xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision 60772e48)
1 /******************************************************************************
2  * emulate.c
3  *
4  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
5  *
6  * Copyright (c) 2005 Keir Fraser
7  *
8  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
9  * privileged instructions:
10  *
11  * Copyright (C) 2006 Qumranet
12  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13  *
14  *   Avi Kivity <avi@qumranet.com>
15  *   Yaniv Kamay <yaniv@qumranet.com>
16  *
17  * This work is licensed under the terms of the GNU GPL, version 2.  See
18  * the COPYING file in the top-level directory.
19  *
20  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
21  */
22 
23 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
25 #include <asm/kvm_emulate.h>
26 #include <linux/stringify.h>
27 #include <asm/debugreg.h>
28 #include <asm/nospec-branch.h>
29 
30 #include "x86.h"
31 #include "tss.h"
32 #include "mmu.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 
180 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
181 
182 #define X2(x...) x, x
183 #define X3(x...) X2(x), x
184 #define X4(x...) X2(x), X2(x)
185 #define X5(x...) X4(x), x
186 #define X6(x...) X4(x), X2(x)
187 #define X7(x...) X4(x), X3(x)
188 #define X8(x...) X4(x), X4(x)
189 #define X16(x...) X8(x), X8(x)
190 
191 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1)
192 #define FASTOP_SIZE 8
193 
194 /*
195  * fastop functions have a special calling convention:
196  *
197  * dst:    rax        (in/out)
198  * src:    rdx        (in/out)
199  * src2:   rcx        (in)
200  * flags:  rflags     (in/out)
201  * ex:     rsi        (in:fastop pointer, out:zero if exception)
202  *
203  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
204  * different operand sizes can be reached by calculation, rather than a jump
205  * table (which would be bigger than the code).
206  *
207  * fastop functions are declared as taking a never-defined fastop parameter,
208  * so they can't be called from C directly.
209  */
210 
211 struct fastop;
212 
213 struct opcode {
214 	u64 flags : 56;
215 	u64 intercept : 8;
216 	union {
217 		int (*execute)(struct x86_emulate_ctxt *ctxt);
218 		const struct opcode *group;
219 		const struct group_dual *gdual;
220 		const struct gprefix *gprefix;
221 		const struct escape *esc;
222 		const struct instr_dual *idual;
223 		const struct mode_dual *mdual;
224 		void (*fastop)(struct fastop *fake);
225 	} u;
226 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
227 };
228 
229 struct group_dual {
230 	struct opcode mod012[8];
231 	struct opcode mod3[8];
232 };
233 
234 struct gprefix {
235 	struct opcode pfx_no;
236 	struct opcode pfx_66;
237 	struct opcode pfx_f2;
238 	struct opcode pfx_f3;
239 };
240 
241 struct escape {
242 	struct opcode op[8];
243 	struct opcode high[64];
244 };
245 
246 struct instr_dual {
247 	struct opcode mod012;
248 	struct opcode mod3;
249 };
250 
251 struct mode_dual {
252 	struct opcode mode32;
253 	struct opcode mode64;
254 };
255 
256 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
257 
258 enum x86_transfer_type {
259 	X86_TRANSFER_NONE,
260 	X86_TRANSFER_CALL_JMP,
261 	X86_TRANSFER_RET,
262 	X86_TRANSFER_TASK_SWITCH,
263 };
264 
265 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
266 {
267 	if (!(ctxt->regs_valid & (1 << nr))) {
268 		ctxt->regs_valid |= 1 << nr;
269 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
270 	}
271 	return ctxt->_regs[nr];
272 }
273 
274 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
275 {
276 	ctxt->regs_valid |= 1 << nr;
277 	ctxt->regs_dirty |= 1 << nr;
278 	return &ctxt->_regs[nr];
279 }
280 
281 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
282 {
283 	reg_read(ctxt, nr);
284 	return reg_write(ctxt, nr);
285 }
286 
287 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
288 {
289 	unsigned reg;
290 
291 	for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16)
292 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
293 }
294 
295 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
296 {
297 	ctxt->regs_dirty = 0;
298 	ctxt->regs_valid = 0;
299 }
300 
301 /*
302  * These EFLAGS bits are restored from saved value during emulation, and
303  * any changes are written back to the saved value after emulation.
304  */
305 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
306 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
307 
308 #ifdef CONFIG_X86_64
309 #define ON64(x) x
310 #else
311 #define ON64(x)
312 #endif
313 
314 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *));
315 
316 #define FOP_FUNC(name) \
317 	".align " __stringify(FASTOP_SIZE) " \n\t" \
318 	".type " name ", @function \n\t" \
319 	name ":\n\t"
320 
321 #define FOP_RET   "ret \n\t"
322 
323 #define FOP_START(op) \
324 	extern void em_##op(struct fastop *fake); \
325 	asm(".pushsection .text, \"ax\" \n\t" \
326 	    ".global em_" #op " \n\t" \
327 	    FOP_FUNC("em_" #op)
328 
329 #define FOP_END \
330 	    ".popsection")
331 
332 #define FOPNOP() \
333 	FOP_FUNC(__stringify(__UNIQUE_ID(nop))) \
334 	FOP_RET
335 
336 #define FOP1E(op,  dst) \
337 	FOP_FUNC(#op "_" #dst) \
338 	"10: " #op " %" #dst " \n\t" FOP_RET
339 
340 #define FOP1EEX(op,  dst) \
341 	FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception)
342 
343 #define FASTOP1(op) \
344 	FOP_START(op) \
345 	FOP1E(op##b, al) \
346 	FOP1E(op##w, ax) \
347 	FOP1E(op##l, eax) \
348 	ON64(FOP1E(op##q, rax))	\
349 	FOP_END
350 
351 /* 1-operand, using src2 (for MUL/DIV r/m) */
352 #define FASTOP1SRC2(op, name) \
353 	FOP_START(name) \
354 	FOP1E(op, cl) \
355 	FOP1E(op, cx) \
356 	FOP1E(op, ecx) \
357 	ON64(FOP1E(op, rcx)) \
358 	FOP_END
359 
360 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
361 #define FASTOP1SRC2EX(op, name) \
362 	FOP_START(name) \
363 	FOP1EEX(op, cl) \
364 	FOP1EEX(op, cx) \
365 	FOP1EEX(op, ecx) \
366 	ON64(FOP1EEX(op, rcx)) \
367 	FOP_END
368 
369 #define FOP2E(op,  dst, src)	   \
370 	FOP_FUNC(#op "_" #dst "_" #src) \
371 	#op " %" #src ", %" #dst " \n\t" FOP_RET
372 
373 #define FASTOP2(op) \
374 	FOP_START(op) \
375 	FOP2E(op##b, al, dl) \
376 	FOP2E(op##w, ax, dx) \
377 	FOP2E(op##l, eax, edx) \
378 	ON64(FOP2E(op##q, rax, rdx)) \
379 	FOP_END
380 
381 /* 2 operand, word only */
382 #define FASTOP2W(op) \
383 	FOP_START(op) \
384 	FOPNOP() \
385 	FOP2E(op##w, ax, dx) \
386 	FOP2E(op##l, eax, edx) \
387 	ON64(FOP2E(op##q, rax, rdx)) \
388 	FOP_END
389 
390 /* 2 operand, src is CL */
391 #define FASTOP2CL(op) \
392 	FOP_START(op) \
393 	FOP2E(op##b, al, cl) \
394 	FOP2E(op##w, ax, cl) \
395 	FOP2E(op##l, eax, cl) \
396 	ON64(FOP2E(op##q, rax, cl)) \
397 	FOP_END
398 
399 /* 2 operand, src and dest are reversed */
400 #define FASTOP2R(op, name) \
401 	FOP_START(name) \
402 	FOP2E(op##b, dl, al) \
403 	FOP2E(op##w, dx, ax) \
404 	FOP2E(op##l, edx, eax) \
405 	ON64(FOP2E(op##q, rdx, rax)) \
406 	FOP_END
407 
408 #define FOP3E(op,  dst, src, src2) \
409 	FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
410 	#op " %" #src2 ", %" #src ", %" #dst " \n\t" FOP_RET
411 
412 /* 3-operand, word-only, src2=cl */
413 #define FASTOP3WCL(op) \
414 	FOP_START(op) \
415 	FOPNOP() \
416 	FOP3E(op##w, ax, dx, cl) \
417 	FOP3E(op##l, eax, edx, cl) \
418 	ON64(FOP3E(op##q, rax, rdx, cl)) \
419 	FOP_END
420 
421 /* Special case for SETcc - 1 instruction per cc */
422 #define FOP_SETCC(op) \
423 	".align 4 \n\t" \
424 	".type " #op ", @function \n\t" \
425 	#op ": \n\t" \
426 	#op " %al \n\t" \
427 	FOP_RET
428 
429 asm(".pushsection .fixup, \"ax\"\n"
430     ".global kvm_fastop_exception \n"
431     "kvm_fastop_exception: xor %esi, %esi; ret\n"
432     ".popsection");
433 
434 FOP_START(setcc)
435 FOP_SETCC(seto)
436 FOP_SETCC(setno)
437 FOP_SETCC(setc)
438 FOP_SETCC(setnc)
439 FOP_SETCC(setz)
440 FOP_SETCC(setnz)
441 FOP_SETCC(setbe)
442 FOP_SETCC(setnbe)
443 FOP_SETCC(sets)
444 FOP_SETCC(setns)
445 FOP_SETCC(setp)
446 FOP_SETCC(setnp)
447 FOP_SETCC(setl)
448 FOP_SETCC(setnl)
449 FOP_SETCC(setle)
450 FOP_SETCC(setnle)
451 FOP_END;
452 
453 FOP_START(salc) "pushf; sbb %al, %al; popf \n\t" FOP_RET
454 FOP_END;
455 
456 /*
457  * XXX: inoutclob user must know where the argument is being expanded.
458  *      Relying on CC_HAVE_ASM_GOTO would allow us to remove _fault.
459  */
460 #define asm_safe(insn, inoutclob...) \
461 ({ \
462 	int _fault = 0; \
463  \
464 	asm volatile("1:" insn "\n" \
465 	             "2:\n" \
466 	             ".pushsection .fixup, \"ax\"\n" \
467 	             "3: movl $1, %[_fault]\n" \
468 	             "   jmp  2b\n" \
469 	             ".popsection\n" \
470 	             _ASM_EXTABLE(1b, 3b) \
471 	             : [_fault] "+qm"(_fault) inoutclob ); \
472  \
473 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
474 })
475 
476 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
477 				    enum x86_intercept intercept,
478 				    enum x86_intercept_stage stage)
479 {
480 	struct x86_instruction_info info = {
481 		.intercept  = intercept,
482 		.rep_prefix = ctxt->rep_prefix,
483 		.modrm_mod  = ctxt->modrm_mod,
484 		.modrm_reg  = ctxt->modrm_reg,
485 		.modrm_rm   = ctxt->modrm_rm,
486 		.src_val    = ctxt->src.val64,
487 		.dst_val    = ctxt->dst.val64,
488 		.src_bytes  = ctxt->src.bytes,
489 		.dst_bytes  = ctxt->dst.bytes,
490 		.ad_bytes   = ctxt->ad_bytes,
491 		.next_rip   = ctxt->eip,
492 	};
493 
494 	return ctxt->ops->intercept(ctxt, &info, stage);
495 }
496 
497 static void assign_masked(ulong *dest, ulong src, ulong mask)
498 {
499 	*dest = (*dest & ~mask) | (src & mask);
500 }
501 
502 static void assign_register(unsigned long *reg, u64 val, int bytes)
503 {
504 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
505 	switch (bytes) {
506 	case 1:
507 		*(u8 *)reg = (u8)val;
508 		break;
509 	case 2:
510 		*(u16 *)reg = (u16)val;
511 		break;
512 	case 4:
513 		*reg = (u32)val;
514 		break;	/* 64b: zero-extend */
515 	case 8:
516 		*reg = val;
517 		break;
518 	}
519 }
520 
521 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
522 {
523 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
524 }
525 
526 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
527 {
528 	u16 sel;
529 	struct desc_struct ss;
530 
531 	if (ctxt->mode == X86EMUL_MODE_PROT64)
532 		return ~0UL;
533 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
534 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
535 }
536 
537 static int stack_size(struct x86_emulate_ctxt *ctxt)
538 {
539 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
540 }
541 
542 /* Access/update address held in a register, based on addressing mode. */
543 static inline unsigned long
544 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
545 {
546 	if (ctxt->ad_bytes == sizeof(unsigned long))
547 		return reg;
548 	else
549 		return reg & ad_mask(ctxt);
550 }
551 
552 static inline unsigned long
553 register_address(struct x86_emulate_ctxt *ctxt, int reg)
554 {
555 	return address_mask(ctxt, reg_read(ctxt, reg));
556 }
557 
558 static void masked_increment(ulong *reg, ulong mask, int inc)
559 {
560 	assign_masked(reg, *reg + inc, mask);
561 }
562 
563 static inline void
564 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
565 {
566 	ulong *preg = reg_rmw(ctxt, reg);
567 
568 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
569 }
570 
571 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
572 {
573 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
574 }
575 
576 static u32 desc_limit_scaled(struct desc_struct *desc)
577 {
578 	u32 limit = get_desc_limit(desc);
579 
580 	return desc->g ? (limit << 12) | 0xfff : limit;
581 }
582 
583 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
584 {
585 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
586 		return 0;
587 
588 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
589 }
590 
591 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
592 			     u32 error, bool valid)
593 {
594 	WARN_ON(vec > 0x1f);
595 	ctxt->exception.vector = vec;
596 	ctxt->exception.error_code = error;
597 	ctxt->exception.error_code_valid = valid;
598 	return X86EMUL_PROPAGATE_FAULT;
599 }
600 
601 static int emulate_db(struct x86_emulate_ctxt *ctxt)
602 {
603 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
604 }
605 
606 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
607 {
608 	return emulate_exception(ctxt, GP_VECTOR, err, true);
609 }
610 
611 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
612 {
613 	return emulate_exception(ctxt, SS_VECTOR, err, true);
614 }
615 
616 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
617 {
618 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
619 }
620 
621 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
622 {
623 	return emulate_exception(ctxt, TS_VECTOR, err, true);
624 }
625 
626 static int emulate_de(struct x86_emulate_ctxt *ctxt)
627 {
628 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
629 }
630 
631 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
632 {
633 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
634 }
635 
636 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
637 {
638 	u16 selector;
639 	struct desc_struct desc;
640 
641 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
642 	return selector;
643 }
644 
645 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
646 				 unsigned seg)
647 {
648 	u16 dummy;
649 	u32 base3;
650 	struct desc_struct desc;
651 
652 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
653 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
654 }
655 
656 /*
657  * x86 defines three classes of vector instructions: explicitly
658  * aligned, explicitly unaligned, and the rest, which change behaviour
659  * depending on whether they're AVX encoded or not.
660  *
661  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
662  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
663  * 512 bytes of data must be aligned to a 16 byte boundary.
664  */
665 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
666 {
667 	u64 alignment = ctxt->d & AlignMask;
668 
669 	if (likely(size < 16))
670 		return 1;
671 
672 	switch (alignment) {
673 	case Unaligned:
674 	case Avx:
675 		return 1;
676 	case Aligned16:
677 		return 16;
678 	case Aligned:
679 	default:
680 		return size;
681 	}
682 }
683 
684 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
685 				       struct segmented_address addr,
686 				       unsigned *max_size, unsigned size,
687 				       bool write, bool fetch,
688 				       enum x86emul_mode mode, ulong *linear)
689 {
690 	struct desc_struct desc;
691 	bool usable;
692 	ulong la;
693 	u32 lim;
694 	u16 sel;
695 	u8  va_bits;
696 
697 	la = seg_base(ctxt, addr.seg) + addr.ea;
698 	*max_size = 0;
699 	switch (mode) {
700 	case X86EMUL_MODE_PROT64:
701 		*linear = la;
702 		va_bits = ctxt_virt_addr_bits(ctxt);
703 		if (get_canonical(la, va_bits) != la)
704 			goto bad;
705 
706 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
707 		if (size > *max_size)
708 			goto bad;
709 		break;
710 	default:
711 		*linear = la = (u32)la;
712 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
713 						addr.seg);
714 		if (!usable)
715 			goto bad;
716 		/* code segment in protected mode or read-only data segment */
717 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
718 					|| !(desc.type & 2)) && write)
719 			goto bad;
720 		/* unreadable code segment */
721 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
722 			goto bad;
723 		lim = desc_limit_scaled(&desc);
724 		if (!(desc.type & 8) && (desc.type & 4)) {
725 			/* expand-down segment */
726 			if (addr.ea <= lim)
727 				goto bad;
728 			lim = desc.d ? 0xffffffff : 0xffff;
729 		}
730 		if (addr.ea > lim)
731 			goto bad;
732 		if (lim == 0xffffffff)
733 			*max_size = ~0u;
734 		else {
735 			*max_size = (u64)lim + 1 - addr.ea;
736 			if (size > *max_size)
737 				goto bad;
738 		}
739 		break;
740 	}
741 	if (la & (insn_alignment(ctxt, size) - 1))
742 		return emulate_gp(ctxt, 0);
743 	return X86EMUL_CONTINUE;
744 bad:
745 	if (addr.seg == VCPU_SREG_SS)
746 		return emulate_ss(ctxt, 0);
747 	else
748 		return emulate_gp(ctxt, 0);
749 }
750 
751 static int linearize(struct x86_emulate_ctxt *ctxt,
752 		     struct segmented_address addr,
753 		     unsigned size, bool write,
754 		     ulong *linear)
755 {
756 	unsigned max_size;
757 	return __linearize(ctxt, addr, &max_size, size, write, false,
758 			   ctxt->mode, linear);
759 }
760 
761 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
762 			     enum x86emul_mode mode)
763 {
764 	ulong linear;
765 	int rc;
766 	unsigned max_size;
767 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
768 					   .ea = dst };
769 
770 	if (ctxt->op_bytes != sizeof(unsigned long))
771 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
772 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
773 	if (rc == X86EMUL_CONTINUE)
774 		ctxt->_eip = addr.ea;
775 	return rc;
776 }
777 
778 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
779 {
780 	return assign_eip(ctxt, dst, ctxt->mode);
781 }
782 
783 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
784 			  const struct desc_struct *cs_desc)
785 {
786 	enum x86emul_mode mode = ctxt->mode;
787 	int rc;
788 
789 #ifdef CONFIG_X86_64
790 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
791 		if (cs_desc->l) {
792 			u64 efer = 0;
793 
794 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
795 			if (efer & EFER_LMA)
796 				mode = X86EMUL_MODE_PROT64;
797 		} else
798 			mode = X86EMUL_MODE_PROT32; /* temporary value */
799 	}
800 #endif
801 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
802 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
803 	rc = assign_eip(ctxt, dst, mode);
804 	if (rc == X86EMUL_CONTINUE)
805 		ctxt->mode = mode;
806 	return rc;
807 }
808 
809 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
810 {
811 	return assign_eip_near(ctxt, ctxt->_eip + rel);
812 }
813 
814 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
815 			      struct segmented_address addr,
816 			      void *data,
817 			      unsigned size)
818 {
819 	int rc;
820 	ulong linear;
821 
822 	rc = linearize(ctxt, addr, size, false, &linear);
823 	if (rc != X86EMUL_CONTINUE)
824 		return rc;
825 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception);
826 }
827 
828 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
829 			       struct segmented_address addr,
830 			       void *data,
831 			       unsigned int size)
832 {
833 	int rc;
834 	ulong linear;
835 
836 	rc = linearize(ctxt, addr, size, true, &linear);
837 	if (rc != X86EMUL_CONTINUE)
838 		return rc;
839 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception);
840 }
841 
842 /*
843  * Prefetch the remaining bytes of the instruction without crossing page
844  * boundary if they are not in fetch_cache yet.
845  */
846 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
847 {
848 	int rc;
849 	unsigned size, max_size;
850 	unsigned long linear;
851 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
852 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
853 					   .ea = ctxt->eip + cur_size };
854 
855 	/*
856 	 * We do not know exactly how many bytes will be needed, and
857 	 * __linearize is expensive, so fetch as much as possible.  We
858 	 * just have to avoid going beyond the 15 byte limit, the end
859 	 * of the segment, or the end of the page.
860 	 *
861 	 * __linearize is called with size 0 so that it does not do any
862 	 * boundary check itself.  Instead, we use max_size to check
863 	 * against op_size.
864 	 */
865 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
866 			 &linear);
867 	if (unlikely(rc != X86EMUL_CONTINUE))
868 		return rc;
869 
870 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
871 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
872 
873 	/*
874 	 * One instruction can only straddle two pages,
875 	 * and one has been loaded at the beginning of
876 	 * x86_decode_insn.  So, if not enough bytes
877 	 * still, we must have hit the 15-byte boundary.
878 	 */
879 	if (unlikely(size < op_size))
880 		return emulate_gp(ctxt, 0);
881 
882 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
883 			      size, &ctxt->exception);
884 	if (unlikely(rc != X86EMUL_CONTINUE))
885 		return rc;
886 	ctxt->fetch.end += size;
887 	return X86EMUL_CONTINUE;
888 }
889 
890 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
891 					       unsigned size)
892 {
893 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
894 
895 	if (unlikely(done_size < size))
896 		return __do_insn_fetch_bytes(ctxt, size - done_size);
897 	else
898 		return X86EMUL_CONTINUE;
899 }
900 
901 /* Fetch next part of the instruction being emulated. */
902 #define insn_fetch(_type, _ctxt)					\
903 ({	_type _x;							\
904 									\
905 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
906 	if (rc != X86EMUL_CONTINUE)					\
907 		goto done;						\
908 	ctxt->_eip += sizeof(_type);					\
909 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
910 	ctxt->fetch.ptr += sizeof(_type);				\
911 	_x;								\
912 })
913 
914 #define insn_fetch_arr(_arr, _size, _ctxt)				\
915 ({									\
916 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
917 	if (rc != X86EMUL_CONTINUE)					\
918 		goto done;						\
919 	ctxt->_eip += (_size);						\
920 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
921 	ctxt->fetch.ptr += (_size);					\
922 })
923 
924 /*
925  * Given the 'reg' portion of a ModRM byte, and a register block, return a
926  * pointer into the block that addresses the relevant register.
927  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
928  */
929 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
930 			     int byteop)
931 {
932 	void *p;
933 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
934 
935 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
936 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
937 	else
938 		p = reg_rmw(ctxt, modrm_reg);
939 	return p;
940 }
941 
942 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
943 			   struct segmented_address addr,
944 			   u16 *size, unsigned long *address, int op_bytes)
945 {
946 	int rc;
947 
948 	if (op_bytes == 2)
949 		op_bytes = 3;
950 	*address = 0;
951 	rc = segmented_read_std(ctxt, addr, size, 2);
952 	if (rc != X86EMUL_CONTINUE)
953 		return rc;
954 	addr.ea += 2;
955 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
956 	return rc;
957 }
958 
959 FASTOP2(add);
960 FASTOP2(or);
961 FASTOP2(adc);
962 FASTOP2(sbb);
963 FASTOP2(and);
964 FASTOP2(sub);
965 FASTOP2(xor);
966 FASTOP2(cmp);
967 FASTOP2(test);
968 
969 FASTOP1SRC2(mul, mul_ex);
970 FASTOP1SRC2(imul, imul_ex);
971 FASTOP1SRC2EX(div, div_ex);
972 FASTOP1SRC2EX(idiv, idiv_ex);
973 
974 FASTOP3WCL(shld);
975 FASTOP3WCL(shrd);
976 
977 FASTOP2W(imul);
978 
979 FASTOP1(not);
980 FASTOP1(neg);
981 FASTOP1(inc);
982 FASTOP1(dec);
983 
984 FASTOP2CL(rol);
985 FASTOP2CL(ror);
986 FASTOP2CL(rcl);
987 FASTOP2CL(rcr);
988 FASTOP2CL(shl);
989 FASTOP2CL(shr);
990 FASTOP2CL(sar);
991 
992 FASTOP2W(bsf);
993 FASTOP2W(bsr);
994 FASTOP2W(bt);
995 FASTOP2W(bts);
996 FASTOP2W(btr);
997 FASTOP2W(btc);
998 
999 FASTOP2(xadd);
1000 
1001 FASTOP2R(cmp, cmp_r);
1002 
1003 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1004 {
1005 	/* If src is zero, do not writeback, but update flags */
1006 	if (ctxt->src.val == 0)
1007 		ctxt->dst.type = OP_NONE;
1008 	return fastop(ctxt, em_bsf);
1009 }
1010 
1011 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1012 {
1013 	/* If src is zero, do not writeback, but update flags */
1014 	if (ctxt->src.val == 0)
1015 		ctxt->dst.type = OP_NONE;
1016 	return fastop(ctxt, em_bsr);
1017 }
1018 
1019 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1020 {
1021 	u8 rc;
1022 	void (*fop)(void) = (void *)em_setcc + 4 * (condition & 0xf);
1023 
1024 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1025 	asm("push %[flags]; popf; " CALL_NOSPEC
1026 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1027 	return rc;
1028 }
1029 
1030 static void fetch_register_operand(struct operand *op)
1031 {
1032 	switch (op->bytes) {
1033 	case 1:
1034 		op->val = *(u8 *)op->addr.reg;
1035 		break;
1036 	case 2:
1037 		op->val = *(u16 *)op->addr.reg;
1038 		break;
1039 	case 4:
1040 		op->val = *(u32 *)op->addr.reg;
1041 		break;
1042 	case 8:
1043 		op->val = *(u64 *)op->addr.reg;
1044 		break;
1045 	}
1046 }
1047 
1048 static void read_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data, int reg)
1049 {
1050 	switch (reg) {
1051 	case 0: asm("movdqa %%xmm0, %0" : "=m"(*data)); break;
1052 	case 1: asm("movdqa %%xmm1, %0" : "=m"(*data)); break;
1053 	case 2: asm("movdqa %%xmm2, %0" : "=m"(*data)); break;
1054 	case 3: asm("movdqa %%xmm3, %0" : "=m"(*data)); break;
1055 	case 4: asm("movdqa %%xmm4, %0" : "=m"(*data)); break;
1056 	case 5: asm("movdqa %%xmm5, %0" : "=m"(*data)); break;
1057 	case 6: asm("movdqa %%xmm6, %0" : "=m"(*data)); break;
1058 	case 7: asm("movdqa %%xmm7, %0" : "=m"(*data)); break;
1059 #ifdef CONFIG_X86_64
1060 	case 8: asm("movdqa %%xmm8, %0" : "=m"(*data)); break;
1061 	case 9: asm("movdqa %%xmm9, %0" : "=m"(*data)); break;
1062 	case 10: asm("movdqa %%xmm10, %0" : "=m"(*data)); break;
1063 	case 11: asm("movdqa %%xmm11, %0" : "=m"(*data)); break;
1064 	case 12: asm("movdqa %%xmm12, %0" : "=m"(*data)); break;
1065 	case 13: asm("movdqa %%xmm13, %0" : "=m"(*data)); break;
1066 	case 14: asm("movdqa %%xmm14, %0" : "=m"(*data)); break;
1067 	case 15: asm("movdqa %%xmm15, %0" : "=m"(*data)); break;
1068 #endif
1069 	default: BUG();
1070 	}
1071 }
1072 
1073 static void write_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data,
1074 			  int reg)
1075 {
1076 	switch (reg) {
1077 	case 0: asm("movdqa %0, %%xmm0" : : "m"(*data)); break;
1078 	case 1: asm("movdqa %0, %%xmm1" : : "m"(*data)); break;
1079 	case 2: asm("movdqa %0, %%xmm2" : : "m"(*data)); break;
1080 	case 3: asm("movdqa %0, %%xmm3" : : "m"(*data)); break;
1081 	case 4: asm("movdqa %0, %%xmm4" : : "m"(*data)); break;
1082 	case 5: asm("movdqa %0, %%xmm5" : : "m"(*data)); break;
1083 	case 6: asm("movdqa %0, %%xmm6" : : "m"(*data)); break;
1084 	case 7: asm("movdqa %0, %%xmm7" : : "m"(*data)); break;
1085 #ifdef CONFIG_X86_64
1086 	case 8: asm("movdqa %0, %%xmm8" : : "m"(*data)); break;
1087 	case 9: asm("movdqa %0, %%xmm9" : : "m"(*data)); break;
1088 	case 10: asm("movdqa %0, %%xmm10" : : "m"(*data)); break;
1089 	case 11: asm("movdqa %0, %%xmm11" : : "m"(*data)); break;
1090 	case 12: asm("movdqa %0, %%xmm12" : : "m"(*data)); break;
1091 	case 13: asm("movdqa %0, %%xmm13" : : "m"(*data)); break;
1092 	case 14: asm("movdqa %0, %%xmm14" : : "m"(*data)); break;
1093 	case 15: asm("movdqa %0, %%xmm15" : : "m"(*data)); break;
1094 #endif
1095 	default: BUG();
1096 	}
1097 }
1098 
1099 static void read_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1100 {
1101 	switch (reg) {
1102 	case 0: asm("movq %%mm0, %0" : "=m"(*data)); break;
1103 	case 1: asm("movq %%mm1, %0" : "=m"(*data)); break;
1104 	case 2: asm("movq %%mm2, %0" : "=m"(*data)); break;
1105 	case 3: asm("movq %%mm3, %0" : "=m"(*data)); break;
1106 	case 4: asm("movq %%mm4, %0" : "=m"(*data)); break;
1107 	case 5: asm("movq %%mm5, %0" : "=m"(*data)); break;
1108 	case 6: asm("movq %%mm6, %0" : "=m"(*data)); break;
1109 	case 7: asm("movq %%mm7, %0" : "=m"(*data)); break;
1110 	default: BUG();
1111 	}
1112 }
1113 
1114 static void write_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1115 {
1116 	switch (reg) {
1117 	case 0: asm("movq %0, %%mm0" : : "m"(*data)); break;
1118 	case 1: asm("movq %0, %%mm1" : : "m"(*data)); break;
1119 	case 2: asm("movq %0, %%mm2" : : "m"(*data)); break;
1120 	case 3: asm("movq %0, %%mm3" : : "m"(*data)); break;
1121 	case 4: asm("movq %0, %%mm4" : : "m"(*data)); break;
1122 	case 5: asm("movq %0, %%mm5" : : "m"(*data)); break;
1123 	case 6: asm("movq %0, %%mm6" : : "m"(*data)); break;
1124 	case 7: asm("movq %0, %%mm7" : : "m"(*data)); break;
1125 	default: BUG();
1126 	}
1127 }
1128 
1129 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1130 {
1131 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1132 		return emulate_nm(ctxt);
1133 
1134 	asm volatile("fninit");
1135 	return X86EMUL_CONTINUE;
1136 }
1137 
1138 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1139 {
1140 	u16 fcw;
1141 
1142 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1143 		return emulate_nm(ctxt);
1144 
1145 	asm volatile("fnstcw %0": "+m"(fcw));
1146 
1147 	ctxt->dst.val = fcw;
1148 
1149 	return X86EMUL_CONTINUE;
1150 }
1151 
1152 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1153 {
1154 	u16 fsw;
1155 
1156 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1157 		return emulate_nm(ctxt);
1158 
1159 	asm volatile("fnstsw %0": "+m"(fsw));
1160 
1161 	ctxt->dst.val = fsw;
1162 
1163 	return X86EMUL_CONTINUE;
1164 }
1165 
1166 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1167 				    struct operand *op)
1168 {
1169 	unsigned reg = ctxt->modrm_reg;
1170 
1171 	if (!(ctxt->d & ModRM))
1172 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1173 
1174 	if (ctxt->d & Sse) {
1175 		op->type = OP_XMM;
1176 		op->bytes = 16;
1177 		op->addr.xmm = reg;
1178 		read_sse_reg(ctxt, &op->vec_val, reg);
1179 		return;
1180 	}
1181 	if (ctxt->d & Mmx) {
1182 		reg &= 7;
1183 		op->type = OP_MM;
1184 		op->bytes = 8;
1185 		op->addr.mm = reg;
1186 		return;
1187 	}
1188 
1189 	op->type = OP_REG;
1190 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1191 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1192 
1193 	fetch_register_operand(op);
1194 	op->orig_val = op->val;
1195 }
1196 
1197 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1198 {
1199 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1200 		ctxt->modrm_seg = VCPU_SREG_SS;
1201 }
1202 
1203 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1204 			struct operand *op)
1205 {
1206 	u8 sib;
1207 	int index_reg, base_reg, scale;
1208 	int rc = X86EMUL_CONTINUE;
1209 	ulong modrm_ea = 0;
1210 
1211 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1212 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1213 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1214 
1215 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1216 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1217 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1218 	ctxt->modrm_seg = VCPU_SREG_DS;
1219 
1220 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1221 		op->type = OP_REG;
1222 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1223 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1224 				ctxt->d & ByteOp);
1225 		if (ctxt->d & Sse) {
1226 			op->type = OP_XMM;
1227 			op->bytes = 16;
1228 			op->addr.xmm = ctxt->modrm_rm;
1229 			read_sse_reg(ctxt, &op->vec_val, ctxt->modrm_rm);
1230 			return rc;
1231 		}
1232 		if (ctxt->d & Mmx) {
1233 			op->type = OP_MM;
1234 			op->bytes = 8;
1235 			op->addr.mm = ctxt->modrm_rm & 7;
1236 			return rc;
1237 		}
1238 		fetch_register_operand(op);
1239 		return rc;
1240 	}
1241 
1242 	op->type = OP_MEM;
1243 
1244 	if (ctxt->ad_bytes == 2) {
1245 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1246 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1247 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1248 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1249 
1250 		/* 16-bit ModR/M decode. */
1251 		switch (ctxt->modrm_mod) {
1252 		case 0:
1253 			if (ctxt->modrm_rm == 6)
1254 				modrm_ea += insn_fetch(u16, ctxt);
1255 			break;
1256 		case 1:
1257 			modrm_ea += insn_fetch(s8, ctxt);
1258 			break;
1259 		case 2:
1260 			modrm_ea += insn_fetch(u16, ctxt);
1261 			break;
1262 		}
1263 		switch (ctxt->modrm_rm) {
1264 		case 0:
1265 			modrm_ea += bx + si;
1266 			break;
1267 		case 1:
1268 			modrm_ea += bx + di;
1269 			break;
1270 		case 2:
1271 			modrm_ea += bp + si;
1272 			break;
1273 		case 3:
1274 			modrm_ea += bp + di;
1275 			break;
1276 		case 4:
1277 			modrm_ea += si;
1278 			break;
1279 		case 5:
1280 			modrm_ea += di;
1281 			break;
1282 		case 6:
1283 			if (ctxt->modrm_mod != 0)
1284 				modrm_ea += bp;
1285 			break;
1286 		case 7:
1287 			modrm_ea += bx;
1288 			break;
1289 		}
1290 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1291 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1292 			ctxt->modrm_seg = VCPU_SREG_SS;
1293 		modrm_ea = (u16)modrm_ea;
1294 	} else {
1295 		/* 32/64-bit ModR/M decode. */
1296 		if ((ctxt->modrm_rm & 7) == 4) {
1297 			sib = insn_fetch(u8, ctxt);
1298 			index_reg |= (sib >> 3) & 7;
1299 			base_reg |= sib & 7;
1300 			scale = sib >> 6;
1301 
1302 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1303 				modrm_ea += insn_fetch(s32, ctxt);
1304 			else {
1305 				modrm_ea += reg_read(ctxt, base_reg);
1306 				adjust_modrm_seg(ctxt, base_reg);
1307 				/* Increment ESP on POP [ESP] */
1308 				if ((ctxt->d & IncSP) &&
1309 				    base_reg == VCPU_REGS_RSP)
1310 					modrm_ea += ctxt->op_bytes;
1311 			}
1312 			if (index_reg != 4)
1313 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1314 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1315 			modrm_ea += insn_fetch(s32, ctxt);
1316 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1317 				ctxt->rip_relative = 1;
1318 		} else {
1319 			base_reg = ctxt->modrm_rm;
1320 			modrm_ea += reg_read(ctxt, base_reg);
1321 			adjust_modrm_seg(ctxt, base_reg);
1322 		}
1323 		switch (ctxt->modrm_mod) {
1324 		case 1:
1325 			modrm_ea += insn_fetch(s8, ctxt);
1326 			break;
1327 		case 2:
1328 			modrm_ea += insn_fetch(s32, ctxt);
1329 			break;
1330 		}
1331 	}
1332 	op->addr.mem.ea = modrm_ea;
1333 	if (ctxt->ad_bytes != 8)
1334 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1335 
1336 done:
1337 	return rc;
1338 }
1339 
1340 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1341 		      struct operand *op)
1342 {
1343 	int rc = X86EMUL_CONTINUE;
1344 
1345 	op->type = OP_MEM;
1346 	switch (ctxt->ad_bytes) {
1347 	case 2:
1348 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1349 		break;
1350 	case 4:
1351 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1352 		break;
1353 	case 8:
1354 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1355 		break;
1356 	}
1357 done:
1358 	return rc;
1359 }
1360 
1361 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1362 {
1363 	long sv = 0, mask;
1364 
1365 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1366 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1367 
1368 		if (ctxt->src.bytes == 2)
1369 			sv = (s16)ctxt->src.val & (s16)mask;
1370 		else if (ctxt->src.bytes == 4)
1371 			sv = (s32)ctxt->src.val & (s32)mask;
1372 		else
1373 			sv = (s64)ctxt->src.val & (s64)mask;
1374 
1375 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1376 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1377 	}
1378 
1379 	/* only subword offset */
1380 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1381 }
1382 
1383 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1384 			 unsigned long addr, void *dest, unsigned size)
1385 {
1386 	int rc;
1387 	struct read_cache *mc = &ctxt->mem_read;
1388 
1389 	if (mc->pos < mc->end)
1390 		goto read_cached;
1391 
1392 	WARN_ON((mc->end + size) >= sizeof(mc->data));
1393 
1394 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1395 				      &ctxt->exception);
1396 	if (rc != X86EMUL_CONTINUE)
1397 		return rc;
1398 
1399 	mc->end += size;
1400 
1401 read_cached:
1402 	memcpy(dest, mc->data + mc->pos, size);
1403 	mc->pos += size;
1404 	return X86EMUL_CONTINUE;
1405 }
1406 
1407 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1408 			  struct segmented_address addr,
1409 			  void *data,
1410 			  unsigned size)
1411 {
1412 	int rc;
1413 	ulong linear;
1414 
1415 	rc = linearize(ctxt, addr, size, false, &linear);
1416 	if (rc != X86EMUL_CONTINUE)
1417 		return rc;
1418 	return read_emulated(ctxt, linear, data, size);
1419 }
1420 
1421 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1422 			   struct segmented_address addr,
1423 			   const void *data,
1424 			   unsigned size)
1425 {
1426 	int rc;
1427 	ulong linear;
1428 
1429 	rc = linearize(ctxt, addr, size, true, &linear);
1430 	if (rc != X86EMUL_CONTINUE)
1431 		return rc;
1432 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1433 					 &ctxt->exception);
1434 }
1435 
1436 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1437 			     struct segmented_address addr,
1438 			     const void *orig_data, const void *data,
1439 			     unsigned size)
1440 {
1441 	int rc;
1442 	ulong linear;
1443 
1444 	rc = linearize(ctxt, addr, size, true, &linear);
1445 	if (rc != X86EMUL_CONTINUE)
1446 		return rc;
1447 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1448 					   size, &ctxt->exception);
1449 }
1450 
1451 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1452 			   unsigned int size, unsigned short port,
1453 			   void *dest)
1454 {
1455 	struct read_cache *rc = &ctxt->io_read;
1456 
1457 	if (rc->pos == rc->end) { /* refill pio read ahead */
1458 		unsigned int in_page, n;
1459 		unsigned int count = ctxt->rep_prefix ?
1460 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1461 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1462 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1463 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1464 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1465 		if (n == 0)
1466 			n = 1;
1467 		rc->pos = rc->end = 0;
1468 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1469 			return 0;
1470 		rc->end = n * size;
1471 	}
1472 
1473 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1474 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1475 		ctxt->dst.data = rc->data + rc->pos;
1476 		ctxt->dst.type = OP_MEM_STR;
1477 		ctxt->dst.count = (rc->end - rc->pos) / size;
1478 		rc->pos = rc->end;
1479 	} else {
1480 		memcpy(dest, rc->data + rc->pos, size);
1481 		rc->pos += size;
1482 	}
1483 	return 1;
1484 }
1485 
1486 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1487 				     u16 index, struct desc_struct *desc)
1488 {
1489 	struct desc_ptr dt;
1490 	ulong addr;
1491 
1492 	ctxt->ops->get_idt(ctxt, &dt);
1493 
1494 	if (dt.size < index * 8 + 7)
1495 		return emulate_gp(ctxt, index << 3 | 0x2);
1496 
1497 	addr = dt.address + index * 8;
1498 	return ctxt->ops->read_std(ctxt, addr, desc, sizeof *desc,
1499 				   &ctxt->exception);
1500 }
1501 
1502 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1503 				     u16 selector, struct desc_ptr *dt)
1504 {
1505 	const struct x86_emulate_ops *ops = ctxt->ops;
1506 	u32 base3 = 0;
1507 
1508 	if (selector & 1 << 2) {
1509 		struct desc_struct desc;
1510 		u16 sel;
1511 
1512 		memset (dt, 0, sizeof *dt);
1513 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1514 				      VCPU_SREG_LDTR))
1515 			return;
1516 
1517 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1518 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1519 	} else
1520 		ops->get_gdt(ctxt, dt);
1521 }
1522 
1523 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1524 			      u16 selector, ulong *desc_addr_p)
1525 {
1526 	struct desc_ptr dt;
1527 	u16 index = selector >> 3;
1528 	ulong addr;
1529 
1530 	get_descriptor_table_ptr(ctxt, selector, &dt);
1531 
1532 	if (dt.size < index * 8 + 7)
1533 		return emulate_gp(ctxt, selector & 0xfffc);
1534 
1535 	addr = dt.address + index * 8;
1536 
1537 #ifdef CONFIG_X86_64
1538 	if (addr >> 32 != 0) {
1539 		u64 efer = 0;
1540 
1541 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1542 		if (!(efer & EFER_LMA))
1543 			addr &= (u32)-1;
1544 	}
1545 #endif
1546 
1547 	*desc_addr_p = addr;
1548 	return X86EMUL_CONTINUE;
1549 }
1550 
1551 /* allowed just for 8 bytes segments */
1552 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1553 				   u16 selector, struct desc_struct *desc,
1554 				   ulong *desc_addr_p)
1555 {
1556 	int rc;
1557 
1558 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1559 	if (rc != X86EMUL_CONTINUE)
1560 		return rc;
1561 
1562 	return ctxt->ops->read_std(ctxt, *desc_addr_p, desc, sizeof(*desc),
1563 				   &ctxt->exception);
1564 }
1565 
1566 /* allowed just for 8 bytes segments */
1567 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1568 				    u16 selector, struct desc_struct *desc)
1569 {
1570 	int rc;
1571 	ulong addr;
1572 
1573 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1574 	if (rc != X86EMUL_CONTINUE)
1575 		return rc;
1576 
1577 	return ctxt->ops->write_std(ctxt, addr, desc, sizeof *desc,
1578 				    &ctxt->exception);
1579 }
1580 
1581 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1582 				     u16 selector, int seg, u8 cpl,
1583 				     enum x86_transfer_type transfer,
1584 				     struct desc_struct *desc)
1585 {
1586 	struct desc_struct seg_desc, old_desc;
1587 	u8 dpl, rpl;
1588 	unsigned err_vec = GP_VECTOR;
1589 	u32 err_code = 0;
1590 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1591 	ulong desc_addr;
1592 	int ret;
1593 	u16 dummy;
1594 	u32 base3 = 0;
1595 
1596 	memset(&seg_desc, 0, sizeof seg_desc);
1597 
1598 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1599 		/* set real mode segment descriptor (keep limit etc. for
1600 		 * unreal mode) */
1601 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1602 		set_desc_base(&seg_desc, selector << 4);
1603 		goto load;
1604 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1605 		/* VM86 needs a clean new segment descriptor */
1606 		set_desc_base(&seg_desc, selector << 4);
1607 		set_desc_limit(&seg_desc, 0xffff);
1608 		seg_desc.type = 3;
1609 		seg_desc.p = 1;
1610 		seg_desc.s = 1;
1611 		seg_desc.dpl = 3;
1612 		goto load;
1613 	}
1614 
1615 	rpl = selector & 3;
1616 
1617 	/* TR should be in GDT only */
1618 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1619 		goto exception;
1620 
1621 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1622 	if (null_selector) {
1623 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1624 			goto exception;
1625 
1626 		if (seg == VCPU_SREG_SS) {
1627 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1628 				goto exception;
1629 
1630 			/*
1631 			 * ctxt->ops->set_segment expects the CPL to be in
1632 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1633 			 */
1634 			seg_desc.type = 3;
1635 			seg_desc.p = 1;
1636 			seg_desc.s = 1;
1637 			seg_desc.dpl = cpl;
1638 			seg_desc.d = 1;
1639 			seg_desc.g = 1;
1640 		}
1641 
1642 		/* Skip all following checks */
1643 		goto load;
1644 	}
1645 
1646 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1647 	if (ret != X86EMUL_CONTINUE)
1648 		return ret;
1649 
1650 	err_code = selector & 0xfffc;
1651 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1652 							   GP_VECTOR;
1653 
1654 	/* can't load system descriptor into segment selector */
1655 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1656 		if (transfer == X86_TRANSFER_CALL_JMP)
1657 			return X86EMUL_UNHANDLEABLE;
1658 		goto exception;
1659 	}
1660 
1661 	if (!seg_desc.p) {
1662 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1663 		goto exception;
1664 	}
1665 
1666 	dpl = seg_desc.dpl;
1667 
1668 	switch (seg) {
1669 	case VCPU_SREG_SS:
1670 		/*
1671 		 * segment is not a writable data segment or segment
1672 		 * selector's RPL != CPL or segment selector's RPL != CPL
1673 		 */
1674 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1675 			goto exception;
1676 		break;
1677 	case VCPU_SREG_CS:
1678 		if (!(seg_desc.type & 8))
1679 			goto exception;
1680 
1681 		if (seg_desc.type & 4) {
1682 			/* conforming */
1683 			if (dpl > cpl)
1684 				goto exception;
1685 		} else {
1686 			/* nonconforming */
1687 			if (rpl > cpl || dpl != cpl)
1688 				goto exception;
1689 		}
1690 		/* in long-mode d/b must be clear if l is set */
1691 		if (seg_desc.d && seg_desc.l) {
1692 			u64 efer = 0;
1693 
1694 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1695 			if (efer & EFER_LMA)
1696 				goto exception;
1697 		}
1698 
1699 		/* CS(RPL) <- CPL */
1700 		selector = (selector & 0xfffc) | cpl;
1701 		break;
1702 	case VCPU_SREG_TR:
1703 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1704 			goto exception;
1705 		old_desc = seg_desc;
1706 		seg_desc.type |= 2; /* busy */
1707 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1708 						  sizeof(seg_desc), &ctxt->exception);
1709 		if (ret != X86EMUL_CONTINUE)
1710 			return ret;
1711 		break;
1712 	case VCPU_SREG_LDTR:
1713 		if (seg_desc.s || seg_desc.type != 2)
1714 			goto exception;
1715 		break;
1716 	default: /*  DS, ES, FS, or GS */
1717 		/*
1718 		 * segment is not a data or readable code segment or
1719 		 * ((segment is a data or nonconforming code segment)
1720 		 * and (both RPL and CPL > DPL))
1721 		 */
1722 		if ((seg_desc.type & 0xa) == 0x8 ||
1723 		    (((seg_desc.type & 0xc) != 0xc) &&
1724 		     (rpl > dpl && cpl > dpl)))
1725 			goto exception;
1726 		break;
1727 	}
1728 
1729 	if (seg_desc.s) {
1730 		/* mark segment as accessed */
1731 		if (!(seg_desc.type & 1)) {
1732 			seg_desc.type |= 1;
1733 			ret = write_segment_descriptor(ctxt, selector,
1734 						       &seg_desc);
1735 			if (ret != X86EMUL_CONTINUE)
1736 				return ret;
1737 		}
1738 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1739 		ret = ctxt->ops->read_std(ctxt, desc_addr+8, &base3,
1740 				sizeof(base3), &ctxt->exception);
1741 		if (ret != X86EMUL_CONTINUE)
1742 			return ret;
1743 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1744 				((u64)base3 << 32), ctxt))
1745 			return emulate_gp(ctxt, 0);
1746 	}
1747 load:
1748 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1749 	if (desc)
1750 		*desc = seg_desc;
1751 	return X86EMUL_CONTINUE;
1752 exception:
1753 	return emulate_exception(ctxt, err_vec, err_code, true);
1754 }
1755 
1756 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1757 				   u16 selector, int seg)
1758 {
1759 	u8 cpl = ctxt->ops->cpl(ctxt);
1760 
1761 	/*
1762 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1763 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1764 	 * but it's wrong).
1765 	 *
1766 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1767 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1768 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1769 	 * and only forbid it here.
1770 	 */
1771 	if (seg == VCPU_SREG_SS && selector == 3 &&
1772 	    ctxt->mode == X86EMUL_MODE_PROT64)
1773 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1774 
1775 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1776 					 X86_TRANSFER_NONE, NULL);
1777 }
1778 
1779 static void write_register_operand(struct operand *op)
1780 {
1781 	return assign_register(op->addr.reg, op->val, op->bytes);
1782 }
1783 
1784 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1785 {
1786 	switch (op->type) {
1787 	case OP_REG:
1788 		write_register_operand(op);
1789 		break;
1790 	case OP_MEM:
1791 		if (ctxt->lock_prefix)
1792 			return segmented_cmpxchg(ctxt,
1793 						 op->addr.mem,
1794 						 &op->orig_val,
1795 						 &op->val,
1796 						 op->bytes);
1797 		else
1798 			return segmented_write(ctxt,
1799 					       op->addr.mem,
1800 					       &op->val,
1801 					       op->bytes);
1802 		break;
1803 	case OP_MEM_STR:
1804 		return segmented_write(ctxt,
1805 				       op->addr.mem,
1806 				       op->data,
1807 				       op->bytes * op->count);
1808 		break;
1809 	case OP_XMM:
1810 		write_sse_reg(ctxt, &op->vec_val, op->addr.xmm);
1811 		break;
1812 	case OP_MM:
1813 		write_mmx_reg(ctxt, &op->mm_val, op->addr.mm);
1814 		break;
1815 	case OP_NONE:
1816 		/* no writeback */
1817 		break;
1818 	default:
1819 		break;
1820 	}
1821 	return X86EMUL_CONTINUE;
1822 }
1823 
1824 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1825 {
1826 	struct segmented_address addr;
1827 
1828 	rsp_increment(ctxt, -bytes);
1829 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1830 	addr.seg = VCPU_SREG_SS;
1831 
1832 	return segmented_write(ctxt, addr, data, bytes);
1833 }
1834 
1835 static int em_push(struct x86_emulate_ctxt *ctxt)
1836 {
1837 	/* Disable writeback. */
1838 	ctxt->dst.type = OP_NONE;
1839 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1840 }
1841 
1842 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1843 		       void *dest, int len)
1844 {
1845 	int rc;
1846 	struct segmented_address addr;
1847 
1848 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1849 	addr.seg = VCPU_SREG_SS;
1850 	rc = segmented_read(ctxt, addr, dest, len);
1851 	if (rc != X86EMUL_CONTINUE)
1852 		return rc;
1853 
1854 	rsp_increment(ctxt, len);
1855 	return rc;
1856 }
1857 
1858 static int em_pop(struct x86_emulate_ctxt *ctxt)
1859 {
1860 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1861 }
1862 
1863 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1864 			void *dest, int len)
1865 {
1866 	int rc;
1867 	unsigned long val, change_mask;
1868 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1869 	int cpl = ctxt->ops->cpl(ctxt);
1870 
1871 	rc = emulate_pop(ctxt, &val, len);
1872 	if (rc != X86EMUL_CONTINUE)
1873 		return rc;
1874 
1875 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1876 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1877 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1878 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1879 
1880 	switch(ctxt->mode) {
1881 	case X86EMUL_MODE_PROT64:
1882 	case X86EMUL_MODE_PROT32:
1883 	case X86EMUL_MODE_PROT16:
1884 		if (cpl == 0)
1885 			change_mask |= X86_EFLAGS_IOPL;
1886 		if (cpl <= iopl)
1887 			change_mask |= X86_EFLAGS_IF;
1888 		break;
1889 	case X86EMUL_MODE_VM86:
1890 		if (iopl < 3)
1891 			return emulate_gp(ctxt, 0);
1892 		change_mask |= X86_EFLAGS_IF;
1893 		break;
1894 	default: /* real mode */
1895 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1896 		break;
1897 	}
1898 
1899 	*(unsigned long *)dest =
1900 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1901 
1902 	return rc;
1903 }
1904 
1905 static int em_popf(struct x86_emulate_ctxt *ctxt)
1906 {
1907 	ctxt->dst.type = OP_REG;
1908 	ctxt->dst.addr.reg = &ctxt->eflags;
1909 	ctxt->dst.bytes = ctxt->op_bytes;
1910 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1911 }
1912 
1913 static int em_enter(struct x86_emulate_ctxt *ctxt)
1914 {
1915 	int rc;
1916 	unsigned frame_size = ctxt->src.val;
1917 	unsigned nesting_level = ctxt->src2.val & 31;
1918 	ulong rbp;
1919 
1920 	if (nesting_level)
1921 		return X86EMUL_UNHANDLEABLE;
1922 
1923 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1924 	rc = push(ctxt, &rbp, stack_size(ctxt));
1925 	if (rc != X86EMUL_CONTINUE)
1926 		return rc;
1927 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1928 		      stack_mask(ctxt));
1929 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1930 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1931 		      stack_mask(ctxt));
1932 	return X86EMUL_CONTINUE;
1933 }
1934 
1935 static int em_leave(struct x86_emulate_ctxt *ctxt)
1936 {
1937 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1938 		      stack_mask(ctxt));
1939 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1940 }
1941 
1942 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1943 {
1944 	int seg = ctxt->src2.val;
1945 
1946 	ctxt->src.val = get_segment_selector(ctxt, seg);
1947 	if (ctxt->op_bytes == 4) {
1948 		rsp_increment(ctxt, -2);
1949 		ctxt->op_bytes = 2;
1950 	}
1951 
1952 	return em_push(ctxt);
1953 }
1954 
1955 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1956 {
1957 	int seg = ctxt->src2.val;
1958 	unsigned long selector;
1959 	int rc;
1960 
1961 	rc = emulate_pop(ctxt, &selector, 2);
1962 	if (rc != X86EMUL_CONTINUE)
1963 		return rc;
1964 
1965 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1966 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1967 	if (ctxt->op_bytes > 2)
1968 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1969 
1970 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1971 	return rc;
1972 }
1973 
1974 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1975 {
1976 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1977 	int rc = X86EMUL_CONTINUE;
1978 	int reg = VCPU_REGS_RAX;
1979 
1980 	while (reg <= VCPU_REGS_RDI) {
1981 		(reg == VCPU_REGS_RSP) ?
1982 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1983 
1984 		rc = em_push(ctxt);
1985 		if (rc != X86EMUL_CONTINUE)
1986 			return rc;
1987 
1988 		++reg;
1989 	}
1990 
1991 	return rc;
1992 }
1993 
1994 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1995 {
1996 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1997 	return em_push(ctxt);
1998 }
1999 
2000 static int em_popa(struct x86_emulate_ctxt *ctxt)
2001 {
2002 	int rc = X86EMUL_CONTINUE;
2003 	int reg = VCPU_REGS_RDI;
2004 	u32 val;
2005 
2006 	while (reg >= VCPU_REGS_RAX) {
2007 		if (reg == VCPU_REGS_RSP) {
2008 			rsp_increment(ctxt, ctxt->op_bytes);
2009 			--reg;
2010 		}
2011 
2012 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2013 		if (rc != X86EMUL_CONTINUE)
2014 			break;
2015 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2016 		--reg;
2017 	}
2018 	return rc;
2019 }
2020 
2021 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2022 {
2023 	const struct x86_emulate_ops *ops = ctxt->ops;
2024 	int rc;
2025 	struct desc_ptr dt;
2026 	gva_t cs_addr;
2027 	gva_t eip_addr;
2028 	u16 cs, eip;
2029 
2030 	/* TODO: Add limit checks */
2031 	ctxt->src.val = ctxt->eflags;
2032 	rc = em_push(ctxt);
2033 	if (rc != X86EMUL_CONTINUE)
2034 		return rc;
2035 
2036 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2037 
2038 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2039 	rc = em_push(ctxt);
2040 	if (rc != X86EMUL_CONTINUE)
2041 		return rc;
2042 
2043 	ctxt->src.val = ctxt->_eip;
2044 	rc = em_push(ctxt);
2045 	if (rc != X86EMUL_CONTINUE)
2046 		return rc;
2047 
2048 	ops->get_idt(ctxt, &dt);
2049 
2050 	eip_addr = dt.address + (irq << 2);
2051 	cs_addr = dt.address + (irq << 2) + 2;
2052 
2053 	rc = ops->read_std(ctxt, cs_addr, &cs, 2, &ctxt->exception);
2054 	if (rc != X86EMUL_CONTINUE)
2055 		return rc;
2056 
2057 	rc = ops->read_std(ctxt, eip_addr, &eip, 2, &ctxt->exception);
2058 	if (rc != X86EMUL_CONTINUE)
2059 		return rc;
2060 
2061 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2062 	if (rc != X86EMUL_CONTINUE)
2063 		return rc;
2064 
2065 	ctxt->_eip = eip;
2066 
2067 	return rc;
2068 }
2069 
2070 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2071 {
2072 	int rc;
2073 
2074 	invalidate_registers(ctxt);
2075 	rc = __emulate_int_real(ctxt, irq);
2076 	if (rc == X86EMUL_CONTINUE)
2077 		writeback_registers(ctxt);
2078 	return rc;
2079 }
2080 
2081 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2082 {
2083 	switch(ctxt->mode) {
2084 	case X86EMUL_MODE_REAL:
2085 		return __emulate_int_real(ctxt, irq);
2086 	case X86EMUL_MODE_VM86:
2087 	case X86EMUL_MODE_PROT16:
2088 	case X86EMUL_MODE_PROT32:
2089 	case X86EMUL_MODE_PROT64:
2090 	default:
2091 		/* Protected mode interrupts unimplemented yet */
2092 		return X86EMUL_UNHANDLEABLE;
2093 	}
2094 }
2095 
2096 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2097 {
2098 	int rc = X86EMUL_CONTINUE;
2099 	unsigned long temp_eip = 0;
2100 	unsigned long temp_eflags = 0;
2101 	unsigned long cs = 0;
2102 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2103 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2104 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2105 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2106 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2107 			     X86_EFLAGS_FIXED;
2108 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2109 				  X86_EFLAGS_VIP;
2110 
2111 	/* TODO: Add stack limit check */
2112 
2113 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2114 
2115 	if (rc != X86EMUL_CONTINUE)
2116 		return rc;
2117 
2118 	if (temp_eip & ~0xffff)
2119 		return emulate_gp(ctxt, 0);
2120 
2121 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2122 
2123 	if (rc != X86EMUL_CONTINUE)
2124 		return rc;
2125 
2126 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2127 
2128 	if (rc != X86EMUL_CONTINUE)
2129 		return rc;
2130 
2131 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2132 
2133 	if (rc != X86EMUL_CONTINUE)
2134 		return rc;
2135 
2136 	ctxt->_eip = temp_eip;
2137 
2138 	if (ctxt->op_bytes == 4)
2139 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2140 	else if (ctxt->op_bytes == 2) {
2141 		ctxt->eflags &= ~0xffff;
2142 		ctxt->eflags |= temp_eflags;
2143 	}
2144 
2145 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2146 	ctxt->eflags |= X86_EFLAGS_FIXED;
2147 	ctxt->ops->set_nmi_mask(ctxt, false);
2148 
2149 	return rc;
2150 }
2151 
2152 static int em_iret(struct x86_emulate_ctxt *ctxt)
2153 {
2154 	switch(ctxt->mode) {
2155 	case X86EMUL_MODE_REAL:
2156 		return emulate_iret_real(ctxt);
2157 	case X86EMUL_MODE_VM86:
2158 	case X86EMUL_MODE_PROT16:
2159 	case X86EMUL_MODE_PROT32:
2160 	case X86EMUL_MODE_PROT64:
2161 	default:
2162 		/* iret from protected mode unimplemented yet */
2163 		return X86EMUL_UNHANDLEABLE;
2164 	}
2165 }
2166 
2167 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2168 {
2169 	int rc;
2170 	unsigned short sel;
2171 	struct desc_struct new_desc;
2172 	u8 cpl = ctxt->ops->cpl(ctxt);
2173 
2174 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2175 
2176 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2177 				       X86_TRANSFER_CALL_JMP,
2178 				       &new_desc);
2179 	if (rc != X86EMUL_CONTINUE)
2180 		return rc;
2181 
2182 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2183 	/* Error handling is not implemented. */
2184 	if (rc != X86EMUL_CONTINUE)
2185 		return X86EMUL_UNHANDLEABLE;
2186 
2187 	return rc;
2188 }
2189 
2190 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2191 {
2192 	return assign_eip_near(ctxt, ctxt->src.val);
2193 }
2194 
2195 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2196 {
2197 	int rc;
2198 	long int old_eip;
2199 
2200 	old_eip = ctxt->_eip;
2201 	rc = assign_eip_near(ctxt, ctxt->src.val);
2202 	if (rc != X86EMUL_CONTINUE)
2203 		return rc;
2204 	ctxt->src.val = old_eip;
2205 	rc = em_push(ctxt);
2206 	return rc;
2207 }
2208 
2209 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2210 {
2211 	u64 old = ctxt->dst.orig_val64;
2212 
2213 	if (ctxt->dst.bytes == 16)
2214 		return X86EMUL_UNHANDLEABLE;
2215 
2216 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2217 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2218 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2219 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2220 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2221 	} else {
2222 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2223 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2224 
2225 		ctxt->eflags |= X86_EFLAGS_ZF;
2226 	}
2227 	return X86EMUL_CONTINUE;
2228 }
2229 
2230 static int em_ret(struct x86_emulate_ctxt *ctxt)
2231 {
2232 	int rc;
2233 	unsigned long eip;
2234 
2235 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2236 	if (rc != X86EMUL_CONTINUE)
2237 		return rc;
2238 
2239 	return assign_eip_near(ctxt, eip);
2240 }
2241 
2242 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2243 {
2244 	int rc;
2245 	unsigned long eip, cs;
2246 	int cpl = ctxt->ops->cpl(ctxt);
2247 	struct desc_struct new_desc;
2248 
2249 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2250 	if (rc != X86EMUL_CONTINUE)
2251 		return rc;
2252 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2253 	if (rc != X86EMUL_CONTINUE)
2254 		return rc;
2255 	/* Outer-privilege level return is not implemented */
2256 	if (ctxt->mode >= X86EMUL_MODE_PROT16 && (cs & 3) > cpl)
2257 		return X86EMUL_UNHANDLEABLE;
2258 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2259 				       X86_TRANSFER_RET,
2260 				       &new_desc);
2261 	if (rc != X86EMUL_CONTINUE)
2262 		return rc;
2263 	rc = assign_eip_far(ctxt, eip, &new_desc);
2264 	/* Error handling is not implemented. */
2265 	if (rc != X86EMUL_CONTINUE)
2266 		return X86EMUL_UNHANDLEABLE;
2267 
2268 	return rc;
2269 }
2270 
2271 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2272 {
2273         int rc;
2274 
2275         rc = em_ret_far(ctxt);
2276         if (rc != X86EMUL_CONTINUE)
2277                 return rc;
2278         rsp_increment(ctxt, ctxt->src.val);
2279         return X86EMUL_CONTINUE;
2280 }
2281 
2282 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2283 {
2284 	/* Save real source value, then compare EAX against destination. */
2285 	ctxt->dst.orig_val = ctxt->dst.val;
2286 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2287 	ctxt->src.orig_val = ctxt->src.val;
2288 	ctxt->src.val = ctxt->dst.orig_val;
2289 	fastop(ctxt, em_cmp);
2290 
2291 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2292 		/* Success: write back to memory; no update of EAX */
2293 		ctxt->src.type = OP_NONE;
2294 		ctxt->dst.val = ctxt->src.orig_val;
2295 	} else {
2296 		/* Failure: write the value we saw to EAX. */
2297 		ctxt->src.type = OP_REG;
2298 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2299 		ctxt->src.val = ctxt->dst.orig_val;
2300 		/* Create write-cycle to dest by writing the same value */
2301 		ctxt->dst.val = ctxt->dst.orig_val;
2302 	}
2303 	return X86EMUL_CONTINUE;
2304 }
2305 
2306 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2307 {
2308 	int seg = ctxt->src2.val;
2309 	unsigned short sel;
2310 	int rc;
2311 
2312 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2313 
2314 	rc = load_segment_descriptor(ctxt, sel, seg);
2315 	if (rc != X86EMUL_CONTINUE)
2316 		return rc;
2317 
2318 	ctxt->dst.val = ctxt->src.val;
2319 	return rc;
2320 }
2321 
2322 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2323 {
2324 	u32 eax, ebx, ecx, edx;
2325 
2326 	eax = 0x80000001;
2327 	ecx = 0;
2328 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
2329 	return edx & bit(X86_FEATURE_LM);
2330 }
2331 
2332 #define GET_SMSTATE(type, smbase, offset)				  \
2333 	({								  \
2334 	 type __val;							  \
2335 	 int r = ctxt->ops->read_phys(ctxt, smbase + offset, &__val,      \
2336 				      sizeof(__val));			  \
2337 	 if (r != X86EMUL_CONTINUE)					  \
2338 		 return X86EMUL_UNHANDLEABLE;				  \
2339 	 __val;								  \
2340 	})
2341 
2342 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2343 {
2344 	desc->g    = (flags >> 23) & 1;
2345 	desc->d    = (flags >> 22) & 1;
2346 	desc->l    = (flags >> 21) & 1;
2347 	desc->avl  = (flags >> 20) & 1;
2348 	desc->p    = (flags >> 15) & 1;
2349 	desc->dpl  = (flags >> 13) & 3;
2350 	desc->s    = (flags >> 12) & 1;
2351 	desc->type = (flags >>  8) & 15;
2352 }
2353 
2354 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, u64 smbase, int n)
2355 {
2356 	struct desc_struct desc;
2357 	int offset;
2358 	u16 selector;
2359 
2360 	selector = GET_SMSTATE(u32, smbase, 0x7fa8 + n * 4);
2361 
2362 	if (n < 3)
2363 		offset = 0x7f84 + n * 12;
2364 	else
2365 		offset = 0x7f2c + (n - 3) * 12;
2366 
2367 	set_desc_base(&desc,      GET_SMSTATE(u32, smbase, offset + 8));
2368 	set_desc_limit(&desc,     GET_SMSTATE(u32, smbase, offset + 4));
2369 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smbase, offset));
2370 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2371 	return X86EMUL_CONTINUE;
2372 }
2373 
2374 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, u64 smbase, int n)
2375 {
2376 	struct desc_struct desc;
2377 	int offset;
2378 	u16 selector;
2379 	u32 base3;
2380 
2381 	offset = 0x7e00 + n * 16;
2382 
2383 	selector =                GET_SMSTATE(u16, smbase, offset);
2384 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smbase, offset + 2) << 8);
2385 	set_desc_limit(&desc,     GET_SMSTATE(u32, smbase, offset + 4));
2386 	set_desc_base(&desc,      GET_SMSTATE(u32, smbase, offset + 8));
2387 	base3 =                   GET_SMSTATE(u32, smbase, offset + 12);
2388 
2389 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2390 	return X86EMUL_CONTINUE;
2391 }
2392 
2393 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2394 				    u64 cr0, u64 cr3, u64 cr4)
2395 {
2396 	int bad;
2397 	u64 pcid;
2398 
2399 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
2400 	pcid = 0;
2401 	if (cr4 & X86_CR4_PCIDE) {
2402 		pcid = cr3 & 0xfff;
2403 		cr3 &= ~0xfff;
2404 	}
2405 
2406 	bad = ctxt->ops->set_cr(ctxt, 3, cr3);
2407 	if (bad)
2408 		return X86EMUL_UNHANDLEABLE;
2409 
2410 	/*
2411 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2412 	 * Then enable protected mode.	However, PCID cannot be enabled
2413 	 * if EFER.LMA=0, so set it separately.
2414 	 */
2415 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2416 	if (bad)
2417 		return X86EMUL_UNHANDLEABLE;
2418 
2419 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2420 	if (bad)
2421 		return X86EMUL_UNHANDLEABLE;
2422 
2423 	if (cr4 & X86_CR4_PCIDE) {
2424 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2425 		if (bad)
2426 			return X86EMUL_UNHANDLEABLE;
2427 		if (pcid) {
2428 			bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid);
2429 			if (bad)
2430 				return X86EMUL_UNHANDLEABLE;
2431 		}
2432 
2433 	}
2434 
2435 	return X86EMUL_CONTINUE;
2436 }
2437 
2438 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt, u64 smbase)
2439 {
2440 	struct desc_struct desc;
2441 	struct desc_ptr dt;
2442 	u16 selector;
2443 	u32 val, cr0, cr3, cr4;
2444 	int i;
2445 
2446 	cr0 =                      GET_SMSTATE(u32, smbase, 0x7ffc);
2447 	cr3 =                      GET_SMSTATE(u32, smbase, 0x7ff8);
2448 	ctxt->eflags =             GET_SMSTATE(u32, smbase, 0x7ff4) | X86_EFLAGS_FIXED;
2449 	ctxt->_eip =               GET_SMSTATE(u32, smbase, 0x7ff0);
2450 
2451 	for (i = 0; i < 8; i++)
2452 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smbase, 0x7fd0 + i * 4);
2453 
2454 	val = GET_SMSTATE(u32, smbase, 0x7fcc);
2455 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2456 	val = GET_SMSTATE(u32, smbase, 0x7fc8);
2457 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2458 
2459 	selector =                 GET_SMSTATE(u32, smbase, 0x7fc4);
2460 	set_desc_base(&desc,       GET_SMSTATE(u32, smbase, 0x7f64));
2461 	set_desc_limit(&desc,      GET_SMSTATE(u32, smbase, 0x7f60));
2462 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smbase, 0x7f5c));
2463 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2464 
2465 	selector =                 GET_SMSTATE(u32, smbase, 0x7fc0);
2466 	set_desc_base(&desc,       GET_SMSTATE(u32, smbase, 0x7f80));
2467 	set_desc_limit(&desc,      GET_SMSTATE(u32, smbase, 0x7f7c));
2468 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smbase, 0x7f78));
2469 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2470 
2471 	dt.address =               GET_SMSTATE(u32, smbase, 0x7f74);
2472 	dt.size =                  GET_SMSTATE(u32, smbase, 0x7f70);
2473 	ctxt->ops->set_gdt(ctxt, &dt);
2474 
2475 	dt.address =               GET_SMSTATE(u32, smbase, 0x7f58);
2476 	dt.size =                  GET_SMSTATE(u32, smbase, 0x7f54);
2477 	ctxt->ops->set_idt(ctxt, &dt);
2478 
2479 	for (i = 0; i < 6; i++) {
2480 		int r = rsm_load_seg_32(ctxt, smbase, i);
2481 		if (r != X86EMUL_CONTINUE)
2482 			return r;
2483 	}
2484 
2485 	cr4 = GET_SMSTATE(u32, smbase, 0x7f14);
2486 
2487 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smbase, 0x7ef8));
2488 
2489 	return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2490 }
2491 
2492 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt, u64 smbase)
2493 {
2494 	struct desc_struct desc;
2495 	struct desc_ptr dt;
2496 	u64 val, cr0, cr3, cr4;
2497 	u32 base3;
2498 	u16 selector;
2499 	int i, r;
2500 
2501 	for (i = 0; i < 16; i++)
2502 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smbase, 0x7ff8 - i * 8);
2503 
2504 	ctxt->_eip   = GET_SMSTATE(u64, smbase, 0x7f78);
2505 	ctxt->eflags = GET_SMSTATE(u32, smbase, 0x7f70) | X86_EFLAGS_FIXED;
2506 
2507 	val = GET_SMSTATE(u32, smbase, 0x7f68);
2508 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2509 	val = GET_SMSTATE(u32, smbase, 0x7f60);
2510 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2511 
2512 	cr0 =                       GET_SMSTATE(u64, smbase, 0x7f58);
2513 	cr3 =                       GET_SMSTATE(u64, smbase, 0x7f50);
2514 	cr4 =                       GET_SMSTATE(u64, smbase, 0x7f48);
2515 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smbase, 0x7f00));
2516 	val =                       GET_SMSTATE(u64, smbase, 0x7ed0);
2517 	ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA);
2518 
2519 	selector =                  GET_SMSTATE(u32, smbase, 0x7e90);
2520 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smbase, 0x7e92) << 8);
2521 	set_desc_limit(&desc,       GET_SMSTATE(u32, smbase, 0x7e94));
2522 	set_desc_base(&desc,        GET_SMSTATE(u32, smbase, 0x7e98));
2523 	base3 =                     GET_SMSTATE(u32, smbase, 0x7e9c);
2524 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2525 
2526 	dt.size =                   GET_SMSTATE(u32, smbase, 0x7e84);
2527 	dt.address =                GET_SMSTATE(u64, smbase, 0x7e88);
2528 	ctxt->ops->set_idt(ctxt, &dt);
2529 
2530 	selector =                  GET_SMSTATE(u32, smbase, 0x7e70);
2531 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smbase, 0x7e72) << 8);
2532 	set_desc_limit(&desc,       GET_SMSTATE(u32, smbase, 0x7e74));
2533 	set_desc_base(&desc,        GET_SMSTATE(u32, smbase, 0x7e78));
2534 	base3 =                     GET_SMSTATE(u32, smbase, 0x7e7c);
2535 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2536 
2537 	dt.size =                   GET_SMSTATE(u32, smbase, 0x7e64);
2538 	dt.address =                GET_SMSTATE(u64, smbase, 0x7e68);
2539 	ctxt->ops->set_gdt(ctxt, &dt);
2540 
2541 	r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2542 	if (r != X86EMUL_CONTINUE)
2543 		return r;
2544 
2545 	for (i = 0; i < 6; i++) {
2546 		r = rsm_load_seg_64(ctxt, smbase, i);
2547 		if (r != X86EMUL_CONTINUE)
2548 			return r;
2549 	}
2550 
2551 	return X86EMUL_CONTINUE;
2552 }
2553 
2554 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2555 {
2556 	unsigned long cr0, cr4, efer;
2557 	u64 smbase;
2558 	int ret;
2559 
2560 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2561 		return emulate_ud(ctxt);
2562 
2563 	/*
2564 	 * Get back to real mode, to prepare a safe state in which to load
2565 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2566 	 * supports long mode.
2567 	 */
2568 	cr4 = ctxt->ops->get_cr(ctxt, 4);
2569 	if (emulator_has_longmode(ctxt)) {
2570 		struct desc_struct cs_desc;
2571 
2572 		/* Zero CR4.PCIDE before CR0.PG.  */
2573 		if (cr4 & X86_CR4_PCIDE) {
2574 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2575 			cr4 &= ~X86_CR4_PCIDE;
2576 		}
2577 
2578 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2579 		memset(&cs_desc, 0, sizeof(cs_desc));
2580 		cs_desc.type = 0xb;
2581 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2582 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2583 	}
2584 
2585 	/* For the 64-bit case, this will clear EFER.LMA.  */
2586 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2587 	if (cr0 & X86_CR0_PE)
2588 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2589 
2590 	/* Now clear CR4.PAE (which must be done before clearing EFER.LME).  */
2591 	if (cr4 & X86_CR4_PAE)
2592 		ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2593 
2594 	/* And finally go back to 32-bit mode.  */
2595 	efer = 0;
2596 	ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2597 
2598 	smbase = ctxt->ops->get_smbase(ctxt);
2599 
2600 	/*
2601 	 * Give pre_leave_smm() a chance to make ISA-specific changes to the
2602 	 * vCPU state (e.g. enter guest mode) before loading state from the SMM
2603 	 * state-save area.
2604 	 */
2605 	if (ctxt->ops->pre_leave_smm(ctxt, smbase))
2606 		return X86EMUL_UNHANDLEABLE;
2607 
2608 	if (emulator_has_longmode(ctxt))
2609 		ret = rsm_load_state_64(ctxt, smbase + 0x8000);
2610 	else
2611 		ret = rsm_load_state_32(ctxt, smbase + 0x8000);
2612 
2613 	if (ret != X86EMUL_CONTINUE) {
2614 		/* FIXME: should triple fault */
2615 		return X86EMUL_UNHANDLEABLE;
2616 	}
2617 
2618 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2619 		ctxt->ops->set_nmi_mask(ctxt, false);
2620 
2621 	ctxt->ops->set_hflags(ctxt, ctxt->ops->get_hflags(ctxt) &
2622 		~(X86EMUL_SMM_INSIDE_NMI_MASK | X86EMUL_SMM_MASK));
2623 	return X86EMUL_CONTINUE;
2624 }
2625 
2626 static void
2627 setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
2628 			struct desc_struct *cs, struct desc_struct *ss)
2629 {
2630 	cs->l = 0;		/* will be adjusted later */
2631 	set_desc_base(cs, 0);	/* flat segment */
2632 	cs->g = 1;		/* 4kb granularity */
2633 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2634 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2635 	cs->s = 1;
2636 	cs->dpl = 0;		/* will be adjusted later */
2637 	cs->p = 1;
2638 	cs->d = 1;
2639 	cs->avl = 0;
2640 
2641 	set_desc_base(ss, 0);	/* flat segment */
2642 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2643 	ss->g = 1;		/* 4kb granularity */
2644 	ss->s = 1;
2645 	ss->type = 0x03;	/* Read/Write, Accessed */
2646 	ss->d = 1;		/* 32bit stack segment */
2647 	ss->dpl = 0;
2648 	ss->p = 1;
2649 	ss->l = 0;
2650 	ss->avl = 0;
2651 }
2652 
2653 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2654 {
2655 	u32 eax, ebx, ecx, edx;
2656 
2657 	eax = ecx = 0;
2658 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
2659 	return ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx
2660 		&& ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx
2661 		&& edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx;
2662 }
2663 
2664 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2665 {
2666 	const struct x86_emulate_ops *ops = ctxt->ops;
2667 	u32 eax, ebx, ecx, edx;
2668 
2669 	/*
2670 	 * syscall should always be enabled in longmode - so only become
2671 	 * vendor specific (cpuid) if other modes are active...
2672 	 */
2673 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2674 		return true;
2675 
2676 	eax = 0x00000000;
2677 	ecx = 0x00000000;
2678 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
2679 	/*
2680 	 * Intel ("GenuineIntel")
2681 	 * remark: Intel CPUs only support "syscall" in 64bit
2682 	 * longmode. Also an 64bit guest with a
2683 	 * 32bit compat-app running will #UD !! While this
2684 	 * behaviour can be fixed (by emulating) into AMD
2685 	 * response - CPUs of AMD can't behave like Intel.
2686 	 */
2687 	if (ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx &&
2688 	    ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx &&
2689 	    edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx)
2690 		return false;
2691 
2692 	/* AMD ("AuthenticAMD") */
2693 	if (ebx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ebx &&
2694 	    ecx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ecx &&
2695 	    edx == X86EMUL_CPUID_VENDOR_AuthenticAMD_edx)
2696 		return true;
2697 
2698 	/* AMD ("AMDisbetter!") */
2699 	if (ebx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ebx &&
2700 	    ecx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ecx &&
2701 	    edx == X86EMUL_CPUID_VENDOR_AMDisbetterI_edx)
2702 		return true;
2703 
2704 	/* default: (not Intel, not AMD), apply Intel's stricter rules... */
2705 	return false;
2706 }
2707 
2708 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2709 {
2710 	const struct x86_emulate_ops *ops = ctxt->ops;
2711 	struct desc_struct cs, ss;
2712 	u64 msr_data;
2713 	u16 cs_sel, ss_sel;
2714 	u64 efer = 0;
2715 
2716 	/* syscall is not available in real mode */
2717 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2718 	    ctxt->mode == X86EMUL_MODE_VM86)
2719 		return emulate_ud(ctxt);
2720 
2721 	if (!(em_syscall_is_enabled(ctxt)))
2722 		return emulate_ud(ctxt);
2723 
2724 	ops->get_msr(ctxt, MSR_EFER, &efer);
2725 	setup_syscalls_segments(ctxt, &cs, &ss);
2726 
2727 	if (!(efer & EFER_SCE))
2728 		return emulate_ud(ctxt);
2729 
2730 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2731 	msr_data >>= 32;
2732 	cs_sel = (u16)(msr_data & 0xfffc);
2733 	ss_sel = (u16)(msr_data + 8);
2734 
2735 	if (efer & EFER_LMA) {
2736 		cs.d = 0;
2737 		cs.l = 1;
2738 	}
2739 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2740 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2741 
2742 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2743 	if (efer & EFER_LMA) {
2744 #ifdef CONFIG_X86_64
2745 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2746 
2747 		ops->get_msr(ctxt,
2748 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2749 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2750 		ctxt->_eip = msr_data;
2751 
2752 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2753 		ctxt->eflags &= ~msr_data;
2754 		ctxt->eflags |= X86_EFLAGS_FIXED;
2755 #endif
2756 	} else {
2757 		/* legacy mode */
2758 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2759 		ctxt->_eip = (u32)msr_data;
2760 
2761 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2762 	}
2763 
2764 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2765 	return X86EMUL_CONTINUE;
2766 }
2767 
2768 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2769 {
2770 	const struct x86_emulate_ops *ops = ctxt->ops;
2771 	struct desc_struct cs, ss;
2772 	u64 msr_data;
2773 	u16 cs_sel, ss_sel;
2774 	u64 efer = 0;
2775 
2776 	ops->get_msr(ctxt, MSR_EFER, &efer);
2777 	/* inject #GP if in real mode */
2778 	if (ctxt->mode == X86EMUL_MODE_REAL)
2779 		return emulate_gp(ctxt, 0);
2780 
2781 	/*
2782 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2783 	 * mode).
2784 	 */
2785 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2786 	    && !vendor_intel(ctxt))
2787 		return emulate_ud(ctxt);
2788 
2789 	/* sysenter/sysexit have not been tested in 64bit mode. */
2790 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2791 		return X86EMUL_UNHANDLEABLE;
2792 
2793 	setup_syscalls_segments(ctxt, &cs, &ss);
2794 
2795 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2796 	if ((msr_data & 0xfffc) == 0x0)
2797 		return emulate_gp(ctxt, 0);
2798 
2799 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2800 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2801 	ss_sel = cs_sel + 8;
2802 	if (efer & EFER_LMA) {
2803 		cs.d = 0;
2804 		cs.l = 1;
2805 	}
2806 
2807 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2808 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2809 
2810 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2811 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2812 
2813 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2814 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2815 							      (u32)msr_data;
2816 
2817 	return X86EMUL_CONTINUE;
2818 }
2819 
2820 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2821 {
2822 	const struct x86_emulate_ops *ops = ctxt->ops;
2823 	struct desc_struct cs, ss;
2824 	u64 msr_data, rcx, rdx;
2825 	int usermode;
2826 	u16 cs_sel = 0, ss_sel = 0;
2827 
2828 	/* inject #GP if in real mode or Virtual 8086 mode */
2829 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2830 	    ctxt->mode == X86EMUL_MODE_VM86)
2831 		return emulate_gp(ctxt, 0);
2832 
2833 	setup_syscalls_segments(ctxt, &cs, &ss);
2834 
2835 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2836 		usermode = X86EMUL_MODE_PROT64;
2837 	else
2838 		usermode = X86EMUL_MODE_PROT32;
2839 
2840 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2841 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2842 
2843 	cs.dpl = 3;
2844 	ss.dpl = 3;
2845 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2846 	switch (usermode) {
2847 	case X86EMUL_MODE_PROT32:
2848 		cs_sel = (u16)(msr_data + 16);
2849 		if ((msr_data & 0xfffc) == 0x0)
2850 			return emulate_gp(ctxt, 0);
2851 		ss_sel = (u16)(msr_data + 24);
2852 		rcx = (u32)rcx;
2853 		rdx = (u32)rdx;
2854 		break;
2855 	case X86EMUL_MODE_PROT64:
2856 		cs_sel = (u16)(msr_data + 32);
2857 		if (msr_data == 0x0)
2858 			return emulate_gp(ctxt, 0);
2859 		ss_sel = cs_sel + 8;
2860 		cs.d = 0;
2861 		cs.l = 1;
2862 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2863 		    emul_is_noncanonical_address(rdx, ctxt))
2864 			return emulate_gp(ctxt, 0);
2865 		break;
2866 	}
2867 	cs_sel |= SEGMENT_RPL_MASK;
2868 	ss_sel |= SEGMENT_RPL_MASK;
2869 
2870 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2871 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2872 
2873 	ctxt->_eip = rdx;
2874 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2875 
2876 	return X86EMUL_CONTINUE;
2877 }
2878 
2879 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2880 {
2881 	int iopl;
2882 	if (ctxt->mode == X86EMUL_MODE_REAL)
2883 		return false;
2884 	if (ctxt->mode == X86EMUL_MODE_VM86)
2885 		return true;
2886 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2887 	return ctxt->ops->cpl(ctxt) > iopl;
2888 }
2889 
2890 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2891 					    u16 port, u16 len)
2892 {
2893 	const struct x86_emulate_ops *ops = ctxt->ops;
2894 	struct desc_struct tr_seg;
2895 	u32 base3;
2896 	int r;
2897 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2898 	unsigned mask = (1 << len) - 1;
2899 	unsigned long base;
2900 
2901 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2902 	if (!tr_seg.p)
2903 		return false;
2904 	if (desc_limit_scaled(&tr_seg) < 103)
2905 		return false;
2906 	base = get_desc_base(&tr_seg);
2907 #ifdef CONFIG_X86_64
2908 	base |= ((u64)base3) << 32;
2909 #endif
2910 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL);
2911 	if (r != X86EMUL_CONTINUE)
2912 		return false;
2913 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2914 		return false;
2915 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL);
2916 	if (r != X86EMUL_CONTINUE)
2917 		return false;
2918 	if ((perm >> bit_idx) & mask)
2919 		return false;
2920 	return true;
2921 }
2922 
2923 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2924 				 u16 port, u16 len)
2925 {
2926 	if (ctxt->perm_ok)
2927 		return true;
2928 
2929 	if (emulator_bad_iopl(ctxt))
2930 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2931 			return false;
2932 
2933 	ctxt->perm_ok = true;
2934 
2935 	return true;
2936 }
2937 
2938 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2939 {
2940 	/*
2941 	 * Intel CPUs mask the counter and pointers in quite strange
2942 	 * manner when ECX is zero due to REP-string optimizations.
2943 	 */
2944 #ifdef CONFIG_X86_64
2945 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2946 		return;
2947 
2948 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2949 
2950 	switch (ctxt->b) {
2951 	case 0xa4:	/* movsb */
2952 	case 0xa5:	/* movsd/w */
2953 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2954 		/* fall through */
2955 	case 0xaa:	/* stosb */
2956 	case 0xab:	/* stosd/w */
2957 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2958 	}
2959 #endif
2960 }
2961 
2962 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2963 				struct tss_segment_16 *tss)
2964 {
2965 	tss->ip = ctxt->_eip;
2966 	tss->flag = ctxt->eflags;
2967 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2968 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2969 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2970 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2971 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2972 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2973 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2974 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2975 
2976 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2977 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2978 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2979 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2980 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2981 }
2982 
2983 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2984 				 struct tss_segment_16 *tss)
2985 {
2986 	int ret;
2987 	u8 cpl;
2988 
2989 	ctxt->_eip = tss->ip;
2990 	ctxt->eflags = tss->flag | 2;
2991 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2992 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2993 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2994 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2995 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2996 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2997 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2998 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2999 
3000 	/*
3001 	 * SDM says that segment selectors are loaded before segment
3002 	 * descriptors
3003 	 */
3004 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
3005 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3006 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3007 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3008 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3009 
3010 	cpl = tss->cs & 3;
3011 
3012 	/*
3013 	 * Now load segment descriptors. If fault happens at this stage
3014 	 * it is handled in a context of new task
3015 	 */
3016 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
3017 					X86_TRANSFER_TASK_SWITCH, NULL);
3018 	if (ret != X86EMUL_CONTINUE)
3019 		return ret;
3020 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3021 					X86_TRANSFER_TASK_SWITCH, NULL);
3022 	if (ret != X86EMUL_CONTINUE)
3023 		return ret;
3024 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3025 					X86_TRANSFER_TASK_SWITCH, NULL);
3026 	if (ret != X86EMUL_CONTINUE)
3027 		return ret;
3028 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3029 					X86_TRANSFER_TASK_SWITCH, NULL);
3030 	if (ret != X86EMUL_CONTINUE)
3031 		return ret;
3032 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3033 					X86_TRANSFER_TASK_SWITCH, NULL);
3034 	if (ret != X86EMUL_CONTINUE)
3035 		return ret;
3036 
3037 	return X86EMUL_CONTINUE;
3038 }
3039 
3040 static int task_switch_16(struct x86_emulate_ctxt *ctxt,
3041 			  u16 tss_selector, u16 old_tss_sel,
3042 			  ulong old_tss_base, struct desc_struct *new_desc)
3043 {
3044 	const struct x86_emulate_ops *ops = ctxt->ops;
3045 	struct tss_segment_16 tss_seg;
3046 	int ret;
3047 	u32 new_tss_base = get_desc_base(new_desc);
3048 
3049 	ret = ops->read_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3050 			    &ctxt->exception);
3051 	if (ret != X86EMUL_CONTINUE)
3052 		return ret;
3053 
3054 	save_state_to_tss16(ctxt, &tss_seg);
3055 
3056 	ret = ops->write_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3057 			     &ctxt->exception);
3058 	if (ret != X86EMUL_CONTINUE)
3059 		return ret;
3060 
3061 	ret = ops->read_std(ctxt, new_tss_base, &tss_seg, sizeof tss_seg,
3062 			    &ctxt->exception);
3063 	if (ret != X86EMUL_CONTINUE)
3064 		return ret;
3065 
3066 	if (old_tss_sel != 0xffff) {
3067 		tss_seg.prev_task_link = old_tss_sel;
3068 
3069 		ret = ops->write_std(ctxt, new_tss_base,
3070 				     &tss_seg.prev_task_link,
3071 				     sizeof tss_seg.prev_task_link,
3072 				     &ctxt->exception);
3073 		if (ret != X86EMUL_CONTINUE)
3074 			return ret;
3075 	}
3076 
3077 	return load_state_from_tss16(ctxt, &tss_seg);
3078 }
3079 
3080 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3081 				struct tss_segment_32 *tss)
3082 {
3083 	/* CR3 and ldt selector are not saved intentionally */
3084 	tss->eip = ctxt->_eip;
3085 	tss->eflags = ctxt->eflags;
3086 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3087 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3088 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3089 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3090 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3091 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3092 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3093 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3094 
3095 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3096 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3097 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3098 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3099 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3100 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3101 }
3102 
3103 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3104 				 struct tss_segment_32 *tss)
3105 {
3106 	int ret;
3107 	u8 cpl;
3108 
3109 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3110 		return emulate_gp(ctxt, 0);
3111 	ctxt->_eip = tss->eip;
3112 	ctxt->eflags = tss->eflags | 2;
3113 
3114 	/* General purpose registers */
3115 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3116 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3117 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3118 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3119 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3120 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3121 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3122 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3123 
3124 	/*
3125 	 * SDM says that segment selectors are loaded before segment
3126 	 * descriptors.  This is important because CPL checks will
3127 	 * use CS.RPL.
3128 	 */
3129 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3130 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3131 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3132 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3133 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3134 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3135 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3136 
3137 	/*
3138 	 * If we're switching between Protected Mode and VM86, we need to make
3139 	 * sure to update the mode before loading the segment descriptors so
3140 	 * that the selectors are interpreted correctly.
3141 	 */
3142 	if (ctxt->eflags & X86_EFLAGS_VM) {
3143 		ctxt->mode = X86EMUL_MODE_VM86;
3144 		cpl = 3;
3145 	} else {
3146 		ctxt->mode = X86EMUL_MODE_PROT32;
3147 		cpl = tss->cs & 3;
3148 	}
3149 
3150 	/*
3151 	 * Now load segment descriptors. If fault happenes at this stage
3152 	 * it is handled in a context of new task
3153 	 */
3154 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3155 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3156 	if (ret != X86EMUL_CONTINUE)
3157 		return ret;
3158 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3159 					X86_TRANSFER_TASK_SWITCH, NULL);
3160 	if (ret != X86EMUL_CONTINUE)
3161 		return ret;
3162 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3163 					X86_TRANSFER_TASK_SWITCH, NULL);
3164 	if (ret != X86EMUL_CONTINUE)
3165 		return ret;
3166 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3167 					X86_TRANSFER_TASK_SWITCH, NULL);
3168 	if (ret != X86EMUL_CONTINUE)
3169 		return ret;
3170 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3171 					X86_TRANSFER_TASK_SWITCH, NULL);
3172 	if (ret != X86EMUL_CONTINUE)
3173 		return ret;
3174 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3175 					X86_TRANSFER_TASK_SWITCH, NULL);
3176 	if (ret != X86EMUL_CONTINUE)
3177 		return ret;
3178 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3179 					X86_TRANSFER_TASK_SWITCH, NULL);
3180 
3181 	return ret;
3182 }
3183 
3184 static int task_switch_32(struct x86_emulate_ctxt *ctxt,
3185 			  u16 tss_selector, u16 old_tss_sel,
3186 			  ulong old_tss_base, struct desc_struct *new_desc)
3187 {
3188 	const struct x86_emulate_ops *ops = ctxt->ops;
3189 	struct tss_segment_32 tss_seg;
3190 	int ret;
3191 	u32 new_tss_base = get_desc_base(new_desc);
3192 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3193 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3194 
3195 	ret = ops->read_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3196 			    &ctxt->exception);
3197 	if (ret != X86EMUL_CONTINUE)
3198 		return ret;
3199 
3200 	save_state_to_tss32(ctxt, &tss_seg);
3201 
3202 	/* Only GP registers and segment selectors are saved */
3203 	ret = ops->write_std(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3204 			     ldt_sel_offset - eip_offset, &ctxt->exception);
3205 	if (ret != X86EMUL_CONTINUE)
3206 		return ret;
3207 
3208 	ret = ops->read_std(ctxt, new_tss_base, &tss_seg, sizeof tss_seg,
3209 			    &ctxt->exception);
3210 	if (ret != X86EMUL_CONTINUE)
3211 		return ret;
3212 
3213 	if (old_tss_sel != 0xffff) {
3214 		tss_seg.prev_task_link = old_tss_sel;
3215 
3216 		ret = ops->write_std(ctxt, new_tss_base,
3217 				     &tss_seg.prev_task_link,
3218 				     sizeof tss_seg.prev_task_link,
3219 				     &ctxt->exception);
3220 		if (ret != X86EMUL_CONTINUE)
3221 			return ret;
3222 	}
3223 
3224 	return load_state_from_tss32(ctxt, &tss_seg);
3225 }
3226 
3227 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3228 				   u16 tss_selector, int idt_index, int reason,
3229 				   bool has_error_code, u32 error_code)
3230 {
3231 	const struct x86_emulate_ops *ops = ctxt->ops;
3232 	struct desc_struct curr_tss_desc, next_tss_desc;
3233 	int ret;
3234 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3235 	ulong old_tss_base =
3236 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3237 	u32 desc_limit;
3238 	ulong desc_addr, dr7;
3239 
3240 	/* FIXME: old_tss_base == ~0 ? */
3241 
3242 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3243 	if (ret != X86EMUL_CONTINUE)
3244 		return ret;
3245 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3246 	if (ret != X86EMUL_CONTINUE)
3247 		return ret;
3248 
3249 	/* FIXME: check that next_tss_desc is tss */
3250 
3251 	/*
3252 	 * Check privileges. The three cases are task switch caused by...
3253 	 *
3254 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3255 	 * 2. Exception/IRQ/iret: No check is performed
3256 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3257 	 *    hardware checks it before exiting.
3258 	 */
3259 	if (reason == TASK_SWITCH_GATE) {
3260 		if (idt_index != -1) {
3261 			/* Software interrupts */
3262 			struct desc_struct task_gate_desc;
3263 			int dpl;
3264 
3265 			ret = read_interrupt_descriptor(ctxt, idt_index,
3266 							&task_gate_desc);
3267 			if (ret != X86EMUL_CONTINUE)
3268 				return ret;
3269 
3270 			dpl = task_gate_desc.dpl;
3271 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3272 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3273 		}
3274 	}
3275 
3276 	desc_limit = desc_limit_scaled(&next_tss_desc);
3277 	if (!next_tss_desc.p ||
3278 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3279 	     desc_limit < 0x2b)) {
3280 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3281 	}
3282 
3283 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3284 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3285 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3286 	}
3287 
3288 	if (reason == TASK_SWITCH_IRET)
3289 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3290 
3291 	/* set back link to prev task only if NT bit is set in eflags
3292 	   note that old_tss_sel is not used after this point */
3293 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3294 		old_tss_sel = 0xffff;
3295 
3296 	if (next_tss_desc.type & 8)
3297 		ret = task_switch_32(ctxt, tss_selector, old_tss_sel,
3298 				     old_tss_base, &next_tss_desc);
3299 	else
3300 		ret = task_switch_16(ctxt, tss_selector, old_tss_sel,
3301 				     old_tss_base, &next_tss_desc);
3302 	if (ret != X86EMUL_CONTINUE)
3303 		return ret;
3304 
3305 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3306 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3307 
3308 	if (reason != TASK_SWITCH_IRET) {
3309 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3310 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3311 	}
3312 
3313 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3314 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3315 
3316 	if (has_error_code) {
3317 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3318 		ctxt->lock_prefix = 0;
3319 		ctxt->src.val = (unsigned long) error_code;
3320 		ret = em_push(ctxt);
3321 	}
3322 
3323 	ops->get_dr(ctxt, 7, &dr7);
3324 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3325 
3326 	return ret;
3327 }
3328 
3329 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3330 			 u16 tss_selector, int idt_index, int reason,
3331 			 bool has_error_code, u32 error_code)
3332 {
3333 	int rc;
3334 
3335 	invalidate_registers(ctxt);
3336 	ctxt->_eip = ctxt->eip;
3337 	ctxt->dst.type = OP_NONE;
3338 
3339 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3340 				     has_error_code, error_code);
3341 
3342 	if (rc == X86EMUL_CONTINUE) {
3343 		ctxt->eip = ctxt->_eip;
3344 		writeback_registers(ctxt);
3345 	}
3346 
3347 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3348 }
3349 
3350 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3351 		struct operand *op)
3352 {
3353 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3354 
3355 	register_address_increment(ctxt, reg, df * op->bytes);
3356 	op->addr.mem.ea = register_address(ctxt, reg);
3357 }
3358 
3359 static int em_das(struct x86_emulate_ctxt *ctxt)
3360 {
3361 	u8 al, old_al;
3362 	bool af, cf, old_cf;
3363 
3364 	cf = ctxt->eflags & X86_EFLAGS_CF;
3365 	al = ctxt->dst.val;
3366 
3367 	old_al = al;
3368 	old_cf = cf;
3369 	cf = false;
3370 	af = ctxt->eflags & X86_EFLAGS_AF;
3371 	if ((al & 0x0f) > 9 || af) {
3372 		al -= 6;
3373 		cf = old_cf | (al >= 250);
3374 		af = true;
3375 	} else {
3376 		af = false;
3377 	}
3378 	if (old_al > 0x99 || old_cf) {
3379 		al -= 0x60;
3380 		cf = true;
3381 	}
3382 
3383 	ctxt->dst.val = al;
3384 	/* Set PF, ZF, SF */
3385 	ctxt->src.type = OP_IMM;
3386 	ctxt->src.val = 0;
3387 	ctxt->src.bytes = 1;
3388 	fastop(ctxt, em_or);
3389 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3390 	if (cf)
3391 		ctxt->eflags |= X86_EFLAGS_CF;
3392 	if (af)
3393 		ctxt->eflags |= X86_EFLAGS_AF;
3394 	return X86EMUL_CONTINUE;
3395 }
3396 
3397 static int em_aam(struct x86_emulate_ctxt *ctxt)
3398 {
3399 	u8 al, ah;
3400 
3401 	if (ctxt->src.val == 0)
3402 		return emulate_de(ctxt);
3403 
3404 	al = ctxt->dst.val & 0xff;
3405 	ah = al / ctxt->src.val;
3406 	al %= ctxt->src.val;
3407 
3408 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3409 
3410 	/* Set PF, ZF, SF */
3411 	ctxt->src.type = OP_IMM;
3412 	ctxt->src.val = 0;
3413 	ctxt->src.bytes = 1;
3414 	fastop(ctxt, em_or);
3415 
3416 	return X86EMUL_CONTINUE;
3417 }
3418 
3419 static int em_aad(struct x86_emulate_ctxt *ctxt)
3420 {
3421 	u8 al = ctxt->dst.val & 0xff;
3422 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3423 
3424 	al = (al + (ah * ctxt->src.val)) & 0xff;
3425 
3426 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3427 
3428 	/* Set PF, ZF, SF */
3429 	ctxt->src.type = OP_IMM;
3430 	ctxt->src.val = 0;
3431 	ctxt->src.bytes = 1;
3432 	fastop(ctxt, em_or);
3433 
3434 	return X86EMUL_CONTINUE;
3435 }
3436 
3437 static int em_call(struct x86_emulate_ctxt *ctxt)
3438 {
3439 	int rc;
3440 	long rel = ctxt->src.val;
3441 
3442 	ctxt->src.val = (unsigned long)ctxt->_eip;
3443 	rc = jmp_rel(ctxt, rel);
3444 	if (rc != X86EMUL_CONTINUE)
3445 		return rc;
3446 	return em_push(ctxt);
3447 }
3448 
3449 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3450 {
3451 	u16 sel, old_cs;
3452 	ulong old_eip;
3453 	int rc;
3454 	struct desc_struct old_desc, new_desc;
3455 	const struct x86_emulate_ops *ops = ctxt->ops;
3456 	int cpl = ctxt->ops->cpl(ctxt);
3457 	enum x86emul_mode prev_mode = ctxt->mode;
3458 
3459 	old_eip = ctxt->_eip;
3460 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3461 
3462 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3463 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3464 				       X86_TRANSFER_CALL_JMP, &new_desc);
3465 	if (rc != X86EMUL_CONTINUE)
3466 		return rc;
3467 
3468 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3469 	if (rc != X86EMUL_CONTINUE)
3470 		goto fail;
3471 
3472 	ctxt->src.val = old_cs;
3473 	rc = em_push(ctxt);
3474 	if (rc != X86EMUL_CONTINUE)
3475 		goto fail;
3476 
3477 	ctxt->src.val = old_eip;
3478 	rc = em_push(ctxt);
3479 	/* If we failed, we tainted the memory, but the very least we should
3480 	   restore cs */
3481 	if (rc != X86EMUL_CONTINUE) {
3482 		pr_warn_once("faulting far call emulation tainted memory\n");
3483 		goto fail;
3484 	}
3485 	return rc;
3486 fail:
3487 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3488 	ctxt->mode = prev_mode;
3489 	return rc;
3490 
3491 }
3492 
3493 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3494 {
3495 	int rc;
3496 	unsigned long eip;
3497 
3498 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3499 	if (rc != X86EMUL_CONTINUE)
3500 		return rc;
3501 	rc = assign_eip_near(ctxt, eip);
3502 	if (rc != X86EMUL_CONTINUE)
3503 		return rc;
3504 	rsp_increment(ctxt, ctxt->src.val);
3505 	return X86EMUL_CONTINUE;
3506 }
3507 
3508 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3509 {
3510 	/* Write back the register source. */
3511 	ctxt->src.val = ctxt->dst.val;
3512 	write_register_operand(&ctxt->src);
3513 
3514 	/* Write back the memory destination with implicit LOCK prefix. */
3515 	ctxt->dst.val = ctxt->src.orig_val;
3516 	ctxt->lock_prefix = 1;
3517 	return X86EMUL_CONTINUE;
3518 }
3519 
3520 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3521 {
3522 	ctxt->dst.val = ctxt->src2.val;
3523 	return fastop(ctxt, em_imul);
3524 }
3525 
3526 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3527 {
3528 	ctxt->dst.type = OP_REG;
3529 	ctxt->dst.bytes = ctxt->src.bytes;
3530 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3531 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3532 
3533 	return X86EMUL_CONTINUE;
3534 }
3535 
3536 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3537 {
3538 	u64 tsc_aux = 0;
3539 
3540 	if (ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux))
3541 		return emulate_gp(ctxt, 0);
3542 	ctxt->dst.val = tsc_aux;
3543 	return X86EMUL_CONTINUE;
3544 }
3545 
3546 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3547 {
3548 	u64 tsc = 0;
3549 
3550 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3551 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3552 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3553 	return X86EMUL_CONTINUE;
3554 }
3555 
3556 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3557 {
3558 	u64 pmc;
3559 
3560 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3561 		return emulate_gp(ctxt, 0);
3562 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3563 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3564 	return X86EMUL_CONTINUE;
3565 }
3566 
3567 static int em_mov(struct x86_emulate_ctxt *ctxt)
3568 {
3569 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3570 	return X86EMUL_CONTINUE;
3571 }
3572 
3573 #define FFL(x) bit(X86_FEATURE_##x)
3574 
3575 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3576 {
3577 	u32 ebx, ecx, edx, eax = 1;
3578 	u16 tmp;
3579 
3580 	/*
3581 	 * Check MOVBE is set in the guest-visible CPUID leaf.
3582 	 */
3583 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3584 	if (!(ecx & FFL(MOVBE)))
3585 		return emulate_ud(ctxt);
3586 
3587 	switch (ctxt->op_bytes) {
3588 	case 2:
3589 		/*
3590 		 * From MOVBE definition: "...When the operand size is 16 bits,
3591 		 * the upper word of the destination register remains unchanged
3592 		 * ..."
3593 		 *
3594 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3595 		 * rules so we have to do the operation almost per hand.
3596 		 */
3597 		tmp = (u16)ctxt->src.val;
3598 		ctxt->dst.val &= ~0xffffUL;
3599 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3600 		break;
3601 	case 4:
3602 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3603 		break;
3604 	case 8:
3605 		ctxt->dst.val = swab64(ctxt->src.val);
3606 		break;
3607 	default:
3608 		BUG();
3609 	}
3610 	return X86EMUL_CONTINUE;
3611 }
3612 
3613 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3614 {
3615 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3616 		return emulate_gp(ctxt, 0);
3617 
3618 	/* Disable writeback. */
3619 	ctxt->dst.type = OP_NONE;
3620 	return X86EMUL_CONTINUE;
3621 }
3622 
3623 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3624 {
3625 	unsigned long val;
3626 
3627 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3628 		val = ctxt->src.val & ~0ULL;
3629 	else
3630 		val = ctxt->src.val & ~0U;
3631 
3632 	/* #UD condition is already handled. */
3633 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3634 		return emulate_gp(ctxt, 0);
3635 
3636 	/* Disable writeback. */
3637 	ctxt->dst.type = OP_NONE;
3638 	return X86EMUL_CONTINUE;
3639 }
3640 
3641 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3642 {
3643 	u64 msr_data;
3644 
3645 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3646 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3647 	if (ctxt->ops->set_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), msr_data))
3648 		return emulate_gp(ctxt, 0);
3649 
3650 	return X86EMUL_CONTINUE;
3651 }
3652 
3653 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3654 {
3655 	u64 msr_data;
3656 
3657 	if (ctxt->ops->get_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &msr_data))
3658 		return emulate_gp(ctxt, 0);
3659 
3660 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3661 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3662 	return X86EMUL_CONTINUE;
3663 }
3664 
3665 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3666 {
3667 	if (segment > VCPU_SREG_GS &&
3668 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3669 	    ctxt->ops->cpl(ctxt) > 0)
3670 		return emulate_gp(ctxt, 0);
3671 
3672 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3673 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3674 		ctxt->dst.bytes = 2;
3675 	return X86EMUL_CONTINUE;
3676 }
3677 
3678 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3679 {
3680 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3681 		return emulate_ud(ctxt);
3682 
3683 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3684 }
3685 
3686 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3687 {
3688 	u16 sel = ctxt->src.val;
3689 
3690 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3691 		return emulate_ud(ctxt);
3692 
3693 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3694 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3695 
3696 	/* Disable writeback. */
3697 	ctxt->dst.type = OP_NONE;
3698 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3699 }
3700 
3701 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3702 {
3703 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3704 }
3705 
3706 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3707 {
3708 	u16 sel = ctxt->src.val;
3709 
3710 	/* Disable writeback. */
3711 	ctxt->dst.type = OP_NONE;
3712 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3713 }
3714 
3715 static int em_str(struct x86_emulate_ctxt *ctxt)
3716 {
3717 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3718 }
3719 
3720 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3721 {
3722 	u16 sel = ctxt->src.val;
3723 
3724 	/* Disable writeback. */
3725 	ctxt->dst.type = OP_NONE;
3726 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3727 }
3728 
3729 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3730 {
3731 	int rc;
3732 	ulong linear;
3733 
3734 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3735 	if (rc == X86EMUL_CONTINUE)
3736 		ctxt->ops->invlpg(ctxt, linear);
3737 	/* Disable writeback. */
3738 	ctxt->dst.type = OP_NONE;
3739 	return X86EMUL_CONTINUE;
3740 }
3741 
3742 static int em_clts(struct x86_emulate_ctxt *ctxt)
3743 {
3744 	ulong cr0;
3745 
3746 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3747 	cr0 &= ~X86_CR0_TS;
3748 	ctxt->ops->set_cr(ctxt, 0, cr0);
3749 	return X86EMUL_CONTINUE;
3750 }
3751 
3752 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3753 {
3754 	int rc = ctxt->ops->fix_hypercall(ctxt);
3755 
3756 	if (rc != X86EMUL_CONTINUE)
3757 		return rc;
3758 
3759 	/* Let the processor re-execute the fixed hypercall */
3760 	ctxt->_eip = ctxt->eip;
3761 	/* Disable writeback. */
3762 	ctxt->dst.type = OP_NONE;
3763 	return X86EMUL_CONTINUE;
3764 }
3765 
3766 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3767 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3768 					      struct desc_ptr *ptr))
3769 {
3770 	struct desc_ptr desc_ptr;
3771 
3772 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3773 	    ctxt->ops->cpl(ctxt) > 0)
3774 		return emulate_gp(ctxt, 0);
3775 
3776 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3777 		ctxt->op_bytes = 8;
3778 	get(ctxt, &desc_ptr);
3779 	if (ctxt->op_bytes == 2) {
3780 		ctxt->op_bytes = 4;
3781 		desc_ptr.address &= 0x00ffffff;
3782 	}
3783 	/* Disable writeback. */
3784 	ctxt->dst.type = OP_NONE;
3785 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3786 				   &desc_ptr, 2 + ctxt->op_bytes);
3787 }
3788 
3789 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3790 {
3791 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3792 }
3793 
3794 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3795 {
3796 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3797 }
3798 
3799 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3800 {
3801 	struct desc_ptr desc_ptr;
3802 	int rc;
3803 
3804 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3805 		ctxt->op_bytes = 8;
3806 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3807 			     &desc_ptr.size, &desc_ptr.address,
3808 			     ctxt->op_bytes);
3809 	if (rc != X86EMUL_CONTINUE)
3810 		return rc;
3811 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3812 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3813 		return emulate_gp(ctxt, 0);
3814 	if (lgdt)
3815 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3816 	else
3817 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3818 	/* Disable writeback. */
3819 	ctxt->dst.type = OP_NONE;
3820 	return X86EMUL_CONTINUE;
3821 }
3822 
3823 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3824 {
3825 	return em_lgdt_lidt(ctxt, true);
3826 }
3827 
3828 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3829 {
3830 	return em_lgdt_lidt(ctxt, false);
3831 }
3832 
3833 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3834 {
3835 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3836 	    ctxt->ops->cpl(ctxt) > 0)
3837 		return emulate_gp(ctxt, 0);
3838 
3839 	if (ctxt->dst.type == OP_MEM)
3840 		ctxt->dst.bytes = 2;
3841 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3842 	return X86EMUL_CONTINUE;
3843 }
3844 
3845 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3846 {
3847 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3848 			  | (ctxt->src.val & 0x0f));
3849 	ctxt->dst.type = OP_NONE;
3850 	return X86EMUL_CONTINUE;
3851 }
3852 
3853 static int em_loop(struct x86_emulate_ctxt *ctxt)
3854 {
3855 	int rc = X86EMUL_CONTINUE;
3856 
3857 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3858 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3859 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3860 		rc = jmp_rel(ctxt, ctxt->src.val);
3861 
3862 	return rc;
3863 }
3864 
3865 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3866 {
3867 	int rc = X86EMUL_CONTINUE;
3868 
3869 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3870 		rc = jmp_rel(ctxt, ctxt->src.val);
3871 
3872 	return rc;
3873 }
3874 
3875 static int em_in(struct x86_emulate_ctxt *ctxt)
3876 {
3877 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3878 			     &ctxt->dst.val))
3879 		return X86EMUL_IO_NEEDED;
3880 
3881 	return X86EMUL_CONTINUE;
3882 }
3883 
3884 static int em_out(struct x86_emulate_ctxt *ctxt)
3885 {
3886 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3887 				    &ctxt->src.val, 1);
3888 	/* Disable writeback. */
3889 	ctxt->dst.type = OP_NONE;
3890 	return X86EMUL_CONTINUE;
3891 }
3892 
3893 static int em_cli(struct x86_emulate_ctxt *ctxt)
3894 {
3895 	if (emulator_bad_iopl(ctxt))
3896 		return emulate_gp(ctxt, 0);
3897 
3898 	ctxt->eflags &= ~X86_EFLAGS_IF;
3899 	return X86EMUL_CONTINUE;
3900 }
3901 
3902 static int em_sti(struct x86_emulate_ctxt *ctxt)
3903 {
3904 	if (emulator_bad_iopl(ctxt))
3905 		return emulate_gp(ctxt, 0);
3906 
3907 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3908 	ctxt->eflags |= X86_EFLAGS_IF;
3909 	return X86EMUL_CONTINUE;
3910 }
3911 
3912 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3913 {
3914 	u32 eax, ebx, ecx, edx;
3915 	u64 msr = 0;
3916 
3917 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3918 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3919 	    ctxt->ops->cpl(ctxt)) {
3920 		return emulate_gp(ctxt, 0);
3921 	}
3922 
3923 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3924 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3925 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
3926 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3927 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3928 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3929 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3930 	return X86EMUL_CONTINUE;
3931 }
3932 
3933 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3934 {
3935 	u32 flags;
3936 
3937 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3938 		X86_EFLAGS_SF;
3939 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3940 
3941 	ctxt->eflags &= ~0xffUL;
3942 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3943 	return X86EMUL_CONTINUE;
3944 }
3945 
3946 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3947 {
3948 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3949 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3950 	return X86EMUL_CONTINUE;
3951 }
3952 
3953 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3954 {
3955 	switch (ctxt->op_bytes) {
3956 #ifdef CONFIG_X86_64
3957 	case 8:
3958 		asm("bswap %0" : "+r"(ctxt->dst.val));
3959 		break;
3960 #endif
3961 	default:
3962 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3963 		break;
3964 	}
3965 	return X86EMUL_CONTINUE;
3966 }
3967 
3968 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3969 {
3970 	/* emulating clflush regardless of cpuid */
3971 	return X86EMUL_CONTINUE;
3972 }
3973 
3974 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3975 {
3976 	ctxt->dst.val = (s32) ctxt->src.val;
3977 	return X86EMUL_CONTINUE;
3978 }
3979 
3980 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3981 {
3982 	u32 eax = 1, ebx, ecx = 0, edx;
3983 
3984 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3985 	if (!(edx & FFL(FXSR)))
3986 		return emulate_ud(ctxt);
3987 
3988 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3989 		return emulate_nm(ctxt);
3990 
3991 	/*
3992 	 * Don't emulate a case that should never be hit, instead of working
3993 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3994 	 */
3995 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3996 		return X86EMUL_UNHANDLEABLE;
3997 
3998 	return X86EMUL_CONTINUE;
3999 }
4000 
4001 /*
4002  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
4003  * and restore MXCSR.
4004  */
4005 static size_t __fxstate_size(int nregs)
4006 {
4007 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
4008 }
4009 
4010 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
4011 {
4012 	bool cr4_osfxsr;
4013 	if (ctxt->mode == X86EMUL_MODE_PROT64)
4014 		return __fxstate_size(16);
4015 
4016 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
4017 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
4018 }
4019 
4020 /*
4021  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
4022  *  1) 16 bit mode
4023  *  2) 32 bit mode
4024  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
4025  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
4026  *       save and restore
4027  *  3) 64-bit mode with REX.W prefix
4028  *     - like (2), but XMM 8-15 are being saved and restored
4029  *  4) 64-bit mode without REX.W prefix
4030  *     - like (3), but FIP and FDP are 64 bit
4031  *
4032  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
4033  * desired result.  (4) is not emulated.
4034  *
4035  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
4036  * and FPU DS) should match.
4037  */
4038 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
4039 {
4040 	struct fxregs_state fx_state;
4041 	int rc;
4042 
4043 	rc = check_fxsr(ctxt);
4044 	if (rc != X86EMUL_CONTINUE)
4045 		return rc;
4046 
4047 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4048 
4049 	if (rc != X86EMUL_CONTINUE)
4050 		return rc;
4051 
4052 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4053 		                   fxstate_size(ctxt));
4054 }
4055 
4056 /*
4057  * FXRSTOR might restore XMM registers not provided by the guest. Fill
4058  * in the host registers (via FXSAVE) instead, so they won't be modified.
4059  * (preemption has to stay disabled until FXRSTOR).
4060  *
4061  * Use noinline to keep the stack for other functions called by callers small.
4062  */
4063 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
4064 				 const size_t used_size)
4065 {
4066 	struct fxregs_state fx_tmp;
4067 	int rc;
4068 
4069 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
4070 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
4071 	       __fxstate_size(16) - used_size);
4072 
4073 	return rc;
4074 }
4075 
4076 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4077 {
4078 	struct fxregs_state fx_state;
4079 	int rc;
4080 	size_t size;
4081 
4082 	rc = check_fxsr(ctxt);
4083 	if (rc != X86EMUL_CONTINUE)
4084 		return rc;
4085 
4086 	size = fxstate_size(ctxt);
4087 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4088 	if (rc != X86EMUL_CONTINUE)
4089 		return rc;
4090 
4091 	if (size < __fxstate_size(16)) {
4092 		rc = fxregs_fixup(&fx_state, size);
4093 		if (rc != X86EMUL_CONTINUE)
4094 			goto out;
4095 	}
4096 
4097 	if (fx_state.mxcsr >> 16) {
4098 		rc = emulate_gp(ctxt, 0);
4099 		goto out;
4100 	}
4101 
4102 	if (rc == X86EMUL_CONTINUE)
4103 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4104 
4105 out:
4106 	return rc;
4107 }
4108 
4109 static bool valid_cr(int nr)
4110 {
4111 	switch (nr) {
4112 	case 0:
4113 	case 2 ... 4:
4114 	case 8:
4115 		return true;
4116 	default:
4117 		return false;
4118 	}
4119 }
4120 
4121 static int check_cr_read(struct x86_emulate_ctxt *ctxt)
4122 {
4123 	if (!valid_cr(ctxt->modrm_reg))
4124 		return emulate_ud(ctxt);
4125 
4126 	return X86EMUL_CONTINUE;
4127 }
4128 
4129 static int check_cr_write(struct x86_emulate_ctxt *ctxt)
4130 {
4131 	u64 new_val = ctxt->src.val64;
4132 	int cr = ctxt->modrm_reg;
4133 	u64 efer = 0;
4134 
4135 	static u64 cr_reserved_bits[] = {
4136 		0xffffffff00000000ULL,
4137 		0, 0, 0, /* CR3 checked later */
4138 		CR4_RESERVED_BITS,
4139 		0, 0, 0,
4140 		CR8_RESERVED_BITS,
4141 	};
4142 
4143 	if (!valid_cr(cr))
4144 		return emulate_ud(ctxt);
4145 
4146 	if (new_val & cr_reserved_bits[cr])
4147 		return emulate_gp(ctxt, 0);
4148 
4149 	switch (cr) {
4150 	case 0: {
4151 		u64 cr4;
4152 		if (((new_val & X86_CR0_PG) && !(new_val & X86_CR0_PE)) ||
4153 		    ((new_val & X86_CR0_NW) && !(new_val & X86_CR0_CD)))
4154 			return emulate_gp(ctxt, 0);
4155 
4156 		cr4 = ctxt->ops->get_cr(ctxt, 4);
4157 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4158 
4159 		if ((new_val & X86_CR0_PG) && (efer & EFER_LME) &&
4160 		    !(cr4 & X86_CR4_PAE))
4161 			return emulate_gp(ctxt, 0);
4162 
4163 		break;
4164 		}
4165 	case 3: {
4166 		u64 rsvd = 0;
4167 
4168 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4169 		if (efer & EFER_LMA) {
4170 			u64 maxphyaddr;
4171 			u32 eax, ebx, ecx, edx;
4172 
4173 			eax = 0x80000008;
4174 			ecx = 0;
4175 			if (ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx,
4176 						 &edx, false))
4177 				maxphyaddr = eax & 0xff;
4178 			else
4179 				maxphyaddr = 36;
4180 			rsvd = rsvd_bits(maxphyaddr, 62);
4181 		}
4182 
4183 		if (new_val & rsvd)
4184 			return emulate_gp(ctxt, 0);
4185 
4186 		break;
4187 		}
4188 	case 4: {
4189 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4190 
4191 		if ((efer & EFER_LMA) && !(new_val & X86_CR4_PAE))
4192 			return emulate_gp(ctxt, 0);
4193 
4194 		break;
4195 		}
4196 	}
4197 
4198 	return X86EMUL_CONTINUE;
4199 }
4200 
4201 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4202 {
4203 	unsigned long dr7;
4204 
4205 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4206 
4207 	/* Check if DR7.Global_Enable is set */
4208 	return dr7 & (1 << 13);
4209 }
4210 
4211 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4212 {
4213 	int dr = ctxt->modrm_reg;
4214 	u64 cr4;
4215 
4216 	if (dr > 7)
4217 		return emulate_ud(ctxt);
4218 
4219 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4220 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4221 		return emulate_ud(ctxt);
4222 
4223 	if (check_dr7_gd(ctxt)) {
4224 		ulong dr6;
4225 
4226 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4227 		dr6 &= ~15;
4228 		dr6 |= DR6_BD | DR6_RTM;
4229 		ctxt->ops->set_dr(ctxt, 6, dr6);
4230 		return emulate_db(ctxt);
4231 	}
4232 
4233 	return X86EMUL_CONTINUE;
4234 }
4235 
4236 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4237 {
4238 	u64 new_val = ctxt->src.val64;
4239 	int dr = ctxt->modrm_reg;
4240 
4241 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4242 		return emulate_gp(ctxt, 0);
4243 
4244 	return check_dr_read(ctxt);
4245 }
4246 
4247 static int check_svme(struct x86_emulate_ctxt *ctxt)
4248 {
4249 	u64 efer = 0;
4250 
4251 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4252 
4253 	if (!(efer & EFER_SVME))
4254 		return emulate_ud(ctxt);
4255 
4256 	return X86EMUL_CONTINUE;
4257 }
4258 
4259 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4260 {
4261 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4262 
4263 	/* Valid physical address? */
4264 	if (rax & 0xffff000000000000ULL)
4265 		return emulate_gp(ctxt, 0);
4266 
4267 	return check_svme(ctxt);
4268 }
4269 
4270 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4271 {
4272 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4273 
4274 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4275 		return emulate_ud(ctxt);
4276 
4277 	return X86EMUL_CONTINUE;
4278 }
4279 
4280 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4281 {
4282 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4283 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4284 
4285 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4286 	    ctxt->ops->check_pmc(ctxt, rcx))
4287 		return emulate_gp(ctxt, 0);
4288 
4289 	return X86EMUL_CONTINUE;
4290 }
4291 
4292 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4293 {
4294 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4295 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4296 		return emulate_gp(ctxt, 0);
4297 
4298 	return X86EMUL_CONTINUE;
4299 }
4300 
4301 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4302 {
4303 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4304 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4305 		return emulate_gp(ctxt, 0);
4306 
4307 	return X86EMUL_CONTINUE;
4308 }
4309 
4310 #define D(_y) { .flags = (_y) }
4311 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4312 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4313 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4314 #define N    D(NotImpl)
4315 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4316 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4317 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4318 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4319 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4320 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4321 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4322 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4323 #define II(_f, _e, _i) \
4324 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4325 #define IIP(_f, _e, _i, _p) \
4326 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4327 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4328 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4329 
4330 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4331 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4332 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4333 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4334 #define I2bvIP(_f, _e, _i, _p) \
4335 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4336 
4337 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4338 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4339 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4340 
4341 static const struct opcode group7_rm0[] = {
4342 	N,
4343 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4344 	N, N, N, N, N, N,
4345 };
4346 
4347 static const struct opcode group7_rm1[] = {
4348 	DI(SrcNone | Priv, monitor),
4349 	DI(SrcNone | Priv, mwait),
4350 	N, N, N, N, N, N,
4351 };
4352 
4353 static const struct opcode group7_rm3[] = {
4354 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4355 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4356 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4357 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4358 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4359 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4360 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4361 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4362 };
4363 
4364 static const struct opcode group7_rm7[] = {
4365 	N,
4366 	DIP(SrcNone, rdtscp, check_rdtsc),
4367 	N, N, N, N, N, N,
4368 };
4369 
4370 static const struct opcode group1[] = {
4371 	F(Lock, em_add),
4372 	F(Lock | PageTable, em_or),
4373 	F(Lock, em_adc),
4374 	F(Lock, em_sbb),
4375 	F(Lock | PageTable, em_and),
4376 	F(Lock, em_sub),
4377 	F(Lock, em_xor),
4378 	F(NoWrite, em_cmp),
4379 };
4380 
4381 static const struct opcode group1A[] = {
4382 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4383 };
4384 
4385 static const struct opcode group2[] = {
4386 	F(DstMem | ModRM, em_rol),
4387 	F(DstMem | ModRM, em_ror),
4388 	F(DstMem | ModRM, em_rcl),
4389 	F(DstMem | ModRM, em_rcr),
4390 	F(DstMem | ModRM, em_shl),
4391 	F(DstMem | ModRM, em_shr),
4392 	F(DstMem | ModRM, em_shl),
4393 	F(DstMem | ModRM, em_sar),
4394 };
4395 
4396 static const struct opcode group3[] = {
4397 	F(DstMem | SrcImm | NoWrite, em_test),
4398 	F(DstMem | SrcImm | NoWrite, em_test),
4399 	F(DstMem | SrcNone | Lock, em_not),
4400 	F(DstMem | SrcNone | Lock, em_neg),
4401 	F(DstXacc | Src2Mem, em_mul_ex),
4402 	F(DstXacc | Src2Mem, em_imul_ex),
4403 	F(DstXacc | Src2Mem, em_div_ex),
4404 	F(DstXacc | Src2Mem, em_idiv_ex),
4405 };
4406 
4407 static const struct opcode group4[] = {
4408 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4409 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4410 	N, N, N, N, N, N,
4411 };
4412 
4413 static const struct opcode group5[] = {
4414 	F(DstMem | SrcNone | Lock,		em_inc),
4415 	F(DstMem | SrcNone | Lock,		em_dec),
4416 	I(SrcMem | NearBranch,			em_call_near_abs),
4417 	I(SrcMemFAddr | ImplicitOps,		em_call_far),
4418 	I(SrcMem | NearBranch,			em_jmp_abs),
4419 	I(SrcMemFAddr | ImplicitOps,		em_jmp_far),
4420 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4421 };
4422 
4423 static const struct opcode group6[] = {
4424 	II(Prot | DstMem,	   em_sldt, sldt),
4425 	II(Prot | DstMem,	   em_str, str),
4426 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4427 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4428 	N, N, N, N,
4429 };
4430 
4431 static const struct group_dual group7 = { {
4432 	II(Mov | DstMem,			em_sgdt, sgdt),
4433 	II(Mov | DstMem,			em_sidt, sidt),
4434 	II(SrcMem | Priv,			em_lgdt, lgdt),
4435 	II(SrcMem | Priv,			em_lidt, lidt),
4436 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4437 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4438 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4439 }, {
4440 	EXT(0, group7_rm0),
4441 	EXT(0, group7_rm1),
4442 	N, EXT(0, group7_rm3),
4443 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4444 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4445 	EXT(0, group7_rm7),
4446 } };
4447 
4448 static const struct opcode group8[] = {
4449 	N, N, N, N,
4450 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4451 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4452 	F(DstMem | SrcImmByte | Lock,			em_btr),
4453 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4454 };
4455 
4456 /*
4457  * The "memory" destination is actually always a register, since we come
4458  * from the register case of group9.
4459  */
4460 static const struct gprefix pfx_0f_c7_7 = {
4461 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdtscp),
4462 };
4463 
4464 
4465 static const struct group_dual group9 = { {
4466 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4467 }, {
4468 	N, N, N, N, N, N, N,
4469 	GP(0, &pfx_0f_c7_7),
4470 } };
4471 
4472 static const struct opcode group11[] = {
4473 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4474 	X7(D(Undefined)),
4475 };
4476 
4477 static const struct gprefix pfx_0f_ae_7 = {
4478 	I(SrcMem | ByteOp, em_clflush), N, N, N,
4479 };
4480 
4481 static const struct group_dual group15 = { {
4482 	I(ModRM | Aligned16, em_fxsave),
4483 	I(ModRM | Aligned16, em_fxrstor),
4484 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4485 }, {
4486 	N, N, N, N, N, N, N, N,
4487 } };
4488 
4489 static const struct gprefix pfx_0f_6f_0f_7f = {
4490 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4491 };
4492 
4493 static const struct instr_dual instr_dual_0f_2b = {
4494 	I(0, em_mov), N
4495 };
4496 
4497 static const struct gprefix pfx_0f_2b = {
4498 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4499 };
4500 
4501 static const struct gprefix pfx_0f_28_0f_29 = {
4502 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4503 };
4504 
4505 static const struct gprefix pfx_0f_e7 = {
4506 	N, I(Sse, em_mov), N, N,
4507 };
4508 
4509 static const struct escape escape_d9 = { {
4510 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4511 }, {
4512 	/* 0xC0 - 0xC7 */
4513 	N, N, N, N, N, N, N, N,
4514 	/* 0xC8 - 0xCF */
4515 	N, N, N, N, N, N, N, N,
4516 	/* 0xD0 - 0xC7 */
4517 	N, N, N, N, N, N, N, N,
4518 	/* 0xD8 - 0xDF */
4519 	N, N, N, N, N, N, N, N,
4520 	/* 0xE0 - 0xE7 */
4521 	N, N, N, N, N, N, N, N,
4522 	/* 0xE8 - 0xEF */
4523 	N, N, N, N, N, N, N, N,
4524 	/* 0xF0 - 0xF7 */
4525 	N, N, N, N, N, N, N, N,
4526 	/* 0xF8 - 0xFF */
4527 	N, N, N, N, N, N, N, N,
4528 } };
4529 
4530 static const struct escape escape_db = { {
4531 	N, N, N, N, N, N, N, N,
4532 }, {
4533 	/* 0xC0 - 0xC7 */
4534 	N, N, N, N, N, N, N, N,
4535 	/* 0xC8 - 0xCF */
4536 	N, N, N, N, N, N, N, N,
4537 	/* 0xD0 - 0xC7 */
4538 	N, N, N, N, N, N, N, N,
4539 	/* 0xD8 - 0xDF */
4540 	N, N, N, N, N, N, N, N,
4541 	/* 0xE0 - 0xE7 */
4542 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4543 	/* 0xE8 - 0xEF */
4544 	N, N, N, N, N, N, N, N,
4545 	/* 0xF0 - 0xF7 */
4546 	N, N, N, N, N, N, N, N,
4547 	/* 0xF8 - 0xFF */
4548 	N, N, N, N, N, N, N, N,
4549 } };
4550 
4551 static const struct escape escape_dd = { {
4552 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4553 }, {
4554 	/* 0xC0 - 0xC7 */
4555 	N, N, N, N, N, N, N, N,
4556 	/* 0xC8 - 0xCF */
4557 	N, N, N, N, N, N, N, N,
4558 	/* 0xD0 - 0xC7 */
4559 	N, N, N, N, N, N, N, N,
4560 	/* 0xD8 - 0xDF */
4561 	N, N, N, N, N, N, N, N,
4562 	/* 0xE0 - 0xE7 */
4563 	N, N, N, N, N, N, N, N,
4564 	/* 0xE8 - 0xEF */
4565 	N, N, N, N, N, N, N, N,
4566 	/* 0xF0 - 0xF7 */
4567 	N, N, N, N, N, N, N, N,
4568 	/* 0xF8 - 0xFF */
4569 	N, N, N, N, N, N, N, N,
4570 } };
4571 
4572 static const struct instr_dual instr_dual_0f_c3 = {
4573 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4574 };
4575 
4576 static const struct mode_dual mode_dual_63 = {
4577 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4578 };
4579 
4580 static const struct opcode opcode_table[256] = {
4581 	/* 0x00 - 0x07 */
4582 	F6ALU(Lock, em_add),
4583 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4584 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4585 	/* 0x08 - 0x0F */
4586 	F6ALU(Lock | PageTable, em_or),
4587 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4588 	N,
4589 	/* 0x10 - 0x17 */
4590 	F6ALU(Lock, em_adc),
4591 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4592 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4593 	/* 0x18 - 0x1F */
4594 	F6ALU(Lock, em_sbb),
4595 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4596 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4597 	/* 0x20 - 0x27 */
4598 	F6ALU(Lock | PageTable, em_and), N, N,
4599 	/* 0x28 - 0x2F */
4600 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4601 	/* 0x30 - 0x37 */
4602 	F6ALU(Lock, em_xor), N, N,
4603 	/* 0x38 - 0x3F */
4604 	F6ALU(NoWrite, em_cmp), N, N,
4605 	/* 0x40 - 0x4F */
4606 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4607 	/* 0x50 - 0x57 */
4608 	X8(I(SrcReg | Stack, em_push)),
4609 	/* 0x58 - 0x5F */
4610 	X8(I(DstReg | Stack, em_pop)),
4611 	/* 0x60 - 0x67 */
4612 	I(ImplicitOps | Stack | No64, em_pusha),
4613 	I(ImplicitOps | Stack | No64, em_popa),
4614 	N, MD(ModRM, &mode_dual_63),
4615 	N, N, N, N,
4616 	/* 0x68 - 0x6F */
4617 	I(SrcImm | Mov | Stack, em_push),
4618 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4619 	I(SrcImmByte | Mov | Stack, em_push),
4620 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4621 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4622 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4623 	/* 0x70 - 0x7F */
4624 	X16(D(SrcImmByte | NearBranch)),
4625 	/* 0x80 - 0x87 */
4626 	G(ByteOp | DstMem | SrcImm, group1),
4627 	G(DstMem | SrcImm, group1),
4628 	G(ByteOp | DstMem | SrcImm | No64, group1),
4629 	G(DstMem | SrcImmByte, group1),
4630 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4631 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4632 	/* 0x88 - 0x8F */
4633 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4634 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4635 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4636 	D(ModRM | SrcMem | NoAccess | DstReg),
4637 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4638 	G(0, group1A),
4639 	/* 0x90 - 0x97 */
4640 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4641 	/* 0x98 - 0x9F */
4642 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4643 	I(SrcImmFAddr | No64, em_call_far), N,
4644 	II(ImplicitOps | Stack, em_pushf, pushf),
4645 	II(ImplicitOps | Stack, em_popf, popf),
4646 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4647 	/* 0xA0 - 0xA7 */
4648 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4649 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4650 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4651 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4652 	/* 0xA8 - 0xAF */
4653 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4654 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4655 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4656 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4657 	/* 0xB0 - 0xB7 */
4658 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4659 	/* 0xB8 - 0xBF */
4660 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4661 	/* 0xC0 - 0xC7 */
4662 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4663 	I(ImplicitOps | NearBranch | SrcImmU16, em_ret_near_imm),
4664 	I(ImplicitOps | NearBranch, em_ret),
4665 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4666 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4667 	G(ByteOp, group11), G(0, group11),
4668 	/* 0xC8 - 0xCF */
4669 	I(Stack | SrcImmU16 | Src2ImmByte, em_enter), I(Stack, em_leave),
4670 	I(ImplicitOps | SrcImmU16, em_ret_far_imm),
4671 	I(ImplicitOps, em_ret_far),
4672 	D(ImplicitOps), DI(SrcImmByte, intn),
4673 	D(ImplicitOps | No64), II(ImplicitOps, em_iret, iret),
4674 	/* 0xD0 - 0xD7 */
4675 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4676 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4677 	I(DstAcc | SrcImmUByte | No64, em_aam),
4678 	I(DstAcc | SrcImmUByte | No64, em_aad),
4679 	F(DstAcc | ByteOp | No64, em_salc),
4680 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4681 	/* 0xD8 - 0xDF */
4682 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4683 	/* 0xE0 - 0xE7 */
4684 	X3(I(SrcImmByte | NearBranch, em_loop)),
4685 	I(SrcImmByte | NearBranch, em_jcxz),
4686 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4687 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4688 	/* 0xE8 - 0xEF */
4689 	I(SrcImm | NearBranch, em_call), D(SrcImm | ImplicitOps | NearBranch),
4690 	I(SrcImmFAddr | No64, em_jmp_far),
4691 	D(SrcImmByte | ImplicitOps | NearBranch),
4692 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4693 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4694 	/* 0xF0 - 0xF7 */
4695 	N, DI(ImplicitOps, icebp), N, N,
4696 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4697 	G(ByteOp, group3), G(0, group3),
4698 	/* 0xF8 - 0xFF */
4699 	D(ImplicitOps), D(ImplicitOps),
4700 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4701 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4702 };
4703 
4704 static const struct opcode twobyte_table[256] = {
4705 	/* 0x00 - 0x0F */
4706 	G(0, group6), GD(0, &group7), N, N,
4707 	N, I(ImplicitOps | EmulateOnUD, em_syscall),
4708 	II(ImplicitOps | Priv, em_clts, clts), N,
4709 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4710 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4711 	/* 0x10 - 0x1F */
4712 	N, N, N, N, N, N, N, N,
4713 	D(ImplicitOps | ModRM | SrcMem | NoAccess),
4714 	N, N, N, N, N, N, D(ImplicitOps | ModRM | SrcMem | NoAccess),
4715 	/* 0x20 - 0x2F */
4716 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_read),
4717 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4718 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4719 						check_cr_write),
4720 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4721 						check_dr_write),
4722 	N, N, N, N,
4723 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4724 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4725 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4726 	N, N, N, N,
4727 	/* 0x30 - 0x3F */
4728 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4729 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4730 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4731 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4732 	I(ImplicitOps | EmulateOnUD, em_sysenter),
4733 	I(ImplicitOps | Priv | EmulateOnUD, em_sysexit),
4734 	N, N,
4735 	N, N, N, N, N, N, N, N,
4736 	/* 0x40 - 0x4F */
4737 	X16(D(DstReg | SrcMem | ModRM)),
4738 	/* 0x50 - 0x5F */
4739 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4740 	/* 0x60 - 0x6F */
4741 	N, N, N, N,
4742 	N, N, N, N,
4743 	N, N, N, N,
4744 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4745 	/* 0x70 - 0x7F */
4746 	N, N, N, N,
4747 	N, N, N, N,
4748 	N, N, N, N,
4749 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4750 	/* 0x80 - 0x8F */
4751 	X16(D(SrcImm | NearBranch)),
4752 	/* 0x90 - 0x9F */
4753 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4754 	/* 0xA0 - 0xA7 */
4755 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4756 	II(ImplicitOps, em_cpuid, cpuid),
4757 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4758 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4759 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4760 	/* 0xA8 - 0xAF */
4761 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4762 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4763 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4764 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4765 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4766 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4767 	/* 0xB0 - 0xB7 */
4768 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4769 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4770 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4771 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4772 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4773 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4774 	/* 0xB8 - 0xBF */
4775 	N, N,
4776 	G(BitOp, group8),
4777 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4778 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4779 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4780 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4781 	/* 0xC0 - 0xC7 */
4782 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4783 	N, ID(0, &instr_dual_0f_c3),
4784 	N, N, N, GD(0, &group9),
4785 	/* 0xC8 - 0xCF */
4786 	X8(I(DstReg, em_bswap)),
4787 	/* 0xD0 - 0xDF */
4788 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4789 	/* 0xE0 - 0xEF */
4790 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4791 	N, N, N, N, N, N, N, N,
4792 	/* 0xF0 - 0xFF */
4793 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4794 };
4795 
4796 static const struct instr_dual instr_dual_0f_38_f0 = {
4797 	I(DstReg | SrcMem | Mov, em_movbe), N
4798 };
4799 
4800 static const struct instr_dual instr_dual_0f_38_f1 = {
4801 	I(DstMem | SrcReg | Mov, em_movbe), N
4802 };
4803 
4804 static const struct gprefix three_byte_0f_38_f0 = {
4805 	ID(0, &instr_dual_0f_38_f0), N, N, N
4806 };
4807 
4808 static const struct gprefix three_byte_0f_38_f1 = {
4809 	ID(0, &instr_dual_0f_38_f1), N, N, N
4810 };
4811 
4812 /*
4813  * Insns below are selected by the prefix which indexed by the third opcode
4814  * byte.
4815  */
4816 static const struct opcode opcode_map_0f_38[256] = {
4817 	/* 0x00 - 0x7f */
4818 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4819 	/* 0x80 - 0xef */
4820 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4821 	/* 0xf0 - 0xf1 */
4822 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4823 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4824 	/* 0xf2 - 0xff */
4825 	N, N, X4(N), X8(N)
4826 };
4827 
4828 #undef D
4829 #undef N
4830 #undef G
4831 #undef GD
4832 #undef I
4833 #undef GP
4834 #undef EXT
4835 #undef MD
4836 #undef ID
4837 
4838 #undef D2bv
4839 #undef D2bvIP
4840 #undef I2bv
4841 #undef I2bvIP
4842 #undef I6ALU
4843 
4844 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4845 {
4846 	unsigned size;
4847 
4848 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4849 	if (size == 8)
4850 		size = 4;
4851 	return size;
4852 }
4853 
4854 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4855 		      unsigned size, bool sign_extension)
4856 {
4857 	int rc = X86EMUL_CONTINUE;
4858 
4859 	op->type = OP_IMM;
4860 	op->bytes = size;
4861 	op->addr.mem.ea = ctxt->_eip;
4862 	/* NB. Immediates are sign-extended as necessary. */
4863 	switch (op->bytes) {
4864 	case 1:
4865 		op->val = insn_fetch(s8, ctxt);
4866 		break;
4867 	case 2:
4868 		op->val = insn_fetch(s16, ctxt);
4869 		break;
4870 	case 4:
4871 		op->val = insn_fetch(s32, ctxt);
4872 		break;
4873 	case 8:
4874 		op->val = insn_fetch(s64, ctxt);
4875 		break;
4876 	}
4877 	if (!sign_extension) {
4878 		switch (op->bytes) {
4879 		case 1:
4880 			op->val &= 0xff;
4881 			break;
4882 		case 2:
4883 			op->val &= 0xffff;
4884 			break;
4885 		case 4:
4886 			op->val &= 0xffffffff;
4887 			break;
4888 		}
4889 	}
4890 done:
4891 	return rc;
4892 }
4893 
4894 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4895 			  unsigned d)
4896 {
4897 	int rc = X86EMUL_CONTINUE;
4898 
4899 	switch (d) {
4900 	case OpReg:
4901 		decode_register_operand(ctxt, op);
4902 		break;
4903 	case OpImmUByte:
4904 		rc = decode_imm(ctxt, op, 1, false);
4905 		break;
4906 	case OpMem:
4907 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4908 	mem_common:
4909 		*op = ctxt->memop;
4910 		ctxt->memopp = op;
4911 		if (ctxt->d & BitOp)
4912 			fetch_bit_operand(ctxt);
4913 		op->orig_val = op->val;
4914 		break;
4915 	case OpMem64:
4916 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4917 		goto mem_common;
4918 	case OpAcc:
4919 		op->type = OP_REG;
4920 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4921 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4922 		fetch_register_operand(op);
4923 		op->orig_val = op->val;
4924 		break;
4925 	case OpAccLo:
4926 		op->type = OP_REG;
4927 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4928 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4929 		fetch_register_operand(op);
4930 		op->orig_val = op->val;
4931 		break;
4932 	case OpAccHi:
4933 		if (ctxt->d & ByteOp) {
4934 			op->type = OP_NONE;
4935 			break;
4936 		}
4937 		op->type = OP_REG;
4938 		op->bytes = ctxt->op_bytes;
4939 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4940 		fetch_register_operand(op);
4941 		op->orig_val = op->val;
4942 		break;
4943 	case OpDI:
4944 		op->type = OP_MEM;
4945 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4946 		op->addr.mem.ea =
4947 			register_address(ctxt, VCPU_REGS_RDI);
4948 		op->addr.mem.seg = VCPU_SREG_ES;
4949 		op->val = 0;
4950 		op->count = 1;
4951 		break;
4952 	case OpDX:
4953 		op->type = OP_REG;
4954 		op->bytes = 2;
4955 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4956 		fetch_register_operand(op);
4957 		break;
4958 	case OpCL:
4959 		op->type = OP_IMM;
4960 		op->bytes = 1;
4961 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4962 		break;
4963 	case OpImmByte:
4964 		rc = decode_imm(ctxt, op, 1, true);
4965 		break;
4966 	case OpOne:
4967 		op->type = OP_IMM;
4968 		op->bytes = 1;
4969 		op->val = 1;
4970 		break;
4971 	case OpImm:
4972 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4973 		break;
4974 	case OpImm64:
4975 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4976 		break;
4977 	case OpMem8:
4978 		ctxt->memop.bytes = 1;
4979 		if (ctxt->memop.type == OP_REG) {
4980 			ctxt->memop.addr.reg = decode_register(ctxt,
4981 					ctxt->modrm_rm, true);
4982 			fetch_register_operand(&ctxt->memop);
4983 		}
4984 		goto mem_common;
4985 	case OpMem16:
4986 		ctxt->memop.bytes = 2;
4987 		goto mem_common;
4988 	case OpMem32:
4989 		ctxt->memop.bytes = 4;
4990 		goto mem_common;
4991 	case OpImmU16:
4992 		rc = decode_imm(ctxt, op, 2, false);
4993 		break;
4994 	case OpImmU:
4995 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4996 		break;
4997 	case OpSI:
4998 		op->type = OP_MEM;
4999 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5000 		op->addr.mem.ea =
5001 			register_address(ctxt, VCPU_REGS_RSI);
5002 		op->addr.mem.seg = ctxt->seg_override;
5003 		op->val = 0;
5004 		op->count = 1;
5005 		break;
5006 	case OpXLat:
5007 		op->type = OP_MEM;
5008 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5009 		op->addr.mem.ea =
5010 			address_mask(ctxt,
5011 				reg_read(ctxt, VCPU_REGS_RBX) +
5012 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
5013 		op->addr.mem.seg = ctxt->seg_override;
5014 		op->val = 0;
5015 		break;
5016 	case OpImmFAddr:
5017 		op->type = OP_IMM;
5018 		op->addr.mem.ea = ctxt->_eip;
5019 		op->bytes = ctxt->op_bytes + 2;
5020 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
5021 		break;
5022 	case OpMemFAddr:
5023 		ctxt->memop.bytes = ctxt->op_bytes + 2;
5024 		goto mem_common;
5025 	case OpES:
5026 		op->type = OP_IMM;
5027 		op->val = VCPU_SREG_ES;
5028 		break;
5029 	case OpCS:
5030 		op->type = OP_IMM;
5031 		op->val = VCPU_SREG_CS;
5032 		break;
5033 	case OpSS:
5034 		op->type = OP_IMM;
5035 		op->val = VCPU_SREG_SS;
5036 		break;
5037 	case OpDS:
5038 		op->type = OP_IMM;
5039 		op->val = VCPU_SREG_DS;
5040 		break;
5041 	case OpFS:
5042 		op->type = OP_IMM;
5043 		op->val = VCPU_SREG_FS;
5044 		break;
5045 	case OpGS:
5046 		op->type = OP_IMM;
5047 		op->val = VCPU_SREG_GS;
5048 		break;
5049 	case OpImplicit:
5050 		/* Special instructions do their own operand decoding. */
5051 	default:
5052 		op->type = OP_NONE; /* Disable writeback. */
5053 		break;
5054 	}
5055 
5056 done:
5057 	return rc;
5058 }
5059 
5060 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len)
5061 {
5062 	int rc = X86EMUL_CONTINUE;
5063 	int mode = ctxt->mode;
5064 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
5065 	bool op_prefix = false;
5066 	bool has_seg_override = false;
5067 	struct opcode opcode;
5068 	u16 dummy;
5069 	struct desc_struct desc;
5070 
5071 	ctxt->memop.type = OP_NONE;
5072 	ctxt->memopp = NULL;
5073 	ctxt->_eip = ctxt->eip;
5074 	ctxt->fetch.ptr = ctxt->fetch.data;
5075 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
5076 	ctxt->opcode_len = 1;
5077 	if (insn_len > 0)
5078 		memcpy(ctxt->fetch.data, insn, insn_len);
5079 	else {
5080 		rc = __do_insn_fetch_bytes(ctxt, 1);
5081 		if (rc != X86EMUL_CONTINUE)
5082 			return rc;
5083 	}
5084 
5085 	switch (mode) {
5086 	case X86EMUL_MODE_REAL:
5087 	case X86EMUL_MODE_VM86:
5088 		def_op_bytes = def_ad_bytes = 2;
5089 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
5090 		if (desc.d)
5091 			def_op_bytes = def_ad_bytes = 4;
5092 		break;
5093 	case X86EMUL_MODE_PROT16:
5094 		def_op_bytes = def_ad_bytes = 2;
5095 		break;
5096 	case X86EMUL_MODE_PROT32:
5097 		def_op_bytes = def_ad_bytes = 4;
5098 		break;
5099 #ifdef CONFIG_X86_64
5100 	case X86EMUL_MODE_PROT64:
5101 		def_op_bytes = 4;
5102 		def_ad_bytes = 8;
5103 		break;
5104 #endif
5105 	default:
5106 		return EMULATION_FAILED;
5107 	}
5108 
5109 	ctxt->op_bytes = def_op_bytes;
5110 	ctxt->ad_bytes = def_ad_bytes;
5111 
5112 	/* Legacy prefixes. */
5113 	for (;;) {
5114 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5115 		case 0x66:	/* operand-size override */
5116 			op_prefix = true;
5117 			/* switch between 2/4 bytes */
5118 			ctxt->op_bytes = def_op_bytes ^ 6;
5119 			break;
5120 		case 0x67:	/* address-size override */
5121 			if (mode == X86EMUL_MODE_PROT64)
5122 				/* switch between 4/8 bytes */
5123 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5124 			else
5125 				/* switch between 2/4 bytes */
5126 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5127 			break;
5128 		case 0x26:	/* ES override */
5129 		case 0x2e:	/* CS override */
5130 		case 0x36:	/* SS override */
5131 		case 0x3e:	/* DS override */
5132 			has_seg_override = true;
5133 			ctxt->seg_override = (ctxt->b >> 3) & 3;
5134 			break;
5135 		case 0x64:	/* FS override */
5136 		case 0x65:	/* GS override */
5137 			has_seg_override = true;
5138 			ctxt->seg_override = ctxt->b & 7;
5139 			break;
5140 		case 0x40 ... 0x4f: /* REX */
5141 			if (mode != X86EMUL_MODE_PROT64)
5142 				goto done_prefixes;
5143 			ctxt->rex_prefix = ctxt->b;
5144 			continue;
5145 		case 0xf0:	/* LOCK */
5146 			ctxt->lock_prefix = 1;
5147 			break;
5148 		case 0xf2:	/* REPNE/REPNZ */
5149 		case 0xf3:	/* REP/REPE/REPZ */
5150 			ctxt->rep_prefix = ctxt->b;
5151 			break;
5152 		default:
5153 			goto done_prefixes;
5154 		}
5155 
5156 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5157 
5158 		ctxt->rex_prefix = 0;
5159 	}
5160 
5161 done_prefixes:
5162 
5163 	/* REX prefix. */
5164 	if (ctxt->rex_prefix & 8)
5165 		ctxt->op_bytes = 8;	/* REX.W */
5166 
5167 	/* Opcode byte(s). */
5168 	opcode = opcode_table[ctxt->b];
5169 	/* Two-byte opcode? */
5170 	if (ctxt->b == 0x0f) {
5171 		ctxt->opcode_len = 2;
5172 		ctxt->b = insn_fetch(u8, ctxt);
5173 		opcode = twobyte_table[ctxt->b];
5174 
5175 		/* 0F_38 opcode map */
5176 		if (ctxt->b == 0x38) {
5177 			ctxt->opcode_len = 3;
5178 			ctxt->b = insn_fetch(u8, ctxt);
5179 			opcode = opcode_map_0f_38[ctxt->b];
5180 		}
5181 	}
5182 	ctxt->d = opcode.flags;
5183 
5184 	if (ctxt->d & ModRM)
5185 		ctxt->modrm = insn_fetch(u8, ctxt);
5186 
5187 	/* vex-prefix instructions are not implemented */
5188 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5189 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5190 		ctxt->d = NotImpl;
5191 	}
5192 
5193 	while (ctxt->d & GroupMask) {
5194 		switch (ctxt->d & GroupMask) {
5195 		case Group:
5196 			goffset = (ctxt->modrm >> 3) & 7;
5197 			opcode = opcode.u.group[goffset];
5198 			break;
5199 		case GroupDual:
5200 			goffset = (ctxt->modrm >> 3) & 7;
5201 			if ((ctxt->modrm >> 6) == 3)
5202 				opcode = opcode.u.gdual->mod3[goffset];
5203 			else
5204 				opcode = opcode.u.gdual->mod012[goffset];
5205 			break;
5206 		case RMExt:
5207 			goffset = ctxt->modrm & 7;
5208 			opcode = opcode.u.group[goffset];
5209 			break;
5210 		case Prefix:
5211 			if (ctxt->rep_prefix && op_prefix)
5212 				return EMULATION_FAILED;
5213 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5214 			switch (simd_prefix) {
5215 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5216 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5217 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5218 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5219 			}
5220 			break;
5221 		case Escape:
5222 			if (ctxt->modrm > 0xbf)
5223 				opcode = opcode.u.esc->high[ctxt->modrm - 0xc0];
5224 			else
5225 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5226 			break;
5227 		case InstrDual:
5228 			if ((ctxt->modrm >> 6) == 3)
5229 				opcode = opcode.u.idual->mod3;
5230 			else
5231 				opcode = opcode.u.idual->mod012;
5232 			break;
5233 		case ModeDual:
5234 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5235 				opcode = opcode.u.mdual->mode64;
5236 			else
5237 				opcode = opcode.u.mdual->mode32;
5238 			break;
5239 		default:
5240 			return EMULATION_FAILED;
5241 		}
5242 
5243 		ctxt->d &= ~(u64)GroupMask;
5244 		ctxt->d |= opcode.flags;
5245 	}
5246 
5247 	/* Unrecognised? */
5248 	if (ctxt->d == 0)
5249 		return EMULATION_FAILED;
5250 
5251 	ctxt->execute = opcode.u.execute;
5252 
5253 	if (unlikely(ctxt->ud) && likely(!(ctxt->d & EmulateOnUD)))
5254 		return EMULATION_FAILED;
5255 
5256 	if (unlikely(ctxt->d &
5257 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5258 	     No16))) {
5259 		/*
5260 		 * These are copied unconditionally here, and checked unconditionally
5261 		 * in x86_emulate_insn.
5262 		 */
5263 		ctxt->check_perm = opcode.check_perm;
5264 		ctxt->intercept = opcode.intercept;
5265 
5266 		if (ctxt->d & NotImpl)
5267 			return EMULATION_FAILED;
5268 
5269 		if (mode == X86EMUL_MODE_PROT64) {
5270 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5271 				ctxt->op_bytes = 8;
5272 			else if (ctxt->d & NearBranch)
5273 				ctxt->op_bytes = 8;
5274 		}
5275 
5276 		if (ctxt->d & Op3264) {
5277 			if (mode == X86EMUL_MODE_PROT64)
5278 				ctxt->op_bytes = 8;
5279 			else
5280 				ctxt->op_bytes = 4;
5281 		}
5282 
5283 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5284 			ctxt->op_bytes = 4;
5285 
5286 		if (ctxt->d & Sse)
5287 			ctxt->op_bytes = 16;
5288 		else if (ctxt->d & Mmx)
5289 			ctxt->op_bytes = 8;
5290 	}
5291 
5292 	/* ModRM and SIB bytes. */
5293 	if (ctxt->d & ModRM) {
5294 		rc = decode_modrm(ctxt, &ctxt->memop);
5295 		if (!has_seg_override) {
5296 			has_seg_override = true;
5297 			ctxt->seg_override = ctxt->modrm_seg;
5298 		}
5299 	} else if (ctxt->d & MemAbs)
5300 		rc = decode_abs(ctxt, &ctxt->memop);
5301 	if (rc != X86EMUL_CONTINUE)
5302 		goto done;
5303 
5304 	if (!has_seg_override)
5305 		ctxt->seg_override = VCPU_SREG_DS;
5306 
5307 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5308 
5309 	/*
5310 	 * Decode and fetch the source operand: register, memory
5311 	 * or immediate.
5312 	 */
5313 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5314 	if (rc != X86EMUL_CONTINUE)
5315 		goto done;
5316 
5317 	/*
5318 	 * Decode and fetch the second source operand: register, memory
5319 	 * or immediate.
5320 	 */
5321 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5322 	if (rc != X86EMUL_CONTINUE)
5323 		goto done;
5324 
5325 	/* Decode and fetch the destination operand: register or memory. */
5326 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5327 
5328 	if (ctxt->rip_relative && likely(ctxt->memopp))
5329 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5330 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5331 
5332 done:
5333 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5334 }
5335 
5336 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5337 {
5338 	return ctxt->d & PageTable;
5339 }
5340 
5341 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5342 {
5343 	/* The second termination condition only applies for REPE
5344 	 * and REPNE. Test if the repeat string operation prefix is
5345 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5346 	 * corresponding termination condition according to:
5347 	 * 	- if REPE/REPZ and ZF = 0 then done
5348 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5349 	 */
5350 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5351 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5352 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5353 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5354 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5355 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5356 		return true;
5357 
5358 	return false;
5359 }
5360 
5361 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5362 {
5363 	int rc;
5364 
5365 	rc = asm_safe("fwait");
5366 
5367 	if (unlikely(rc != X86EMUL_CONTINUE))
5368 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5369 
5370 	return X86EMUL_CONTINUE;
5371 }
5372 
5373 static void fetch_possible_mmx_operand(struct x86_emulate_ctxt *ctxt,
5374 				       struct operand *op)
5375 {
5376 	if (op->type == OP_MM)
5377 		read_mmx_reg(ctxt, &op->mm_val, op->addr.mm);
5378 }
5379 
5380 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *))
5381 {
5382 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5383 
5384 	if (!(ctxt->d & ByteOp))
5385 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5386 
5387 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5388 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5389 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5390 	    : "c"(ctxt->src2.val));
5391 
5392 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5393 	if (!fop) /* exception is returned in fop variable */
5394 		return emulate_de(ctxt);
5395 	return X86EMUL_CONTINUE;
5396 }
5397 
5398 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5399 {
5400 	memset(&ctxt->rip_relative, 0,
5401 	       (void *)&ctxt->modrm - (void *)&ctxt->rip_relative);
5402 
5403 	ctxt->io_read.pos = 0;
5404 	ctxt->io_read.end = 0;
5405 	ctxt->mem_read.end = 0;
5406 }
5407 
5408 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5409 {
5410 	const struct x86_emulate_ops *ops = ctxt->ops;
5411 	int rc = X86EMUL_CONTINUE;
5412 	int saved_dst_type = ctxt->dst.type;
5413 	unsigned emul_flags;
5414 
5415 	ctxt->mem_read.pos = 0;
5416 
5417 	/* LOCK prefix is allowed only with some instructions */
5418 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5419 		rc = emulate_ud(ctxt);
5420 		goto done;
5421 	}
5422 
5423 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5424 		rc = emulate_ud(ctxt);
5425 		goto done;
5426 	}
5427 
5428 	emul_flags = ctxt->ops->get_hflags(ctxt);
5429 	if (unlikely(ctxt->d &
5430 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5431 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5432 				(ctxt->d & Undefined)) {
5433 			rc = emulate_ud(ctxt);
5434 			goto done;
5435 		}
5436 
5437 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5438 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5439 			rc = emulate_ud(ctxt);
5440 			goto done;
5441 		}
5442 
5443 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5444 			rc = emulate_nm(ctxt);
5445 			goto done;
5446 		}
5447 
5448 		if (ctxt->d & Mmx) {
5449 			rc = flush_pending_x87_faults(ctxt);
5450 			if (rc != X86EMUL_CONTINUE)
5451 				goto done;
5452 			/*
5453 			 * Now that we know the fpu is exception safe, we can fetch
5454 			 * operands from it.
5455 			 */
5456 			fetch_possible_mmx_operand(ctxt, &ctxt->src);
5457 			fetch_possible_mmx_operand(ctxt, &ctxt->src2);
5458 			if (!(ctxt->d & Mov))
5459 				fetch_possible_mmx_operand(ctxt, &ctxt->dst);
5460 		}
5461 
5462 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5463 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5464 						      X86_ICPT_PRE_EXCEPT);
5465 			if (rc != X86EMUL_CONTINUE)
5466 				goto done;
5467 		}
5468 
5469 		/* Instruction can only be executed in protected mode */
5470 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5471 			rc = emulate_ud(ctxt);
5472 			goto done;
5473 		}
5474 
5475 		/* Privileged instruction can be executed only in CPL=0 */
5476 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5477 			if (ctxt->d & PrivUD)
5478 				rc = emulate_ud(ctxt);
5479 			else
5480 				rc = emulate_gp(ctxt, 0);
5481 			goto done;
5482 		}
5483 
5484 		/* Do instruction specific permission checks */
5485 		if (ctxt->d & CheckPerm) {
5486 			rc = ctxt->check_perm(ctxt);
5487 			if (rc != X86EMUL_CONTINUE)
5488 				goto done;
5489 		}
5490 
5491 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5492 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5493 						      X86_ICPT_POST_EXCEPT);
5494 			if (rc != X86EMUL_CONTINUE)
5495 				goto done;
5496 		}
5497 
5498 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5499 			/* All REP prefixes have the same first termination condition */
5500 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5501 				string_registers_quirk(ctxt);
5502 				ctxt->eip = ctxt->_eip;
5503 				ctxt->eflags &= ~X86_EFLAGS_RF;
5504 				goto done;
5505 			}
5506 		}
5507 	}
5508 
5509 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5510 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5511 				    ctxt->src.valptr, ctxt->src.bytes);
5512 		if (rc != X86EMUL_CONTINUE)
5513 			goto done;
5514 		ctxt->src.orig_val64 = ctxt->src.val64;
5515 	}
5516 
5517 	if (ctxt->src2.type == OP_MEM) {
5518 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5519 				    &ctxt->src2.val, ctxt->src2.bytes);
5520 		if (rc != X86EMUL_CONTINUE)
5521 			goto done;
5522 	}
5523 
5524 	if ((ctxt->d & DstMask) == ImplicitOps)
5525 		goto special_insn;
5526 
5527 
5528 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5529 		/* optimisation - avoid slow emulated read if Mov */
5530 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5531 				   &ctxt->dst.val, ctxt->dst.bytes);
5532 		if (rc != X86EMUL_CONTINUE) {
5533 			if (!(ctxt->d & NoWrite) &&
5534 			    rc == X86EMUL_PROPAGATE_FAULT &&
5535 			    ctxt->exception.vector == PF_VECTOR)
5536 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5537 			goto done;
5538 		}
5539 	}
5540 	/* Copy full 64-bit value for CMPXCHG8B.  */
5541 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5542 
5543 special_insn:
5544 
5545 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5546 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5547 					      X86_ICPT_POST_MEMACCESS);
5548 		if (rc != X86EMUL_CONTINUE)
5549 			goto done;
5550 	}
5551 
5552 	if (ctxt->rep_prefix && (ctxt->d & String))
5553 		ctxt->eflags |= X86_EFLAGS_RF;
5554 	else
5555 		ctxt->eflags &= ~X86_EFLAGS_RF;
5556 
5557 	if (ctxt->execute) {
5558 		if (ctxt->d & Fastop) {
5559 			void (*fop)(struct fastop *) = (void *)ctxt->execute;
5560 			rc = fastop(ctxt, fop);
5561 			if (rc != X86EMUL_CONTINUE)
5562 				goto done;
5563 			goto writeback;
5564 		}
5565 		rc = ctxt->execute(ctxt);
5566 		if (rc != X86EMUL_CONTINUE)
5567 			goto done;
5568 		goto writeback;
5569 	}
5570 
5571 	if (ctxt->opcode_len == 2)
5572 		goto twobyte_insn;
5573 	else if (ctxt->opcode_len == 3)
5574 		goto threebyte_insn;
5575 
5576 	switch (ctxt->b) {
5577 	case 0x70 ... 0x7f: /* jcc (short) */
5578 		if (test_cc(ctxt->b, ctxt->eflags))
5579 			rc = jmp_rel(ctxt, ctxt->src.val);
5580 		break;
5581 	case 0x8d: /* lea r16/r32, m */
5582 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5583 		break;
5584 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5585 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5586 			ctxt->dst.type = OP_NONE;
5587 		else
5588 			rc = em_xchg(ctxt);
5589 		break;
5590 	case 0x98: /* cbw/cwde/cdqe */
5591 		switch (ctxt->op_bytes) {
5592 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5593 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5594 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5595 		}
5596 		break;
5597 	case 0xcc:		/* int3 */
5598 		rc = emulate_int(ctxt, 3);
5599 		break;
5600 	case 0xcd:		/* int n */
5601 		rc = emulate_int(ctxt, ctxt->src.val);
5602 		break;
5603 	case 0xce:		/* into */
5604 		if (ctxt->eflags & X86_EFLAGS_OF)
5605 			rc = emulate_int(ctxt, 4);
5606 		break;
5607 	case 0xe9: /* jmp rel */
5608 	case 0xeb: /* jmp rel short */
5609 		rc = jmp_rel(ctxt, ctxt->src.val);
5610 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5611 		break;
5612 	case 0xf4:              /* hlt */
5613 		ctxt->ops->halt(ctxt);
5614 		break;
5615 	case 0xf5:	/* cmc */
5616 		/* complement carry flag from eflags reg */
5617 		ctxt->eflags ^= X86_EFLAGS_CF;
5618 		break;
5619 	case 0xf8: /* clc */
5620 		ctxt->eflags &= ~X86_EFLAGS_CF;
5621 		break;
5622 	case 0xf9: /* stc */
5623 		ctxt->eflags |= X86_EFLAGS_CF;
5624 		break;
5625 	case 0xfc: /* cld */
5626 		ctxt->eflags &= ~X86_EFLAGS_DF;
5627 		break;
5628 	case 0xfd: /* std */
5629 		ctxt->eflags |= X86_EFLAGS_DF;
5630 		break;
5631 	default:
5632 		goto cannot_emulate;
5633 	}
5634 
5635 	if (rc != X86EMUL_CONTINUE)
5636 		goto done;
5637 
5638 writeback:
5639 	if (ctxt->d & SrcWrite) {
5640 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5641 		rc = writeback(ctxt, &ctxt->src);
5642 		if (rc != X86EMUL_CONTINUE)
5643 			goto done;
5644 	}
5645 	if (!(ctxt->d & NoWrite)) {
5646 		rc = writeback(ctxt, &ctxt->dst);
5647 		if (rc != X86EMUL_CONTINUE)
5648 			goto done;
5649 	}
5650 
5651 	/*
5652 	 * restore dst type in case the decoding will be reused
5653 	 * (happens for string instruction )
5654 	 */
5655 	ctxt->dst.type = saved_dst_type;
5656 
5657 	if ((ctxt->d & SrcMask) == SrcSI)
5658 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5659 
5660 	if ((ctxt->d & DstMask) == DstDI)
5661 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5662 
5663 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5664 		unsigned int count;
5665 		struct read_cache *r = &ctxt->io_read;
5666 		if ((ctxt->d & SrcMask) == SrcSI)
5667 			count = ctxt->src.count;
5668 		else
5669 			count = ctxt->dst.count;
5670 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5671 
5672 		if (!string_insn_completed(ctxt)) {
5673 			/*
5674 			 * Re-enter guest when pio read ahead buffer is empty
5675 			 * or, if it is not used, after each 1024 iteration.
5676 			 */
5677 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5678 			    (r->end == 0 || r->end != r->pos)) {
5679 				/*
5680 				 * Reset read cache. Usually happens before
5681 				 * decode, but since instruction is restarted
5682 				 * we have to do it here.
5683 				 */
5684 				ctxt->mem_read.end = 0;
5685 				writeback_registers(ctxt);
5686 				return EMULATION_RESTART;
5687 			}
5688 			goto done; /* skip rip writeback */
5689 		}
5690 		ctxt->eflags &= ~X86_EFLAGS_RF;
5691 	}
5692 
5693 	ctxt->eip = ctxt->_eip;
5694 
5695 done:
5696 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5697 		WARN_ON(ctxt->exception.vector > 0x1f);
5698 		ctxt->have_exception = true;
5699 	}
5700 	if (rc == X86EMUL_INTERCEPTED)
5701 		return EMULATION_INTERCEPTED;
5702 
5703 	if (rc == X86EMUL_CONTINUE)
5704 		writeback_registers(ctxt);
5705 
5706 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5707 
5708 twobyte_insn:
5709 	switch (ctxt->b) {
5710 	case 0x09:		/* wbinvd */
5711 		(ctxt->ops->wbinvd)(ctxt);
5712 		break;
5713 	case 0x08:		/* invd */
5714 	case 0x0d:		/* GrpP (prefetch) */
5715 	case 0x18:		/* Grp16 (prefetch/nop) */
5716 	case 0x1f:		/* nop */
5717 		break;
5718 	case 0x20: /* mov cr, reg */
5719 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5720 		break;
5721 	case 0x21: /* mov from dr to reg */
5722 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5723 		break;
5724 	case 0x40 ... 0x4f:	/* cmov */
5725 		if (test_cc(ctxt->b, ctxt->eflags))
5726 			ctxt->dst.val = ctxt->src.val;
5727 		else if (ctxt->op_bytes != 4)
5728 			ctxt->dst.type = OP_NONE; /* no writeback */
5729 		break;
5730 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5731 		if (test_cc(ctxt->b, ctxt->eflags))
5732 			rc = jmp_rel(ctxt, ctxt->src.val);
5733 		break;
5734 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5735 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5736 		break;
5737 	case 0xb6 ... 0xb7:	/* movzx */
5738 		ctxt->dst.bytes = ctxt->op_bytes;
5739 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5740 						       : (u16) ctxt->src.val;
5741 		break;
5742 	case 0xbe ... 0xbf:	/* movsx */
5743 		ctxt->dst.bytes = ctxt->op_bytes;
5744 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5745 							(s16) ctxt->src.val;
5746 		break;
5747 	default:
5748 		goto cannot_emulate;
5749 	}
5750 
5751 threebyte_insn:
5752 
5753 	if (rc != X86EMUL_CONTINUE)
5754 		goto done;
5755 
5756 	goto writeback;
5757 
5758 cannot_emulate:
5759 	return EMULATION_FAILED;
5760 }
5761 
5762 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5763 {
5764 	invalidate_registers(ctxt);
5765 }
5766 
5767 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5768 {
5769 	writeback_registers(ctxt);
5770 }
5771 
5772 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5773 {
5774 	if (ctxt->rep_prefix && (ctxt->d & String))
5775 		return false;
5776 
5777 	if (ctxt->d & TwoMemOp)
5778 		return false;
5779 
5780 	return true;
5781 }
5782