830 	edx = reg_read(ctxt, VCPU_REGS_RDX);
3831 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3832 
3833 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
3834 		return emulate_gp(ctxt, 0);
3835 
3836 	return X86EMUL_CONTINUE;
3837 }
3838 
3839 static bool valid_cr(int nr)
3840 {
3841 	switch (nr) {
3842 	case 0:
3843 	case 2 ... 4:
3844 	case 8:
3845 		return true;
3846 	default:
3847 		return false;
3848 	}
3849 }
3850 
3851 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
3852 {
3853 	if (!valid_cr(ctxt->modrm_reg))
3854 		return emulate_ud(ctxt);
3855 
3856 	return X86EMUL_CONTINUE;
3857 }
3858 
3859 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
3860 {
3861 	unsigned long dr7;
3862 
3863 	ctxt->ops->get_dr(ctxt, 7, &dr7);
3864 
3865 	return dr7 & DR7_GD;
3866 }
3867 
3868 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
3869 {
3870 	int dr = ctxt->modrm_reg;
3871 	u64 cr4;
3872 
3873 	if (dr > 7)
3874 		return emulate_ud(ctxt);
3875 
3876 	cr4 = ctxt->ops->get_cr(ctxt, 4);
3877 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
3878 		return emulate_ud(ctxt);
3879 
3880 	if (check_dr7_gd(ctxt)) {
3881 		ulong dr6;
3882 
3883 		ctxt->ops->get_dr(ctxt, 6, &dr6);
3884 		dr6 &= ~DR_TRAP_BITS;
3885 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
3886 		ctxt->ops->set_dr(ctxt, 6, dr6);
3887 		return emulate_db(ctxt);
3888 	}
3889 
3890 	return X86EMUL_CONTINUE;
3891 }
3892 
3893 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
3894 {
3895 	u64 new_val = ctxt->src.val64;
3896 	int dr = ctxt->modrm_reg;
3897 
3898 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
3899 		return emulate_gp(ctxt, 0);
3900 
3901 	return check_dr_read(ctxt);
3902 }
3903 
3904 static int check_svme(struct x86_emulate_ctxt *ctxt)
3905 {
3906 	u64 efer = 0;
3907 
3908 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
3909 
3910 	if (!(efer & EFER_SVME))
3911 		return emulate_ud(ctxt);
3912 
3913 	return X86EMUL_CONTINUE;
3914 }
3915 
3916 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
3917 {
3918 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
3919 
3920 	/* Valid physical address? */
3921 	if (rax & 0xffff000000000000ULL)
3922 		return emulate_gp(ctxt, 0);
3923 
3924 	return check_svme(ctxt);
3925 }
3926 
3927 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
3928 {
3929 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
3930 
3931 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
3932 		return emulate_gp(ctxt, 0);
3933 
3934 	return X86EMUL_CONTINUE;
3935 }
3936 
3937 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
3938 {
3939 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
3940 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
3941 
3942 	/*
3943 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
3944 	 * in Ring3 when CR4.PCE=0.
3945 	 */
3946 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
3947 		return X86EMUL_CONTINUE;
3948 
3949 	/*
3950 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
3951 	 * check however is unnecessary because CPL is always 0 outside
3952 	 * protected mode.
3953 	 */
3954 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
3955 	    ctxt->ops->check_pmc(ctxt, rcx))
3956 		return emulate_gp(ctxt, 0);
3957 
3958 	return X86EMUL_CONTINUE;
3959 }
3960 
3961 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
3962 {
3963 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
3964 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
3965 		return emulate_gp(ctxt, 0);
3966 
3967 	return X86EMUL_CONTINUE;
3968 }
3969 
3970 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
3971 {
3972 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
3973 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
3974 		return emulate_gp(ctxt, 0);
3975 
3976 	return X86EMUL_CONTINUE;
3977 }
3978 
3979 #define D(_y) { .flags = (_y) }
3980 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
3981 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
3982 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
3983 #define N    D(NotImpl)
3984 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
3985 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
3986 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
3987 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
3988 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
3989 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
3990 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
3991 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
3992 #define II(_f, _e, _i) \
3993 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
3994 #define IIP(_f, _e, _i, _p) \
3995 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
3996 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
3997 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
3998 
3999 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4000 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4001 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4002 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4003 #define I2bvIP(_f, _e, _i, _p) \
4004 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4005 
4006 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4007 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4008 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4009 
4010 static const struct opcode group7_rm0[] = {
4011 	N,
4012 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4013 	N, N, N, N, N, N,
4014 };
4015 
4016 static const struct opcode group7_rm1[] = {
4017 	DI(SrcNone | Priv, monitor),
4018 	DI(SrcNone | Priv, mwait),
4019 	N, N, N, N, N, N,
4020 };
4021 
4022 static const struct opcode group7_rm2[] = {
4023 	N,
4024 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4025 	N, N, N, N, N, N,
4026 };
4027 
4028 static const struct opcode group7_rm3[] = {
4029 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4030 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4031 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4032 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4033 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4034 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4035 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4036 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4037 };
4038 
4039 static const struct opcode group7_rm7[] = {
4040 	N,
4041 	DIP(SrcNone, rdtscp, check_rdtsc),
4042 	N, N, N, N, N, N,
4043 };
4044 
4045 static const struct opcode group1[] = {
4046 	F(Lock, em_add),
4047 	F(Lock | PageTable, em_or),
4048 	F(Lock, em_adc),
4049 	F(Lock, em_sbb),
4050 	F(Lock | PageTable, em_and),
4051 	F(Lock, em_sub),
4052 	F(Lock, em_xor),
4053 	F(NoWrite, em_cmp),
4054 };
4055 
4056 static const struct opcode group1A[] = {
4057 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4058 };
4059 
4060 static const struct opcode group2[] = {
4061 	F(DstMem | ModRM, em_rol),
4062 	F(DstMem | ModRM, em_ror),
4063 	F(DstMem | ModRM, em_rcl),
4064 	F(DstMem | ModRM, em_rcr),
4065 	F(DstMem | ModRM, em_shl),
4066 	F(DstMem | ModRM, em_shr),
4067 	F(DstMem | ModRM, em_shl),
4068 	F(DstMem | ModRM, em_sar),
4069 };
4070 
4071 static const struct opcode group3[] = {
4072 	F(DstMem | SrcImm | NoWrite, em_test),
4073 	F(DstMem | SrcImm | NoWrite, em_test),
4074 	F(DstMem | SrcNone | Lock, em_not),
4075 	F(DstMem | SrcNone | Lock, em_neg),
4076 	F(DstXacc | Src2Mem, em_mul_ex),
4077 	F(DstXacc | Src2Mem, em_imul_ex),
4078 	F(DstXacc | Src2Mem, em_div_ex),
4079 	F(DstXacc | Src2Mem, em_idiv_ex),
4080 };
4081 
4082 static const struct opcode group4[] = {
4083 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4084 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4085 	N, N, N, N, N, N,
4086 };
4087 
4088 static const struct opcode group5[] = {
4089 	F(DstMem | SrcNone | Lock,		em_inc),
4090 	F(DstMem | SrcNone | Lock,		em_dec),
4091 	I(SrcMem | NearBranch | IsBranch,       em_call_near_abs),
4092 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_call_far),
4093 	I(SrcMem | NearBranch | IsBranch,       em_jmp_abs),
4094 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_jmp_far),
4095 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4096 };
4097 
4098 static const struct opcode group6[] = {
4099 	II(Prot | DstMem,	   em_sldt, sldt),
4100 	II(Prot | DstMem,	   em_str, str),
4101 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4102 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4103 	N, N, N, N,
4104 };
4105 
4106 static const struct group_dual group7 = { {
4107 	II(Mov | DstMem,			em_sgdt, sgdt),
4108 	II(Mov | DstMem,			em_sidt, sidt),
4109 	II(SrcMem | Priv,			em_lgdt, lgdt),
4110 	II(SrcMem | Priv,			em_lidt, lidt),
4111 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4112 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4113 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4114 }, {
4115 	EXT(0, group7_rm0),
4116 	EXT(0, group7_rm1),
4117 	EXT(0, group7_rm2),
4118 	EXT(0, group7_rm3),
4119 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4120 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4121 	EXT(0, group7_rm7),
4122 } };
4123 
4124 static const struct opcode group8[] = {
4125 	N, N, N, N,
4126 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4127 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4128 	F(DstMem | SrcImmByte | Lock,			em_btr),
4129 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4130 };
4131 
4132 /*
4133  * The "memory" destination is actually always a register, since we come
4134  * from the register case of group9.
4135  */
4136 static const struct gprefix pfx_0f_c7_7 = {
4137 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4138 };
4139 
4140 
4141 static const struct group_dual group9 = { {
4142 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4143 }, {
4144 	N, N, N, N, N, N, N,
4145 	GP(0, &pfx_0f_c7_7),
4146 } };
4147 
4148 static const struct opcode group11[] = {
4149 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4150 	X7(D(Undefined)),
4151 };
4152 
4153 static const struct gprefix pfx_0f_ae_7 = {
4154 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4155 };
4156 
4157 static const struct group_dual group15 = { {
4158 	I(ModRM | Aligned16, em_fxsave),
4159 	I(ModRM | Aligned16, em_fxrstor),
4160 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4161 }, {
4162 	N, N, N, N, N, N, N, N,
4163 } };
4164 
4165 static const struct gprefix pfx_0f_6f_0f_7f = {
4166 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4167 };
4168 
4169 static const struct instr_dual instr_dual_0f_2b = {
4170 	I(0, em_mov), N
4171 };
4172 
4173 static const struct gprefix pfx_0f_2b = {
4174 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4175 };
4176 
4177 static const struct gprefix pfx_0f_10_0f_11 = {
4178 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4179 };
4180 
4181 static const struct gprefix pfx_0f_28_0f_29 = {
4182 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4183 };
4184 
4185 static const struct gprefix pfx_0f_e7 = {
4186 	N, I(Sse, em_mov), N, N,
4187 };
4188 
4189 static const struct escape escape_d9 = { {
4190 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4191 }, {
4192 	/* 0xC0 - 0xC7 */
4193 	N, N, N, N, N, N, N, N,
4194 	/* 0xC8 - 0xCF */
4195 	N, N, N, N, N, N, N, N,
4196 	/* 0xD0 - 0xC7 */
4197 	N, N, N, N, N, N, N, N,
4198 	/* 0xD8 - 0xDF */
4199 	N, N, N, N, N, N, N, N,
4200 	/* 0xE0 - 0xE7 */
4201 	N, N, N, N, N, N, N, N,
4202 	/* 0xE8 - 0xEF */
4203 	N, N, N, N, N, N, N, N,
4204 	/* 0xF0 - 0xF7 */
4205 	N, N, N, N, N, N, N, N,
4206 	/* 0xF8 - 0xFF */
4207 	N, N, N, N, N, N, N, N,
4208 } };
4209 
4210 static const struct escape escape_db = { {
4211 	N, N, N, N, N, N, N, N,
4212 }, {
4213 	/* 0xC0 - 0xC7 */
4214 	N, N, N, N, N, N, N, N,
4215 	/* 0xC8 - 0xCF */
4216 	N, N, N, N, N, N, N, N,
4217 	/* 0xD0 - 0xC7 */
4218 	N, N, N, N, N, N, N, N,
4219 	/* 0xD8 - 0xDF */
4220 	N, N, N, N, N, N, N, N,
4221 	/* 0xE0 - 0xE7 */
4222 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4223 	/* 0xE8 - 0xEF */
4224 	N, N, N, N, N, N, N, N,
4225 	/* 0xF0 - 0xF7 */
4226 	N, N, N, N, N, N, N, N,
4227 	/* 0xF8 - 0xFF */
4228 	N, N, N, N, N, N, N, N,
4229 } };
4230 
4231 static const struct escape escape_dd = { {
4232 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4233 }, {
4234 	/* 0xC0 - 0xC7 */
4235 	N, N, N, N, N, N, N, N,
4236 	/* 0xC8 - 0xCF */
4237 	N, N, N, N, N, N, N, N,
4238 	/* 0xD0 - 0xC7 */
4239 	N, N, N, N, N, N, N, N,
4240 	/* 0xD8 - 0xDF */
4241 	N, N, N, N, N, N, N, N,
4242 	/* 0xE0 - 0xE7 */
4243 	N, N, N, N, N, N, N, N,
4244 	/* 0xE8 - 0xEF */
4245 	N, N, N, N, N, N, N, N,
4246 	/* 0xF0 - 0xF7 */
4247 	N, N, N, N, N, N, N, N,
4248 	/* 0xF8 - 0xFF */
4249 	N, N, N, N, N, N, N, N,
4250 } };
4251 
4252 static const struct instr_dual instr_dual_0f_c3 = {
4253 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4254 };
4255 
4256 static const struct mode_dual mode_dual_63 = {
4257 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4258 };
4259 
4260 static const struct instr_dual instr_dual_8d = {
4261 	D(DstReg | SrcMem | ModRM | NoAccess), N
4262 };
4263 
4264 static const struct opcode opcode_table[256] = {
4265 	/* 0x00 - 0x07 */
4266 	F6ALU(Lock, em_add),
4267 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4268 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4269 	/* 0x08 - 0x0F */
4270 	F6ALU(Lock | PageTable, em_or),
4271 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4272 	N,
4273 	/* 0x10 - 0x17 */
4274 	F6ALU(Lock, em_adc),
4275 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4276 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4277 	/* 0x18 - 0x1F */
4278 	F6ALU(Lock, em_sbb),
4279 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4280 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4281 	/* 0x20 - 0x27 */
4282 	F6ALU(Lock | PageTable, em_and), N, N,
4283 	/* 0x28 - 0x2F */
4284 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4285 	/* 0x30 - 0x37 */
4286 	F6ALU(Lock, em_xor), N, N,
4287 	/* 0x38 - 0x3F */
4288 	F6ALU(NoWrite, em_cmp), N, N,
4289 	/* 0x40 - 0x4F */
4290 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4291 	/* 0x50 - 0x57 */
4292 	X8(I(SrcReg | Stack, em_push)),
4293 	/* 0x58 - 0x5F */
4294 	X8(I(DstReg | Stack, em_pop)),
4295 	/* 0x60 - 0x67 */
4296 	I(ImplicitOps | Stack | No64, em_pusha),
4297 	I(ImplicitOps | Stack | No64, em_popa),
4298 	N, MD(ModRM, &mode_dual_63),
4299 	N, N, N, N,
4300 	/* 0x68 - 0x6F */
4301 	I(SrcImm | Mov | Stack, em_push),
4302 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4303 	I(SrcImmByte | Mov | Stack, em_push),
4304 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4305 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4306 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4307 	/* 0x70 - 0x7F */
4308 	X16(D(SrcImmByte | NearBranch | IsBranch)),
4309 	/* 0x80 - 0x87 */
4310 	G(ByteOp | DstMem | SrcImm, group1),
4311 	G(DstMem | SrcImm, group1),
4312 	G(ByteOp | DstMem | SrcImm | No64, group1),
4313 	G(DstMem | SrcImmByte, group1),
4314 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4315 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4316 	/* 0x88 - 0x8F */
4317 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4318 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4319 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4320 	ID(0, &instr_dual_8d),
4321 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4322 	G(0, group1A),
4323 	/* 0x90 - 0x97 */
4324 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4325 	/* 0x98 - 0x9F */
4326 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4327 	I(SrcImmFAddr | No64 | IsBranch, em_call_far), N,
4328 	II(ImplicitOps | Stack, em_pushf, pushf),
4329 	II(ImplicitOps | Stack, em_popf, popf),
4330 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4331 	/* 0xA0 - 0xA7 */
4332 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4333 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4334 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4335 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4336 	/* 0xA8 - 0xAF */
4337 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4338 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4339 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4340 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4341 	/* 0xB0 - 0xB7 */
4342 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4343 	/* 0xB8 - 0xBF */
4344 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4345 	/* 0xC0 - 0xC7 */
4346 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4347 	I(ImplicitOps | NearBranch | SrcImmU16 | IsBranch, em_ret_near_imm),
4348 	I(ImplicitOps | NearBranch | IsBranch, em_ret),
4349 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4350 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4351 	G(ByteOp, group11), G(0, group11),
4352 	/* 0xC8 - 0xCF */
4353 	I(Stack | SrcImmU16 | Src2ImmByte | IsBranch, em_enter),
4354 	I(Stack | IsBranch, em_leave),
4355 	I(ImplicitOps | SrcImmU16 | IsBranch, em_ret_far_imm),
4356 	I(ImplicitOps | IsBranch, em_ret_far),
4357 	D(ImplicitOps | IsBranch), DI(SrcImmByte | IsBranch, intn),
4358 	D(ImplicitOps | No64 | IsBranch),
4359 	II(ImplicitOps | IsBranch, em_iret, iret),
4360 	/* 0xD0 - 0xD7 */
4361 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4362 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4363 	I(DstAcc | SrcImmUByte | No64, em_aam),
4364 	I(DstAcc | SrcImmUByte | No64, em_aad),
4365 	F(DstAcc | ByteOp | No64, em_salc),
4366 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4367 	/* 0xD8 - 0xDF */
4368 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4369 	/* 0xE0 - 0xE7 */
4370 	X3(I(SrcImmByte | NearBranch | IsBranch, em_loop)),
4371 	I(SrcImmByte | NearBranch | IsBranch, em_jcxz),
4372 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4373 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4374 	/* 0xE8 - 0xEF */
4375 	I(SrcImm | NearBranch | IsBranch, em_call),
4376 	D(SrcImm | ImplicitOps | NearBranch | IsBranch),
4377 	I(SrcImmFAddr | No64 | IsBranch, em_jmp_far),
4378 	D(SrcImmByte | ImplicitOps | NearBranch | IsBranch),
4379 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4380 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4381 	/* 0xF0 - 0xF7 */
4382 	N, DI(ImplicitOps, icebp), N, N,
4383 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4384 	G(ByteOp, group3), G(0, group3),
4385 	/* 0xF8 - 0xFF */
4386 	D(ImplicitOps), D(ImplicitOps),
4387 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4388 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4389 };
4390 
4391 static const struct opcode twobyte_table[256] = {
4392 	/* 0x00 - 0x0F */
4393 	G(0, group6), GD(0, &group7), N, N,
4394 	N, I(ImplicitOps | EmulateOnUD | IsBranch, em_syscall),
4395 	II(ImplicitOps | Priv, em_clts, clts), N,
4396 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4397 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4398 	/* 0x10 - 0x1F */
4399 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4400 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4401 	N, N, N, N, N, N,
4402 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4403 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4404 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4405 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4406 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4407 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4408 	/* 0x20 - 0x2F */
4409 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4410 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4411 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4412 						check_cr_access),
4413 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4414 						check_dr_write),
4415 	N, N, N, N,
4416 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4417 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4418 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4419 	N, N, N, N,
4420 	/* 0x30 - 0x3F */
4421 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4422 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4423 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4424 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4425 	I(ImplicitOps | EmulateOnUD | IsBranch, em_sysenter),
4426 	I(ImplicitOps | Priv | EmulateOnUD | IsBranch, em_sysexit),
4427 	N, N,
4428 	N, N, N, N, N, N, N, N,
4429 	/* 0x40 - 0x4F */
4430 	X16(D(DstReg | SrcMem | ModRM)),
4431 	/* 0x50 - 0x5F */
4432 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4433 	/* 0x60 - 0x6F */
4434 	N, N, N, N,
4435 	N, N, N, N,
4436 	N, N, N, N,
4437 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4438 	/* 0x70 - 0x7F */
4439 	N, N, N, N,
4440 	N, N, N, N,
4441 	N, N, N, N,
4442 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4443 	/* 0x80 - 0x8F */
4444 	X16(D(SrcImm | NearBranch | IsBranch)),
4445 	/* 0x90 - 0x9F */
4446 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4447 	/* 0xA0 - 0xA7 */
4448 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4449 	II(ImplicitOps, em_cpuid, cpuid),
4450 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4451 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4452 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4453 	/* 0xA8 - 0xAF */
4454 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4455 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4456 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4457 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4458 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4459 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4460 	/* 0xB0 - 0xB7 */
4461 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4462 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4463 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4464 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4465 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4466 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4467 	/* 0xB8 - 0xBF */
4468 	N, N,
4469 	G(BitOp, group8),
4470 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4471 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4472 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4473 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4474 	/* 0xC0 - 0xC7 */
4475 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4476 	N, ID(0, &instr_dual_0f_c3),
4477 	N, N, N, GD(0, &group9),
4478 	/* 0xC8 - 0xCF */
4479 	X8(I(DstReg, em_bswap)),
4480 	/* 0xD0 - 0xDF */
4481 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4482 	/* 0xE0 - 0xEF */
4483 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4484 	N, N, N, N, N, N, N, N,
4485 	/* 0xF0 - 0xFF */
4486 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4487 };
4488 
4489 static const struct instr_dual instr_dual_0f_38_f0 = {
4490 	I(DstReg | SrcMem | Mov, em_movbe), N
4491 };
4492 
4493 static const struct instr_dual instr_dual_0f_38_f1 = {
4494 	I(DstMem | SrcReg | Mov, em_movbe), N
4495 };
4496 
4497 static const struct gprefix three_byte_0f_38_f0 = {
4498 	ID(0, &instr_dual_0f_38_f0), N, N, N
4499 };
4500 
4501 static const struct gprefix three_byte_0f_38_f1 = {
4502 	ID(0, &instr_dual_0f_38_f1), N, N, N
4503 };
4504 
4505 /*
4506  * Insns below are selected by the prefix which indexed by the third opcode
4507  * byte.
4508  */
4509 static const struct opcode opcode_map_0f_38[256] = {
4510 	/* 0x00 - 0x7f */
4511 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4512 	/* 0x80 - 0xef */
4513 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4514 	/* 0xf0 - 0xf1 */
4515 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4516 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4517 	/* 0xf2 - 0xff */
4518 	N, N, X4(N), X8(N)
4519 };
4520 
4521 #undef D
4522 #undef N
4523 #undef G
4524 #undef GD
4525 #undef I
4526 #undef GP
4527 #undef EXT
4528 #undef MD
4529 #undef ID
4530 
4531 #undef D2bv
4532 #undef D2bvIP
4533 #undef I2bv
4534 #undef I2bvIP
4535 #undef I6ALU
4536 
4537 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4538 {
4539 	unsigned size;
4540 
4541 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4542 	if (size == 8)
4543 		size = 4;
4544 	return size;
4545 }
4546 
4547 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4548 		      unsigned size, bool sign_extension)
4549 {
4550 	int rc = X86EMUL_CONTINUE;
4551 
4552 	op->type = OP_IMM;
4553 	op->bytes = size;
4554 	op->addr.mem.ea = ctxt->_eip;
4555 	/* NB. Immediates are sign-extended as necessary. */
4556 	switch (op->bytes) {
4557 	case 1:
4558 		op->val = insn_fetch(s8, ctxt);
4559 		break;
4560 	case 2:
4561 		op->val = insn_fetch(s16, ctxt);
4562 		break;
4563 	case 4:
4564 		op->val = insn_fetch(s32, ctxt);
4565 		break;
4566 	case 8:
4567 		op->val = insn_fetch(s64, ctxt);
4568 		break;
4569 	}
4570 	if (!sign_extension) {
4571 		switch (op->bytes) {
4572 		case 1:
4573 			op->val &= 0xff;
4574 			break;
4575 		case 2:
4576 			op->val &= 0xffff;
4577 			break;
4578 		case 4:
4579 			op->val &= 0xffffffff;
4580 			break;
4581 		}
4582 	}
4583 done:
4584 	return rc;
4585 }
4586 
4587 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4588 			  unsigned d)
4589 {
4590 	int rc = X86EMUL_CONTINUE;
4591 
4592 	switch (d) {
4593 	case OpReg:
4594 		decode_register_operand(ctxt, op);
4595 		break;
4596 	case OpImmUByte:
4597 		rc = decode_imm(ctxt, op, 1, false);
4598 		break;
4599 	case OpMem:
4600 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4601 	mem_common:
4602 		*op = ctxt->memop;
4603 		ctxt->memopp = op;
4604 		if (ctxt->d & BitOp)
4605 			fetch_bit_operand(ctxt);
4606 		op->orig_val = op->val;
4607 		break;
4608 	case OpMem64:
4609 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4610 		goto mem_common;
4611 	case OpAcc:
4612 		op->type = OP_REG;
4613 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4614 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4615 		fetch_register_operand(op);
4616 		op->orig_val = op->val;
4617 		break;
4618 	case OpAccLo:
4619 		op->type = OP_REG;
4620 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4621 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4622 		fetch_register_operand(op);
4623 		op->orig_val = op->val;
4624 		break;
4625 	case OpAccHi:
4626 		if (ctxt->d & ByteOp) {
4627 			op->type = OP_NONE;
4628 			break;
4629 		}
4630 		op->type = OP_REG;
4631 		op->bytes = ctxt->op_bytes;
4632 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4633 		fetch_register_operand(op);
4634 		op->orig_val = op->val;
4635 		break;
4636 	case OpDI:
4637 		op->type = OP_MEM;
4638 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4639 		op->addr.mem.ea =
4640 			register_address(ctxt, VCPU_REGS_RDI);
4641 		op->addr.mem.seg = VCPU_SREG_ES;
4642 		op->val = 0;
4643 		op->count = 1;
4644 		break;
4645 	case OpDX:
4646 		op->type = OP_REG;
4647 		op->bytes = 2;
4648 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4649 		fetch_register_operand(op);
4650 		break;
4651 	case OpCL:
4652 		op->type = OP_IMM;
4653 		op->bytes = 1;
4654 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4655 		break;
4656 	case OpImmByte:
4657 		rc = decode_imm(ctxt, op, 1, true);
4658 		break;
4659 	case OpOne:
4660 		op->type = OP_IMM;
4661 		op->bytes = 1;
4662 		op->val = 1;
4663 		break;
4664 	case OpImm:
4665 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4666 		break;
4667 	case OpImm64:
4668 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4669 		break;
4670 	case OpMem8:
4671 		ctxt->memop.bytes = 1;
4672 		if (ctxt->memop.type == OP_REG) {
4673 			ctxt->memop.addr.reg = decode_register(ctxt,
4674 					ctxt->modrm_rm, true);
4675 			fetch_register_operand(&ctxt->memop);
4676 		}
4677 		goto mem_common;
4678 	case OpMem16:
4679 		ctxt->memop.bytes = 2;
4680 		goto mem_common;
4681 	case OpMem32:
4682 		ctxt->memop.bytes = 4;
4683 		goto mem_common;
4684 	case OpImmU16:
4685 		rc = decode_imm(ctxt, op, 2, false);
4686 		break;
4687 	case OpImmU:
4688 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4689 		break;
4690 	case OpSI:
4691 		op->type = OP_MEM;
4692 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4693 		op->addr.mem.ea =
4694 			register_address(ctxt, VCPU_REGS_RSI);
4695 		op->addr.mem.seg = ctxt->seg_override;
4696 		op->val = 0;
4697 		op->count = 1;
4698 		break;
4699 	case OpXLat:
4700 		op->type = OP_MEM;
4701 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4702 		op->addr.mem.ea =
4703 			address_mask(ctxt,
4704 				reg_read(ctxt, VCPU_REGS_RBX) +
4705 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
4706 		op->addr.mem.seg = ctxt->seg_override;
4707 		op->val = 0;
4708 		break;
4709 	case OpImmFAddr:
4710 		op->type = OP_IMM;
4711 		op->addr.mem.ea = ctxt->_eip;
4712 		op->bytes = ctxt->op_bytes + 2;
4713 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
4714 		break;
4715 	case OpMemFAddr:
4716 		ctxt->memop.bytes = ctxt->op_bytes + 2;
4717 		goto mem_common;
4718 	case OpES:
4719 		op->type = OP_IMM;
4720 		op->val = VCPU_SREG_ES;
4721 		break;
4722 	case OpCS:
4723 		op->type = OP_IMM;
4724 		op->val = VCPU_SREG_CS;
4725 		break;
4726 	case OpSS:
4727 		op->type = OP_IMM;
4728 		op->val = VCPU_SREG_SS;
4729 		break;
4730 	case OpDS:
4731 		op->type = OP_IMM;
4732 		op->val = VCPU_SREG_DS;
4733 		break;
4734 	case OpFS:
4735 		op->type = OP_IMM;
4736 		op->val = VCPU_SREG_FS;
4737 		break;
4738 	case OpGS:
4739 		op->type = OP_IMM;
4740 		op->val = VCPU_SREG_GS;
4741 		break;
4742 	case OpImplicit:
4743 		/* Special instructions do their own operand decoding. */
4744 	default:
4745 		op->type = OP_NONE; /* Disable writeback. */
4746 		break;
4747 	}
4748 
4749 done:
4750 	return rc;
4751 }
4752 
4753 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
4754 {
4755 	int rc = X86EMUL_CONTINUE;
4756 	int mode = ctxt->mode;
4757 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
4758 	bool op_prefix = false;
4759 	bool has_seg_override = false;
4760 	struct opcode opcode;
4761 	u16 dummy;
4762 	struct desc_struct desc;
4763 
4764 	ctxt->memop.type = OP_NONE;
4765 	ctxt->memopp = NULL;
4766 	ctxt->_eip = ctxt->eip;
4767 	ctxt->fetch.ptr = ctxt->fetch.data;
4768 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
4769 	ctxt->opcode_len = 1;
4770 	ctxt->intercept = x86_intercept_none;
4771 	if (insn_len > 0)
4772 		memcpy(ctxt->fetch.data, insn, insn_len);
4773 	else {
4774 		rc = __do_insn_fetch_bytes(ctxt, 1);
4775 		if (rc != X86EMUL_CONTINUE)
4776 			goto done;
4777 	}
4778 
4779 	switch (mode) {
4780 	case X86EMUL_MODE_REAL:
4781 	case X86EMUL_MODE_VM86:
4782 		def_op_bytes = def_ad_bytes = 2;
4783 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
4784 		if (desc.d)
4785 			def_op_bytes = def_ad_bytes = 4;
4786 		break;
4787 	case X86EMUL_MODE_PROT16:
4788 		def_op_bytes = def_ad_bytes = 2;
4789 		break;
4790 	case X86EMUL_MODE_PROT32:
4791 		def_op_bytes = def_ad_bytes = 4;
4792 		break;
4793 #ifdef CONFIG_X86_64
4794 	case X86EMUL_MODE_PROT64:
4795 		def_op_bytes = 4;
4796 		def_ad_bytes = 8;
4797 		break;
4798 #endif
4799 	default:
4800 		return EMULATION_FAILED;
4801 	}
4802 
4803 	ctxt->op_bytes = def_op_bytes;
4804 	ctxt->ad_bytes = def_ad_bytes;
4805 
4806 	/* Legacy prefixes. */
4807 	for (;;) {
4808 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
4809 		case 0x66:	/* operand-size override */
4810 			op_prefix = true;
4811 			/* switch between 2/4 bytes */
4812 			ctxt->op_bytes = def_op_bytes ^ 6;
4813 			break;
4814 		case 0x67:	/* address-size override */
4815 			if (mode == X86EMUL_MODE_PROT64)
4816 				/* switch between 4/8 bytes */
4817 				ctxt->ad_bytes = def_ad_bytes ^ 12;
4818 			else
4819 				/* switch between 2/4 bytes */
4820 				ctxt->ad_bytes = def_ad_bytes ^ 6;
4821 			break;
4822 		case 0x26:	/* ES override */
4823 			has_seg_override = true;
4824 			ctxt->seg_override = VCPU_SREG_ES;
4825 			break;
4826 		case 0x2e:	/* CS override */
4827 			has_seg_override = true;
4828 			ctxt->seg_override = VCPU_SREG_CS;
4829 			break;
4830 		case 0x36:	/* SS override */
4831 			has_seg_override = true;
4832 			ctxt->seg_override = VCPU_SREG_SS;
4833 			break;
4834 		case 0x3e:	/* DS override */
4835 			has_seg_override = true;
4836 			ctxt->seg_override = VCPU_SREG_DS;
4837 			break;
4838 		case 0x64:	/* FS override */
4839 			has_seg_override = true;
4840 			ctxt->seg_override = VCPU_SREG_FS;
4841 			break;
4842 		case 0x65:	/* GS override */
4843 			has_seg_override = true;
4844 			ctxt->seg_override = VCPU_SREG_GS;
4845 			break;
4846 		case 0x40 ... 0x4f: /* REX */
4847 			if (mode != X86EMUL_MODE_PROT64)
4848 				goto done_prefixes;
4849 			ctxt->rex_prefix = ctxt->b;
4850 			continue;
4851 		case 0xf0:	/* LOCK */
4852 			ctxt->lock_prefix = 1;
4853 			break;
4854 		case 0xf2:	/* REPNE/REPNZ */
4855 		case 0xf3:	/* REP/REPE/REPZ */
4856 			ctxt->rep_prefix = ctxt->b;
4857 			break;
4858 		default:
4859 			goto done_prefixes;
4860 		}
4861 
4862 		/* Any legacy prefix after a REX prefix nullifies its effect. */
4863 
4864 		ctxt->rex_prefix = 0;
4865 	}
4866 
4867 done_prefixes:
4868 
4869 	/* REX prefix. */
4870 	if (ctxt->rex_prefix & 8)
4871 		ctxt->op_bytes = 8;	/* REX.W */
4872 
4873 	/* Opcode byte(s). */
4874 	opcode = opcode_table[ctxt->b];
4875 	/* Two-byte opcode? */
4876 	if (ctxt->b == 0x0f) {
4877 		ctxt->opcode_len = 2;
4878 		ctxt->b = insn_fetch(u8, ctxt);
4879 		opcode = twobyte_table[ctxt->b];
4880 
4881 		/* 0F_38 opcode map */
4882 		if (ctxt->b == 0x38) {
4883 			ctxt->opcode_len = 3;
4884 			ctxt->b = insn_fetch(u8, ctxt);
4885 			opcode = opcode_map_0f_38[ctxt->b];
4886 		}
4887 	}
4888 	ctxt->d = opcode.flags;
4889 
4890 	if (ctxt->d & ModRM)
4891 		ctxt->modrm = insn_fetch(u8, ctxt);
4892 
4893 	/* vex-prefix instructions are not implemented */
4894 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
4895 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
4896 		ctxt->d = NotImpl;
4897 	}
4898 
4899 	while (ctxt->d & GroupMask) {
4900 		switch (ctxt->d & GroupMask) {
4901 		case Group:
4902 			goffset = (ctxt->modrm >> 3) & 7;
4903 			opcode = opcode.u.group[goffset];
4904 			break;
4905 		case GroupDual:
4906 			goffset = (ctxt->modrm >> 3) & 7;
4907 			if ((ctxt->modrm >> 6) == 3)
4908 				opcode = opcode.u.gdual->mod3[goffset];
4909 			else
4910 				opcode = opcode.u.gdual->mod012[goffset];
4911 			break;
4912 		case RMExt:
4913 			goffset = ctxt->modrm & 7;
4914 			opcode = opcode.u.group[goffset];
4915 			break;
4916 		case Prefix:
4917 			if (ctxt->rep_prefix && op_prefix)
4918 				return EMULATION_FAILED;
4919 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
4920 			switch (simd_prefix) {
4921 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
4922 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
4923 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
4924 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
4925 			}
4926 			break;
4927 		case Escape:
4928 			if (ctxt->modrm > 0xbf) {
4929 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
4930 				u32 index = array_index_nospec(
4931 					ctxt->modrm - 0xc0, size);
4932 
4933 				opcode = opcode.u.esc->high[index];
4934 			} else {
4935 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
4936 			}
4937 			break;
4938 		case InstrDual:
4939 			if ((ctxt->modrm >> 6) == 3)
4940 				opcode = opcode.u.idual->mod3;
4941 			else
4942 				opcode = opcode.u.idual->mod012;
4943 			break;
4944 		case ModeDual:
4945 			if (ctxt->mode == X86EMUL_MODE_PROT64)
4946 				opcode = opcode.u.mdual->mode64;
4947 			else
4948 				opcode = opcode.u.mdual->mode32;
4949 			break;
4950 		default:
4951 			return EMULATION_FAILED;
4952 		}
4953 
4954 		ctxt->d &= ~(u64)GroupMask;
4955 		ctxt->d |= opcode.flags;
4956 	}
4957 
4958 	ctxt->is_branch = opcode.flags & IsBranch;
4959 
4960 	/* Unrecognised? */
4961 	if (ctxt->d == 0)
4962 		return EMULATION_FAILED;
4963 
4964 	ctxt->execute = opcode.u.execute;
4965 
4966 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
4967 	    likely(!(ctxt->d & EmulateOnUD)))
4968 		return EMULATION_FAILED;
4969 
4970 	if (unlikely(ctxt->d &
4971 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
4972 	     No16))) {
4973 		/*
4974 		 * These are copied unconditionally here, and checked unconditionally
4975 		 * in x86_emulate_insn.
4976 		 */
4977 		ctxt->check_perm = opcode.check_perm;
4978 		ctxt->intercept = opcode.intercept;
4979 
4980 		if (ctxt->d & NotImpl)
4981 			return EMULATION_FAILED;
4982 
4983 		if (mode == X86EMUL_MODE_PROT64) {
4984 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
4985 				ctxt->op_bytes = 8;
4986 			else if (ctxt->d & NearBranch)
4987 				ctxt->op_bytes = 8;
4988 		}
4989 
4990 		if (ctxt->d & Op3264) {
4991 			if (mode == X86EMUL_MODE_PROT64)
4992 				ctxt->op_bytes = 8;
4993 			else
4994 				ctxt->op_bytes = 4;
4995 		}
4996 
4997 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
4998 			ctxt->op_bytes = 4;
4999 
5000 		if (ctxt->d & Sse)
5001 			ctxt->op_bytes = 16;
5002 		else if (ctxt->d & Mmx)
5003 			ctxt->op_bytes = 8;
5004 	}
5005 
5006 	/* ModRM and SIB bytes. */
5007 	if (ctxt->d & ModRM) {
5008 		rc = decode_modrm(ctxt, &ctxt->memop);
5009 		if (!has_seg_override) {
5010 			has_seg_override = true;
5011 			ctxt->seg_override = ctxt->modrm_seg;
5012 		}
5013 	} else if (ctxt->d & MemAbs)
5014 		rc = decode_abs(ctxt, &ctxt->memop);
5015 	if (rc != X86EMUL_CONTINUE)
5016 		goto done;
5017 
5018 	if (!has_seg_override)
5019 		ctxt->seg_override = VCPU_SREG_DS;
5020 
5021 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5022 
5023 	/*
5024 	 * Decode and fetch the source operand: register, memory
5025 	 * or immediate.
5026 	 */
5027 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5028 	if (rc != X86EMUL_CONTINUE)
5029 		goto done;
5030 
5031 	/*
5032 	 * Decode and fetch the second source operand: register, memory
5033 	 * or immediate.
5034 	 */
5035 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5036 	if (rc != X86EMUL_CONTINUE)
5037 		goto done;
5038 
5039 	/* Decode and fetch the destination operand: register or memory. */
5040 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5041 
5042 	if (ctxt->rip_relative && likely(ctxt->memopp))
5043 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5044 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5045 
5046 done:
5047 	if (rc == X86EMUL_PROPAGATE_FAULT)
5048 		ctxt->have_exception = true;
5049 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5050 }
5051 
5052 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5053 {
5054 	return ctxt->d & PageTable;
5055 }
5056 
5057 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5058 {
5059 	/* The second termination condition only applies for REPE
5060 	 * and REPNE. Test if the repeat string operation prefix is
5061 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5062 	 * corresponding termination condition according to:
5063 	 * 	- if REPE/REPZ and ZF = 0 then done
5064 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5065 	 */
5066 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5067 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5068 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5069 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5070 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5071 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5072 		return true;
5073 
5074 	return false;
5075 }
5076 
5077 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5078 {
5079 	int rc;
5080 
5081 	kvm_fpu_get();
5082 	rc = asm_safe("fwait");
5083 	kvm_fpu_put();
5084 
5085 	if (unlikely(rc != X86EMUL_CONTINUE))
5086 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5087 
5088 	return X86EMUL_CONTINUE;
5089 }
5090 
5091 static void fetch_possible_mmx_operand(struct operand *op)
5092 {
5093 	if (op->type == OP_MM)
5094 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5095 }
5096 
5097 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5098 {
5099 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5100 
5101 	if (!(ctxt->d & ByteOp))
5102 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5103 
5104 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5105 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5106 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5107 	    : "c"(ctxt->src2.val));
5108 
5109 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5110 	if (!fop) /* exception is returned in fop variable */
5111 		return emulate_de(ctxt);
5112 	return X86EMUL_CONTINUE;
5113 }
5114 
5115 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5116 {
5117 	/* Clear fields that are set conditionally but read without a guard. */
5118 	ctxt->rip_relative = false;
5119 	ctxt->rex_prefix = 0;
5120 	ctxt->lock_prefix = 0;
5121 	ctxt->rep_prefix = 0;
5122 	ctxt->regs_valid = 0;
5123 	ctxt->regs_dirty = 0;
5124 
5125 	ctxt->io_read.pos = 0;
5126 	ctxt->io_read.end = 0;
5127 	ctxt->mem_read.end = 0;
5128 }
5129 
5130 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5131 {
5132 	const struct x86_emulate_ops *ops = ctxt->ops;
5133 	int rc = X86EMUL_CONTINUE;
5134 	int saved_dst_type = ctxt->dst.type;
5135 	unsigned emul_flags;
5136 
5137 	ctxt->mem_read.pos = 0;
5138 
5139 	/* LOCK prefix is allowed only with some instructions */
5140 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5141 		rc = emulate_ud(ctxt);
5142 		goto done;
5143 	}
5144 
5145 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5146 		rc = emulate_ud(ctxt);
5147 		goto done;
5148 	}
5149 
5150 	emul_flags = ctxt->ops->get_hflags(ctxt);
5151 	if (unlikely(ctxt->d &
5152 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5153 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5154 				(ctxt->d & Undefined)) {
5155 			rc = emulate_ud(ctxt);
5156 			goto done;
5157 		}
5158 
5159 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5160 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5161 			rc = emulate_ud(ctxt);
5162 			goto done;
5163 		}
5164 
5165 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5166 			rc = emulate_nm(ctxt);
5167 			goto done;
5168 		}
5169 
5170 		if (ctxt->d & Mmx) {
5171 			rc = flush_pending_x87_faults(ctxt);
5172 			if (rc != X86EMUL_CONTINUE)
5173 				goto done;
5174 			/*
5175 			 * Now that we know the fpu is exception safe, we can fetch
5176 			 * operands from it.
5177 			 */
5178 			fetch_possible_mmx_operand(&ctxt->src);
5179 			fetch_possible_mmx_operand(&ctxt->src2);
5180 			if (!(ctxt->d & Mov))
5181 				fetch_possible_mmx_operand(&ctxt->dst);
5182 		}
5183 
5184 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5185 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5186 						      X86_ICPT_PRE_EXCEPT);
5187 			if (rc != X86EMUL_CONTINUE)
5188 				goto done;
5189 		}
5190 
5191 		/* Instruction can only be executed in protected mode */
5192 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5193 			rc = emulate_ud(ctxt);
5194 			goto done;
5195 		}
5196 
5197 		/* Privileged instruction can be executed only in CPL=0 */
5198 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5199 			if (ctxt->d & PrivUD)
5200 				rc = emulate_ud(ctxt);
5201 			else
5202 				rc = emulate_gp(ctxt, 0);
5203 			goto done;
5204 		}
5205 
5206 		/* Do instruction specific permission checks */
5207 		if (ctxt->d & CheckPerm) {
5208 			rc = ctxt->check_perm(ctxt);
5209 			if (rc != X86EMUL_CONTINUE)
5210 				goto done;
5211 		}
5212 
5213 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5214 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5215 						      X86_ICPT_POST_EXCEPT);
5216 			if (rc != X86EMUL_CONTINUE)
5217 				goto done;
5218 		}
5219 
5220 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5221 			/* All REP prefixes have the same first termination condition */
5222 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5223 				string_registers_quirk(ctxt);
5224 				ctxt->eip = ctxt->_eip;
5225 				ctxt->eflags &= ~X86_EFLAGS_RF;
5226 				goto done;
5227 			}
5228 		}
5229 	}
5230 
5231 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5232 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5233 				    ctxt->src.valptr, ctxt->src.bytes);
5234 		if (rc != X86EMUL_CONTINUE)
5235 			goto done;
5236 		ctxt->src.orig_val64 = ctxt->src.val64;
5237 	}
5238 
5239 	if (ctxt->src2.type == OP_MEM) {
5240 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5241 				    &ctxt->src2.val, ctxt->src2.bytes);
5242 		if (rc != X86EMUL_CONTINUE)
5243 			goto done;
5244 	}
5245 
5246 	if ((ctxt->d & DstMask) == ImplicitOps)
5247 		goto special_insn;
5248 
5249 
5250 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5251 		/* optimisation - avoid slow emulated read if Mov */
5252 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5253 				   &ctxt->dst.val, ctxt->dst.bytes);
5254 		if (rc != X86EMUL_CONTINUE) {
5255 			if (!(ctxt->d & NoWrite) &&
5256 			    rc == X86EMUL_PROPAGATE_FAULT &&
5257 			    ctxt->exception.vector == PF_VECTOR)
5258 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5259 			goto done;
5260 		}
5261 	}
5262 	/* Copy full 64-bit value for CMPXCHG8B.  */
5263 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5264 
5265 special_insn:
5266 
5267 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5268 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5269 					      X86_ICPT_POST_MEMACCESS);
5270 		if (rc != X86EMUL_CONTINUE)
5271 			goto done;
5272 	}
5273 
5274 	if (ctxt->rep_prefix && (ctxt->d & String))
5275 		ctxt->eflags |= X86_EFLAGS_RF;
5276 	else
5277 		ctxt->eflags &= ~X86_EFLAGS_RF;
5278 
5279 	if (ctxt->execute) {
5280 		if (ctxt->d & Fastop)
5281 			rc = fastop(ctxt, ctxt->fop);
5282 		else
5283 			rc = ctxt->execute(ctxt);
5284 		if (rc != X86EMUL_CONTINUE)
5285 			goto done;
5286 		goto writeback;
5287 	}
5288 
5289 	if (ctxt->opcode_len == 2)
5290 		goto twobyte_insn;
5291 	else if (ctxt->opcode_len == 3)
5292 		goto threebyte_insn;
5293 
5294 	switch (ctxt->b) {
5295 	case 0x70 ... 0x7f: /* jcc (short) */
5296 		if (test_cc(ctxt->b, ctxt->eflags))
5297 			rc = jmp_rel(ctxt, ctxt->src.val);
5298 		break;
5299 	case 0x8d: /* lea r16/r32, m */
5300 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5301 		break;
5302 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5303 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5304 			ctxt->dst.type = OP_NONE;
5305 		else
5306 			rc = em_xchg(ctxt);
5307 		break;
5308 	case 0x98: /* cbw/cwde/cdqe */
5309 		switch (ctxt->op_bytes) {
5310 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5311 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5312 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5313 		}
5314 		break;
5315 	case 0xcc:		/* int3 */
5316 		rc = emulate_int(ctxt, 3);
5317 		break;
5318 	case 0xcd:		/* int n */
5319 		rc = emulate_int(ctxt, ctxt->src.val);
5320 		break;
5321 	case 0xce:		/* into */
5322 		if (ctxt->eflags & X86_EFLAGS_OF)
5323 			rc = emulate_int(ctxt, 4);
5324 		break;
5325 	case 0xe9: /* jmp rel */
5326 	case 0xeb: /* jmp rel short */
5327 		rc = jmp_rel(ctxt, ctxt->src.val);
5328 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5329 		break;
5330 	case 0xf4:              /* hlt */
5331 		ctxt->ops->halt(ctxt);
5332 		break;
5333 	case 0xf5:	/* cmc */
5334 		/* complement carry flag from eflags reg */
5335 		ctxt->eflags ^= X86_EFLAGS_CF;
5336 		break;
5337 	case 0xf8: /* clc */
5338 		ctxt->eflags &= ~X86_EFLAGS_CF;
5339 		break;
5340 	case 0xf9: /* stc */
5341 		ctxt->eflags |= X86_EFLAGS_CF;
5342 		break;
5343 	case 0xfc: /* cld */
5344 		ctxt->eflags &= ~X86_EFLAGS_DF;
5345 		break;
5346 	case 0xfd: /* std */
5347 		ctxt->eflags |= X86_EFLAGS_DF;
5348 		break;
5349 	default:
5350 		goto cannot_emulate;
5351 	}
5352 
5353 	if (rc != X86EMUL_CONTINUE)
5354 		goto done;
5355 
5356 writeback:
5357 	if (ctxt->d & SrcWrite) {
5358 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5359 		rc = writeback(ctxt, &ctxt->src);
5360 		if (rc != X86EMUL_CONTINUE)
5361 			goto done;
5362 	}
5363 	if (!(ctxt->d & NoWrite)) {
5364 		rc = writeback(ctxt, &ctxt->dst);
5365 		if (rc != X86EMUL_CONTINUE)
5366 			goto done;
5367 	}
5368 
5369 	/*
5370 	 * restore dst type in case the decoding will be reused
5371 	 * (happens for string instruction )
5372 	 */
5373 	ctxt->dst.type = saved_dst_type;
5374 
5375 	if ((ctxt->d & SrcMask) == SrcSI)
5376 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5377 
5378 	if ((ctxt->d & DstMask) == DstDI)
5379 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5380 
5381 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5382 		unsigned int count;
5383 		struct read_cache *r = &ctxt->io_read;
5384 		if ((ctxt->d & SrcMask) == SrcSI)
5385 			count = ctxt->src.count;
5386 		else
5387 			count = ctxt->dst.count;
5388 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5389 
5390 		if (!string_insn_completed(ctxt)) {
5391 			/*
5392 			 * Re-enter guest when pio read ahead buffer is empty
5393 			 * or, if it is not used, after each 1024 iteration.
5394 			 */
5395 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5396 			    (r->end == 0 || r->end != r->pos)) {
5397 				/*
5398 				 * Reset read cache. Usually happens before
5399 				 * decode, but since instruction is restarted
5400 				 * we have to do it here.
5401 				 */
5402 				ctxt->mem_read.end = 0;
5403 				writeback_registers(ctxt);
5404 				return EMULATION_RESTART;
5405 			}
5406 			goto done; /* skip rip writeback */
5407 		}
5408 		ctxt->eflags &= ~X86_EFLAGS_RF;
5409 	}
5410 
5411 	ctxt->eip = ctxt->_eip;
5412 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5413 		ctxt->eip = (u32)ctxt->_eip;
5414 
5415 done:
5416 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5417 		if (KVM_EMULATOR_BUG_ON(ctxt->exception.vector > 0x1f, ctxt))
5418 			return EMULATION_FAILED;
5419 		ctxt->have_exception = true;
5420 	}
5421 	if (rc == X86EMUL_INTERCEPTED)
5422 		return EMULATION_INTERCEPTED;
5423 
5424 	if (rc == X86EMUL_CONTINUE)
5425 		writeback_registers(ctxt);
5426 
5427 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5428 
5429 twobyte_insn:
5430 	switch (ctxt->b) {
5431 	case 0x09:		/* wbinvd */
5432 		(ctxt->ops->wbinvd)(ctxt);
5433 		break;
5434 	case 0x08:		/* invd */
5435 	case 0x0d:		/* GrpP (prefetch) */
5436 	case 0x18:		/* Grp16 (prefetch/nop) */
5437 	case 0x1f:		/* nop */
5438 		break;
5439 	case 0x20: /* mov cr, reg */
5440 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5441 		break;
5442 	case 0x21: /* mov from dr to reg */
5443 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5444 		break;
5445 	case 0x40 ... 0x4f:	/* cmov */
5446 		if (test_cc(ctxt->b, ctxt->eflags))
5447 			ctxt->dst.val = ctxt->src.val;
5448 		else if (ctxt->op_bytes != 4)
5449 			ctxt->dst.type = OP_NONE; /* no writeback */
5450 		break;
5451 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5452 		if (test_cc(ctxt->b, ctxt->eflags))
5453 			rc = jmp_rel(ctxt, ctxt->src.val);
5454 		break;
5455 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5456 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5457 		break;
5458 	case 0xb6 ... 0xb7:	/* movzx */
5459 		ctxt->dst.bytes = ctxt->op_bytes;
5460 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5461 						       : (u16) ctxt->src.val;
5462 		break;
5463 	case 0xbe ... 0xbf:	/* movsx */
5464 		ctxt->dst.bytes = ctxt->op_bytes;
5465 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5466 							(s16) ctxt->src.val;
5467 		break;
5468 	default:
5469 		goto cannot_emulate;
5470 	}
5471 
5472 threebyte_insn:
5473 
5474 	if (rc != X86EMUL_CONTINUE)
5475 		goto done;
5476 
5477 	goto writeback;
5478 
5479 cannot_emulate:
5480 	return EMULATION_FAILED;
5481 }
5482 
5483 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5484 {
5485 	invalidate_registers(ctxt);
5486 }
5487 
5488 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5489 {
5490 	writeback_registers(ctxt);
5491 }
5492 
5493 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5494 {
5495 	if (ctxt->rep_prefix && (ctxt->d & String))
5496 		return false;
5497 
5498 	if (ctxt->d & TwoMemOp)
5499 		return false;
5500 
5501 	return true;
5502 }
5503