xref: /linux/arch/x86/kvm/emulate.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
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 
29 #include "x86.h"
30 #include "tss.h"
31 
32 /*
33  * Operand types
34  */
35 #define OpNone             0ull
36 #define OpImplicit         1ull  /* No generic decode */
37 #define OpReg              2ull  /* Register */
38 #define OpMem              3ull  /* Memory */
39 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
40 #define OpDI               5ull  /* ES:DI/EDI/RDI */
41 #define OpMem64            6ull  /* Memory, 64-bit */
42 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
43 #define OpDX               8ull  /* DX register */
44 #define OpCL               9ull  /* CL register (for shifts) */
45 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
46 #define OpOne             11ull  /* Implied 1 */
47 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
48 #define OpMem16           13ull  /* Memory operand (16-bit). */
49 #define OpMem32           14ull  /* Memory operand (32-bit). */
50 #define OpImmU            15ull  /* Immediate operand, zero extended */
51 #define OpSI              16ull  /* SI/ESI/RSI */
52 #define OpImmFAddr        17ull  /* Immediate far address */
53 #define OpMemFAddr        18ull  /* Far address in memory */
54 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
55 #define OpES              20ull  /* ES */
56 #define OpCS              21ull  /* CS */
57 #define OpSS              22ull  /* SS */
58 #define OpDS              23ull  /* DS */
59 #define OpFS              24ull  /* FS */
60 #define OpGS              25ull  /* GS */
61 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
62 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
63 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
64 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
65 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
66 
67 #define OpBits             5  /* Width of operand field */
68 #define OpMask             ((1ull << OpBits) - 1)
69 
70 /*
71  * Opcode effective-address decode tables.
72  * Note that we only emulate instructions that have at least one memory
73  * operand (excluding implicit stack references). We assume that stack
74  * references and instruction fetches will never occur in special memory
75  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
76  * not be handled.
77  */
78 
79 /* Operand sizes: 8-bit operands or specified/overridden size. */
80 #define ByteOp      (1<<0)	/* 8-bit operands. */
81 /* Destination operand type. */
82 #define DstShift    1
83 #define ImplicitOps (OpImplicit << DstShift)
84 #define DstReg      (OpReg << DstShift)
85 #define DstMem      (OpMem << DstShift)
86 #define DstAcc      (OpAcc << DstShift)
87 #define DstDI       (OpDI << DstShift)
88 #define DstMem64    (OpMem64 << DstShift)
89 #define DstMem16    (OpMem16 << DstShift)
90 #define DstImmUByte (OpImmUByte << DstShift)
91 #define DstDX       (OpDX << DstShift)
92 #define DstAccLo    (OpAccLo << DstShift)
93 #define DstMask     (OpMask << DstShift)
94 /* Source operand type. */
95 #define SrcShift    6
96 #define SrcNone     (OpNone << SrcShift)
97 #define SrcReg      (OpReg << SrcShift)
98 #define SrcMem      (OpMem << SrcShift)
99 #define SrcMem16    (OpMem16 << SrcShift)
100 #define SrcMem32    (OpMem32 << SrcShift)
101 #define SrcImm      (OpImm << SrcShift)
102 #define SrcImmByte  (OpImmByte << SrcShift)
103 #define SrcOne      (OpOne << SrcShift)
104 #define SrcImmUByte (OpImmUByte << SrcShift)
105 #define SrcImmU     (OpImmU << SrcShift)
106 #define SrcSI       (OpSI << SrcShift)
107 #define SrcXLat     (OpXLat << SrcShift)
108 #define SrcImmFAddr (OpImmFAddr << SrcShift)
109 #define SrcMemFAddr (OpMemFAddr << SrcShift)
110 #define SrcAcc      (OpAcc << SrcShift)
111 #define SrcImmU16   (OpImmU16 << SrcShift)
112 #define SrcImm64    (OpImm64 << SrcShift)
113 #define SrcDX       (OpDX << SrcShift)
114 #define SrcMem8     (OpMem8 << SrcShift)
115 #define SrcAccHi    (OpAccHi << SrcShift)
116 #define SrcMask     (OpMask << SrcShift)
117 #define BitOp       (1<<11)
118 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
119 #define String      (1<<13)     /* String instruction (rep capable) */
120 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
121 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
122 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
123 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
124 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
125 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
126 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
127 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
128 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
129 #define Sse         (1<<18)     /* SSE Vector instruction */
130 /* Generic ModRM decode. */
131 #define ModRM       (1<<19)
132 /* Destination is only written; never read. */
133 #define Mov         (1<<20)
134 /* Misc flags */
135 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
136 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
137 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
138 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
139 #define Undefined   (1<<25) /* No Such Instruction */
140 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
141 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
142 #define No64	    (1<<28)
143 #define PageTable   (1 << 29)   /* instruction used to write page table */
144 #define NotImpl     (1 << 30)   /* instruction is not implemented */
145 /* Source 2 operand type */
146 #define Src2Shift   (31)
147 #define Src2None    (OpNone << Src2Shift)
148 #define Src2Mem     (OpMem << Src2Shift)
149 #define Src2CL      (OpCL << Src2Shift)
150 #define Src2ImmByte (OpImmByte << Src2Shift)
151 #define Src2One     (OpOne << Src2Shift)
152 #define Src2Imm     (OpImm << Src2Shift)
153 #define Src2ES      (OpES << Src2Shift)
154 #define Src2CS      (OpCS << Src2Shift)
155 #define Src2SS      (OpSS << Src2Shift)
156 #define Src2DS      (OpDS << Src2Shift)
157 #define Src2FS      (OpFS << Src2Shift)
158 #define Src2GS      (OpGS << Src2Shift)
159 #define Src2Mask    (OpMask << Src2Shift)
160 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
161 #define AlignMask   ((u64)7 << 41)
162 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
163 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
164 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
165 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
166 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
167 #define NoWrite     ((u64)1 << 45)  /* No writeback */
168 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
169 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
170 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
171 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
172 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
173 #define NearBranch  ((u64)1 << 52)  /* Near branches */
174 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
175 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
176 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
177 
178 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
179 
180 #define X2(x...) x, x
181 #define X3(x...) X2(x), x
182 #define X4(x...) X2(x), X2(x)
183 #define X5(x...) X4(x), x
184 #define X6(x...) X4(x), X2(x)
185 #define X7(x...) X4(x), X3(x)
186 #define X8(x...) X4(x), X4(x)
187 #define X16(x...) X8(x), X8(x)
188 
189 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1)
190 #define FASTOP_SIZE 8
191 
192 /*
193  * fastop functions have a special calling convention:
194  *
195  * dst:    rax        (in/out)
196  * src:    rdx        (in/out)
197  * src2:   rcx        (in)
198  * flags:  rflags     (in/out)
199  * ex:     rsi        (in:fastop pointer, out:zero if exception)
200  *
201  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
202  * different operand sizes can be reached by calculation, rather than a jump
203  * table (which would be bigger than the code).
204  *
205  * fastop functions are declared as taking a never-defined fastop parameter,
206  * so they can't be called from C directly.
207  */
208 
209 struct fastop;
210 
211 struct opcode {
212 	u64 flags : 56;
213 	u64 intercept : 8;
214 	union {
215 		int (*execute)(struct x86_emulate_ctxt *ctxt);
216 		const struct opcode *group;
217 		const struct group_dual *gdual;
218 		const struct gprefix *gprefix;
219 		const struct escape *esc;
220 		const struct instr_dual *idual;
221 		const struct mode_dual *mdual;
222 		void (*fastop)(struct fastop *fake);
223 	} u;
224 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
225 };
226 
227 struct group_dual {
228 	struct opcode mod012[8];
229 	struct opcode mod3[8];
230 };
231 
232 struct gprefix {
233 	struct opcode pfx_no;
234 	struct opcode pfx_66;
235 	struct opcode pfx_f2;
236 	struct opcode pfx_f3;
237 };
238 
239 struct escape {
240 	struct opcode op[8];
241 	struct opcode high[64];
242 };
243 
244 struct instr_dual {
245 	struct opcode mod012;
246 	struct opcode mod3;
247 };
248 
249 struct mode_dual {
250 	struct opcode mode32;
251 	struct opcode mode64;
252 };
253 
254 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
255 
256 enum x86_transfer_type {
257 	X86_TRANSFER_NONE,
258 	X86_TRANSFER_CALL_JMP,
259 	X86_TRANSFER_RET,
260 	X86_TRANSFER_TASK_SWITCH,
261 };
262 
263 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
264 {
265 	if (!(ctxt->regs_valid & (1 << nr))) {
266 		ctxt->regs_valid |= 1 << nr;
267 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
268 	}
269 	return ctxt->_regs[nr];
270 }
271 
272 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
273 {
274 	ctxt->regs_valid |= 1 << nr;
275 	ctxt->regs_dirty |= 1 << nr;
276 	return &ctxt->_regs[nr];
277 }
278 
279 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
280 {
281 	reg_read(ctxt, nr);
282 	return reg_write(ctxt, nr);
283 }
284 
285 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
286 {
287 	unsigned reg;
288 
289 	for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16)
290 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
291 }
292 
293 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
294 {
295 	ctxt->regs_dirty = 0;
296 	ctxt->regs_valid = 0;
297 }
298 
299 /*
300  * These EFLAGS bits are restored from saved value during emulation, and
301  * any changes are written back to the saved value after emulation.
302  */
303 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
304 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
305 
306 #ifdef CONFIG_X86_64
307 #define ON64(x) x
308 #else
309 #define ON64(x)
310 #endif
311 
312 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *));
313 
314 #define FOP_FUNC(name) \
315 	".align " __stringify(FASTOP_SIZE) " \n\t" \
316 	".type " name ", @function \n\t" \
317 	name ":\n\t"
318 
319 #define FOP_RET   "ret \n\t"
320 
321 #define FOP_START(op) \
322 	extern void em_##op(struct fastop *fake); \
323 	asm(".pushsection .text, \"ax\" \n\t" \
324 	    ".global em_" #op " \n\t" \
325 	    FOP_FUNC("em_" #op)
326 
327 #define FOP_END \
328 	    ".popsection")
329 
330 #define FOPNOP() \
331 	FOP_FUNC(__stringify(__UNIQUE_ID(nop))) \
332 	FOP_RET
333 
334 #define FOP1E(op,  dst) \
335 	FOP_FUNC(#op "_" #dst) \
336 	"10: " #op " %" #dst " \n\t" FOP_RET
337 
338 #define FOP1EEX(op,  dst) \
339 	FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception)
340 
341 #define FASTOP1(op) \
342 	FOP_START(op) \
343 	FOP1E(op##b, al) \
344 	FOP1E(op##w, ax) \
345 	FOP1E(op##l, eax) \
346 	ON64(FOP1E(op##q, rax))	\
347 	FOP_END
348 
349 /* 1-operand, using src2 (for MUL/DIV r/m) */
350 #define FASTOP1SRC2(op, name) \
351 	FOP_START(name) \
352 	FOP1E(op, cl) \
353 	FOP1E(op, cx) \
354 	FOP1E(op, ecx) \
355 	ON64(FOP1E(op, rcx)) \
356 	FOP_END
357 
358 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
359 #define FASTOP1SRC2EX(op, name) \
360 	FOP_START(name) \
361 	FOP1EEX(op, cl) \
362 	FOP1EEX(op, cx) \
363 	FOP1EEX(op, ecx) \
364 	ON64(FOP1EEX(op, rcx)) \
365 	FOP_END
366 
367 #define FOP2E(op,  dst, src)	   \
368 	FOP_FUNC(#op "_" #dst "_" #src) \
369 	#op " %" #src ", %" #dst " \n\t" FOP_RET
370 
371 #define FASTOP2(op) \
372 	FOP_START(op) \
373 	FOP2E(op##b, al, dl) \
374 	FOP2E(op##w, ax, dx) \
375 	FOP2E(op##l, eax, edx) \
376 	ON64(FOP2E(op##q, rax, rdx)) \
377 	FOP_END
378 
379 /* 2 operand, word only */
380 #define FASTOP2W(op) \
381 	FOP_START(op) \
382 	FOPNOP() \
383 	FOP2E(op##w, ax, dx) \
384 	FOP2E(op##l, eax, edx) \
385 	ON64(FOP2E(op##q, rax, rdx)) \
386 	FOP_END
387 
388 /* 2 operand, src is CL */
389 #define FASTOP2CL(op) \
390 	FOP_START(op) \
391 	FOP2E(op##b, al, cl) \
392 	FOP2E(op##w, ax, cl) \
393 	FOP2E(op##l, eax, cl) \
394 	ON64(FOP2E(op##q, rax, cl)) \
395 	FOP_END
396 
397 /* 2 operand, src and dest are reversed */
398 #define FASTOP2R(op, name) \
399 	FOP_START(name) \
400 	FOP2E(op##b, dl, al) \
401 	FOP2E(op##w, dx, ax) \
402 	FOP2E(op##l, edx, eax) \
403 	ON64(FOP2E(op##q, rdx, rax)) \
404 	FOP_END
405 
406 #define FOP3E(op,  dst, src, src2) \
407 	FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
408 	#op " %" #src2 ", %" #src ", %" #dst " \n\t" FOP_RET
409 
410 /* 3-operand, word-only, src2=cl */
411 #define FASTOP3WCL(op) \
412 	FOP_START(op) \
413 	FOPNOP() \
414 	FOP3E(op##w, ax, dx, cl) \
415 	FOP3E(op##l, eax, edx, cl) \
416 	ON64(FOP3E(op##q, rax, rdx, cl)) \
417 	FOP_END
418 
419 /* Special case for SETcc - 1 instruction per cc */
420 #define FOP_SETCC(op) \
421 	".align 4 \n\t" \
422 	".type " #op ", @function \n\t" \
423 	#op ": \n\t" \
424 	#op " %al \n\t" \
425 	FOP_RET
426 
427 asm(".global kvm_fastop_exception \n"
428     "kvm_fastop_exception: xor %esi, %esi; ret");
429 
430 FOP_START(setcc)
431 FOP_SETCC(seto)
432 FOP_SETCC(setno)
433 FOP_SETCC(setc)
434 FOP_SETCC(setnc)
435 FOP_SETCC(setz)
436 FOP_SETCC(setnz)
437 FOP_SETCC(setbe)
438 FOP_SETCC(setnbe)
439 FOP_SETCC(sets)
440 FOP_SETCC(setns)
441 FOP_SETCC(setp)
442 FOP_SETCC(setnp)
443 FOP_SETCC(setl)
444 FOP_SETCC(setnl)
445 FOP_SETCC(setle)
446 FOP_SETCC(setnle)
447 FOP_END;
448 
449 FOP_START(salc) "pushf; sbb %al, %al; popf \n\t" FOP_RET
450 FOP_END;
451 
452 /*
453  * XXX: inoutclob user must know where the argument is being expanded.
454  *      Relying on CC_HAVE_ASM_GOTO would allow us to remove _fault.
455  */
456 #define asm_safe(insn, inoutclob...) \
457 ({ \
458 	int _fault = 0; \
459  \
460 	asm volatile("1:" insn "\n" \
461 	             "2:\n" \
462 	             ".pushsection .fixup, \"ax\"\n" \
463 	             "3: movl $1, %[_fault]\n" \
464 	             "   jmp  2b\n" \
465 	             ".popsection\n" \
466 	             _ASM_EXTABLE(1b, 3b) \
467 	             : [_fault] "+qm"(_fault) inoutclob ); \
468  \
469 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
470 })
471 
472 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
473 				    enum x86_intercept intercept,
474 				    enum x86_intercept_stage stage)
475 {
476 	struct x86_instruction_info info = {
477 		.intercept  = intercept,
478 		.rep_prefix = ctxt->rep_prefix,
479 		.modrm_mod  = ctxt->modrm_mod,
480 		.modrm_reg  = ctxt->modrm_reg,
481 		.modrm_rm   = ctxt->modrm_rm,
482 		.src_val    = ctxt->src.val64,
483 		.dst_val    = ctxt->dst.val64,
484 		.src_bytes  = ctxt->src.bytes,
485 		.dst_bytes  = ctxt->dst.bytes,
486 		.ad_bytes   = ctxt->ad_bytes,
487 		.next_rip   = ctxt->eip,
488 	};
489 
490 	return ctxt->ops->intercept(ctxt, &info, stage);
491 }
492 
493 static void assign_masked(ulong *dest, ulong src, ulong mask)
494 {
495 	*dest = (*dest & ~mask) | (src & mask);
496 }
497 
498 static void assign_register(unsigned long *reg, u64 val, int bytes)
499 {
500 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
501 	switch (bytes) {
502 	case 1:
503 		*(u8 *)reg = (u8)val;
504 		break;
505 	case 2:
506 		*(u16 *)reg = (u16)val;
507 		break;
508 	case 4:
509 		*reg = (u32)val;
510 		break;	/* 64b: zero-extend */
511 	case 8:
512 		*reg = val;
513 		break;
514 	}
515 }
516 
517 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
518 {
519 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
520 }
521 
522 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
523 {
524 	u16 sel;
525 	struct desc_struct ss;
526 
527 	if (ctxt->mode == X86EMUL_MODE_PROT64)
528 		return ~0UL;
529 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
530 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
531 }
532 
533 static int stack_size(struct x86_emulate_ctxt *ctxt)
534 {
535 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
536 }
537 
538 /* Access/update address held in a register, based on addressing mode. */
539 static inline unsigned long
540 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
541 {
542 	if (ctxt->ad_bytes == sizeof(unsigned long))
543 		return reg;
544 	else
545 		return reg & ad_mask(ctxt);
546 }
547 
548 static inline unsigned long
549 register_address(struct x86_emulate_ctxt *ctxt, int reg)
550 {
551 	return address_mask(ctxt, reg_read(ctxt, reg));
552 }
553 
554 static void masked_increment(ulong *reg, ulong mask, int inc)
555 {
556 	assign_masked(reg, *reg + inc, mask);
557 }
558 
559 static inline void
560 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
561 {
562 	ulong *preg = reg_rmw(ctxt, reg);
563 
564 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
565 }
566 
567 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
568 {
569 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
570 }
571 
572 static u32 desc_limit_scaled(struct desc_struct *desc)
573 {
574 	u32 limit = get_desc_limit(desc);
575 
576 	return desc->g ? (limit << 12) | 0xfff : limit;
577 }
578 
579 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
580 {
581 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
582 		return 0;
583 
584 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
585 }
586 
587 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
588 			     u32 error, bool valid)
589 {
590 	WARN_ON(vec > 0x1f);
591 	ctxt->exception.vector = vec;
592 	ctxt->exception.error_code = error;
593 	ctxt->exception.error_code_valid = valid;
594 	return X86EMUL_PROPAGATE_FAULT;
595 }
596 
597 static int emulate_db(struct x86_emulate_ctxt *ctxt)
598 {
599 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
600 }
601 
602 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
603 {
604 	return emulate_exception(ctxt, GP_VECTOR, err, true);
605 }
606 
607 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
608 {
609 	return emulate_exception(ctxt, SS_VECTOR, err, true);
610 }
611 
612 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
613 {
614 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
615 }
616 
617 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
618 {
619 	return emulate_exception(ctxt, TS_VECTOR, err, true);
620 }
621 
622 static int emulate_de(struct x86_emulate_ctxt *ctxt)
623 {
624 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
625 }
626 
627 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
628 {
629 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
630 }
631 
632 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
633 {
634 	u16 selector;
635 	struct desc_struct desc;
636 
637 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
638 	return selector;
639 }
640 
641 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
642 				 unsigned seg)
643 {
644 	u16 dummy;
645 	u32 base3;
646 	struct desc_struct desc;
647 
648 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
649 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
650 }
651 
652 /*
653  * x86 defines three classes of vector instructions: explicitly
654  * aligned, explicitly unaligned, and the rest, which change behaviour
655  * depending on whether they're AVX encoded or not.
656  *
657  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
658  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
659  * 512 bytes of data must be aligned to a 16 byte boundary.
660  */
661 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
662 {
663 	u64 alignment = ctxt->d & AlignMask;
664 
665 	if (likely(size < 16))
666 		return 1;
667 
668 	switch (alignment) {
669 	case Unaligned:
670 	case Avx:
671 		return 1;
672 	case Aligned16:
673 		return 16;
674 	case Aligned:
675 	default:
676 		return size;
677 	}
678 }
679 
680 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
681 				       struct segmented_address addr,
682 				       unsigned *max_size, unsigned size,
683 				       bool write, bool fetch,
684 				       enum x86emul_mode mode, ulong *linear)
685 {
686 	struct desc_struct desc;
687 	bool usable;
688 	ulong la;
689 	u32 lim;
690 	u16 sel;
691 
692 	la = seg_base(ctxt, addr.seg) + addr.ea;
693 	*max_size = 0;
694 	switch (mode) {
695 	case X86EMUL_MODE_PROT64:
696 		*linear = la;
697 		if (is_noncanonical_address(la))
698 			goto bad;
699 
700 		*max_size = min_t(u64, ~0u, (1ull << 48) - la);
701 		if (size > *max_size)
702 			goto bad;
703 		break;
704 	default:
705 		*linear = la = (u32)la;
706 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
707 						addr.seg);
708 		if (!usable)
709 			goto bad;
710 		/* code segment in protected mode or read-only data segment */
711 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
712 					|| !(desc.type & 2)) && write)
713 			goto bad;
714 		/* unreadable code segment */
715 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
716 			goto bad;
717 		lim = desc_limit_scaled(&desc);
718 		if (!(desc.type & 8) && (desc.type & 4)) {
719 			/* expand-down segment */
720 			if (addr.ea <= lim)
721 				goto bad;
722 			lim = desc.d ? 0xffffffff : 0xffff;
723 		}
724 		if (addr.ea > lim)
725 			goto bad;
726 		if (lim == 0xffffffff)
727 			*max_size = ~0u;
728 		else {
729 			*max_size = (u64)lim + 1 - addr.ea;
730 			if (size > *max_size)
731 				goto bad;
732 		}
733 		break;
734 	}
735 	if (la & (insn_alignment(ctxt, size) - 1))
736 		return emulate_gp(ctxt, 0);
737 	return X86EMUL_CONTINUE;
738 bad:
739 	if (addr.seg == VCPU_SREG_SS)
740 		return emulate_ss(ctxt, 0);
741 	else
742 		return emulate_gp(ctxt, 0);
743 }
744 
745 static int linearize(struct x86_emulate_ctxt *ctxt,
746 		     struct segmented_address addr,
747 		     unsigned size, bool write,
748 		     ulong *linear)
749 {
750 	unsigned max_size;
751 	return __linearize(ctxt, addr, &max_size, size, write, false,
752 			   ctxt->mode, linear);
753 }
754 
755 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
756 			     enum x86emul_mode mode)
757 {
758 	ulong linear;
759 	int rc;
760 	unsigned max_size;
761 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
762 					   .ea = dst };
763 
764 	if (ctxt->op_bytes != sizeof(unsigned long))
765 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
766 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
767 	if (rc == X86EMUL_CONTINUE)
768 		ctxt->_eip = addr.ea;
769 	return rc;
770 }
771 
772 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
773 {
774 	return assign_eip(ctxt, dst, ctxt->mode);
775 }
776 
777 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
778 			  const struct desc_struct *cs_desc)
779 {
780 	enum x86emul_mode mode = ctxt->mode;
781 	int rc;
782 
783 #ifdef CONFIG_X86_64
784 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
785 		if (cs_desc->l) {
786 			u64 efer = 0;
787 
788 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
789 			if (efer & EFER_LMA)
790 				mode = X86EMUL_MODE_PROT64;
791 		} else
792 			mode = X86EMUL_MODE_PROT32; /* temporary value */
793 	}
794 #endif
795 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
796 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
797 	rc = assign_eip(ctxt, dst, mode);
798 	if (rc == X86EMUL_CONTINUE)
799 		ctxt->mode = mode;
800 	return rc;
801 }
802 
803 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
804 {
805 	return assign_eip_near(ctxt, ctxt->_eip + rel);
806 }
807 
808 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
809 			      struct segmented_address addr,
810 			      void *data,
811 			      unsigned size)
812 {
813 	int rc;
814 	ulong linear;
815 
816 	rc = linearize(ctxt, addr, size, false, &linear);
817 	if (rc != X86EMUL_CONTINUE)
818 		return rc;
819 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception);
820 }
821 
822 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
823 			       struct segmented_address addr,
824 			       void *data,
825 			       unsigned int size)
826 {
827 	int rc;
828 	ulong linear;
829 
830 	rc = linearize(ctxt, addr, size, true, &linear);
831 	if (rc != X86EMUL_CONTINUE)
832 		return rc;
833 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception);
834 }
835 
836 /*
837  * Prefetch the remaining bytes of the instruction without crossing page
838  * boundary if they are not in fetch_cache yet.
839  */
840 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
841 {
842 	int rc;
843 	unsigned size, max_size;
844 	unsigned long linear;
845 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
846 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
847 					   .ea = ctxt->eip + cur_size };
848 
849 	/*
850 	 * We do not know exactly how many bytes will be needed, and
851 	 * __linearize is expensive, so fetch as much as possible.  We
852 	 * just have to avoid going beyond the 15 byte limit, the end
853 	 * of the segment, or the end of the page.
854 	 *
855 	 * __linearize is called with size 0 so that it does not do any
856 	 * boundary check itself.  Instead, we use max_size to check
857 	 * against op_size.
858 	 */
859 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
860 			 &linear);
861 	if (unlikely(rc != X86EMUL_CONTINUE))
862 		return rc;
863 
864 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
865 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
866 
867 	/*
868 	 * One instruction can only straddle two pages,
869 	 * and one has been loaded at the beginning of
870 	 * x86_decode_insn.  So, if not enough bytes
871 	 * still, we must have hit the 15-byte boundary.
872 	 */
873 	if (unlikely(size < op_size))
874 		return emulate_gp(ctxt, 0);
875 
876 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
877 			      size, &ctxt->exception);
878 	if (unlikely(rc != X86EMUL_CONTINUE))
879 		return rc;
880 	ctxt->fetch.end += size;
881 	return X86EMUL_CONTINUE;
882 }
883 
884 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
885 					       unsigned size)
886 {
887 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
888 
889 	if (unlikely(done_size < size))
890 		return __do_insn_fetch_bytes(ctxt, size - done_size);
891 	else
892 		return X86EMUL_CONTINUE;
893 }
894 
895 /* Fetch next part of the instruction being emulated. */
896 #define insn_fetch(_type, _ctxt)					\
897 ({	_type _x;							\
898 									\
899 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
900 	if (rc != X86EMUL_CONTINUE)					\
901 		goto done;						\
902 	ctxt->_eip += sizeof(_type);					\
903 	_x = *(_type __aligned(1) *) ctxt->fetch.ptr;			\
904 	ctxt->fetch.ptr += sizeof(_type);				\
905 	_x;								\
906 })
907 
908 #define insn_fetch_arr(_arr, _size, _ctxt)				\
909 ({									\
910 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
911 	if (rc != X86EMUL_CONTINUE)					\
912 		goto done;						\
913 	ctxt->_eip += (_size);						\
914 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
915 	ctxt->fetch.ptr += (_size);					\
916 })
917 
918 /*
919  * Given the 'reg' portion of a ModRM byte, and a register block, return a
920  * pointer into the block that addresses the relevant register.
921  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
922  */
923 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
924 			     int byteop)
925 {
926 	void *p;
927 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
928 
929 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
930 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
931 	else
932 		p = reg_rmw(ctxt, modrm_reg);
933 	return p;
934 }
935 
936 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
937 			   struct segmented_address addr,
938 			   u16 *size, unsigned long *address, int op_bytes)
939 {
940 	int rc;
941 
942 	if (op_bytes == 2)
943 		op_bytes = 3;
944 	*address = 0;
945 	rc = segmented_read_std(ctxt, addr, size, 2);
946 	if (rc != X86EMUL_CONTINUE)
947 		return rc;
948 	addr.ea += 2;
949 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
950 	return rc;
951 }
952 
953 FASTOP2(add);
954 FASTOP2(or);
955 FASTOP2(adc);
956 FASTOP2(sbb);
957 FASTOP2(and);
958 FASTOP2(sub);
959 FASTOP2(xor);
960 FASTOP2(cmp);
961 FASTOP2(test);
962 
963 FASTOP1SRC2(mul, mul_ex);
964 FASTOP1SRC2(imul, imul_ex);
965 FASTOP1SRC2EX(div, div_ex);
966 FASTOP1SRC2EX(idiv, idiv_ex);
967 
968 FASTOP3WCL(shld);
969 FASTOP3WCL(shrd);
970 
971 FASTOP2W(imul);
972 
973 FASTOP1(not);
974 FASTOP1(neg);
975 FASTOP1(inc);
976 FASTOP1(dec);
977 
978 FASTOP2CL(rol);
979 FASTOP2CL(ror);
980 FASTOP2CL(rcl);
981 FASTOP2CL(rcr);
982 FASTOP2CL(shl);
983 FASTOP2CL(shr);
984 FASTOP2CL(sar);
985 
986 FASTOP2W(bsf);
987 FASTOP2W(bsr);
988 FASTOP2W(bt);
989 FASTOP2W(bts);
990 FASTOP2W(btr);
991 FASTOP2W(btc);
992 
993 FASTOP2(xadd);
994 
995 FASTOP2R(cmp, cmp_r);
996 
997 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
998 {
999 	/* If src is zero, do not writeback, but update flags */
1000 	if (ctxt->src.val == 0)
1001 		ctxt->dst.type = OP_NONE;
1002 	return fastop(ctxt, em_bsf);
1003 }
1004 
1005 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1006 {
1007 	/* If src is zero, do not writeback, but update flags */
1008 	if (ctxt->src.val == 0)
1009 		ctxt->dst.type = OP_NONE;
1010 	return fastop(ctxt, em_bsr);
1011 }
1012 
1013 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1014 {
1015 	u8 rc;
1016 	void (*fop)(void) = (void *)em_setcc + 4 * (condition & 0xf);
1017 
1018 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1019 	asm("push %[flags]; popf; call *%[fastop]"
1020 	    : "=a"(rc) : [fastop]"r"(fop), [flags]"r"(flags));
1021 	return rc;
1022 }
1023 
1024 static void fetch_register_operand(struct operand *op)
1025 {
1026 	switch (op->bytes) {
1027 	case 1:
1028 		op->val = *(u8 *)op->addr.reg;
1029 		break;
1030 	case 2:
1031 		op->val = *(u16 *)op->addr.reg;
1032 		break;
1033 	case 4:
1034 		op->val = *(u32 *)op->addr.reg;
1035 		break;
1036 	case 8:
1037 		op->val = *(u64 *)op->addr.reg;
1038 		break;
1039 	}
1040 }
1041 
1042 static void read_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data, int reg)
1043 {
1044 	ctxt->ops->get_fpu(ctxt);
1045 	switch (reg) {
1046 	case 0: asm("movdqa %%xmm0, %0" : "=m"(*data)); break;
1047 	case 1: asm("movdqa %%xmm1, %0" : "=m"(*data)); break;
1048 	case 2: asm("movdqa %%xmm2, %0" : "=m"(*data)); break;
1049 	case 3: asm("movdqa %%xmm3, %0" : "=m"(*data)); break;
1050 	case 4: asm("movdqa %%xmm4, %0" : "=m"(*data)); break;
1051 	case 5: asm("movdqa %%xmm5, %0" : "=m"(*data)); break;
1052 	case 6: asm("movdqa %%xmm6, %0" : "=m"(*data)); break;
1053 	case 7: asm("movdqa %%xmm7, %0" : "=m"(*data)); break;
1054 #ifdef CONFIG_X86_64
1055 	case 8: asm("movdqa %%xmm8, %0" : "=m"(*data)); break;
1056 	case 9: asm("movdqa %%xmm9, %0" : "=m"(*data)); break;
1057 	case 10: asm("movdqa %%xmm10, %0" : "=m"(*data)); break;
1058 	case 11: asm("movdqa %%xmm11, %0" : "=m"(*data)); break;
1059 	case 12: asm("movdqa %%xmm12, %0" : "=m"(*data)); break;
1060 	case 13: asm("movdqa %%xmm13, %0" : "=m"(*data)); break;
1061 	case 14: asm("movdqa %%xmm14, %0" : "=m"(*data)); break;
1062 	case 15: asm("movdqa %%xmm15, %0" : "=m"(*data)); break;
1063 #endif
1064 	default: BUG();
1065 	}
1066 	ctxt->ops->put_fpu(ctxt);
1067 }
1068 
1069 static void write_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data,
1070 			  int reg)
1071 {
1072 	ctxt->ops->get_fpu(ctxt);
1073 	switch (reg) {
1074 	case 0: asm("movdqa %0, %%xmm0" : : "m"(*data)); break;
1075 	case 1: asm("movdqa %0, %%xmm1" : : "m"(*data)); break;
1076 	case 2: asm("movdqa %0, %%xmm2" : : "m"(*data)); break;
1077 	case 3: asm("movdqa %0, %%xmm3" : : "m"(*data)); break;
1078 	case 4: asm("movdqa %0, %%xmm4" : : "m"(*data)); break;
1079 	case 5: asm("movdqa %0, %%xmm5" : : "m"(*data)); break;
1080 	case 6: asm("movdqa %0, %%xmm6" : : "m"(*data)); break;
1081 	case 7: asm("movdqa %0, %%xmm7" : : "m"(*data)); break;
1082 #ifdef CONFIG_X86_64
1083 	case 8: asm("movdqa %0, %%xmm8" : : "m"(*data)); break;
1084 	case 9: asm("movdqa %0, %%xmm9" : : "m"(*data)); break;
1085 	case 10: asm("movdqa %0, %%xmm10" : : "m"(*data)); break;
1086 	case 11: asm("movdqa %0, %%xmm11" : : "m"(*data)); break;
1087 	case 12: asm("movdqa %0, %%xmm12" : : "m"(*data)); break;
1088 	case 13: asm("movdqa %0, %%xmm13" : : "m"(*data)); break;
1089 	case 14: asm("movdqa %0, %%xmm14" : : "m"(*data)); break;
1090 	case 15: asm("movdqa %0, %%xmm15" : : "m"(*data)); break;
1091 #endif
1092 	default: BUG();
1093 	}
1094 	ctxt->ops->put_fpu(ctxt);
1095 }
1096 
1097 static void read_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1098 {
1099 	ctxt->ops->get_fpu(ctxt);
1100 	switch (reg) {
1101 	case 0: asm("movq %%mm0, %0" : "=m"(*data)); break;
1102 	case 1: asm("movq %%mm1, %0" : "=m"(*data)); break;
1103 	case 2: asm("movq %%mm2, %0" : "=m"(*data)); break;
1104 	case 3: asm("movq %%mm3, %0" : "=m"(*data)); break;
1105 	case 4: asm("movq %%mm4, %0" : "=m"(*data)); break;
1106 	case 5: asm("movq %%mm5, %0" : "=m"(*data)); break;
1107 	case 6: asm("movq %%mm6, %0" : "=m"(*data)); break;
1108 	case 7: asm("movq %%mm7, %0" : "=m"(*data)); break;
1109 	default: BUG();
1110 	}
1111 	ctxt->ops->put_fpu(ctxt);
1112 }
1113 
1114 static void write_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1115 {
1116 	ctxt->ops->get_fpu(ctxt);
1117 	switch (reg) {
1118 	case 0: asm("movq %0, %%mm0" : : "m"(*data)); break;
1119 	case 1: asm("movq %0, %%mm1" : : "m"(*data)); break;
1120 	case 2: asm("movq %0, %%mm2" : : "m"(*data)); break;
1121 	case 3: asm("movq %0, %%mm3" : : "m"(*data)); break;
1122 	case 4: asm("movq %0, %%mm4" : : "m"(*data)); break;
1123 	case 5: asm("movq %0, %%mm5" : : "m"(*data)); break;
1124 	case 6: asm("movq %0, %%mm6" : : "m"(*data)); break;
1125 	case 7: asm("movq %0, %%mm7" : : "m"(*data)); break;
1126 	default: BUG();
1127 	}
1128 	ctxt->ops->put_fpu(ctxt);
1129 }
1130 
1131 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1132 {
1133 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1134 		return emulate_nm(ctxt);
1135 
1136 	ctxt->ops->get_fpu(ctxt);
1137 	asm volatile("fninit");
1138 	ctxt->ops->put_fpu(ctxt);
1139 	return X86EMUL_CONTINUE;
1140 }
1141 
1142 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1143 {
1144 	u16 fcw;
1145 
1146 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1147 		return emulate_nm(ctxt);
1148 
1149 	ctxt->ops->get_fpu(ctxt);
1150 	asm volatile("fnstcw %0": "+m"(fcw));
1151 	ctxt->ops->put_fpu(ctxt);
1152 
1153 	ctxt->dst.val = fcw;
1154 
1155 	return X86EMUL_CONTINUE;
1156 }
1157 
1158 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1159 {
1160 	u16 fsw;
1161 
1162 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1163 		return emulate_nm(ctxt);
1164 
1165 	ctxt->ops->get_fpu(ctxt);
1166 	asm volatile("fnstsw %0": "+m"(fsw));
1167 	ctxt->ops->put_fpu(ctxt);
1168 
1169 	ctxt->dst.val = fsw;
1170 
1171 	return X86EMUL_CONTINUE;
1172 }
1173 
1174 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1175 				    struct operand *op)
1176 {
1177 	unsigned reg = ctxt->modrm_reg;
1178 
1179 	if (!(ctxt->d & ModRM))
1180 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1181 
1182 	if (ctxt->d & Sse) {
1183 		op->type = OP_XMM;
1184 		op->bytes = 16;
1185 		op->addr.xmm = reg;
1186 		read_sse_reg(ctxt, &op->vec_val, reg);
1187 		return;
1188 	}
1189 	if (ctxt->d & Mmx) {
1190 		reg &= 7;
1191 		op->type = OP_MM;
1192 		op->bytes = 8;
1193 		op->addr.mm = reg;
1194 		return;
1195 	}
1196 
1197 	op->type = OP_REG;
1198 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1199 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1200 
1201 	fetch_register_operand(op);
1202 	op->orig_val = op->val;
1203 }
1204 
1205 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1206 {
1207 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1208 		ctxt->modrm_seg = VCPU_SREG_SS;
1209 }
1210 
1211 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1212 			struct operand *op)
1213 {
1214 	u8 sib;
1215 	int index_reg, base_reg, scale;
1216 	int rc = X86EMUL_CONTINUE;
1217 	ulong modrm_ea = 0;
1218 
1219 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1220 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1221 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1222 
1223 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1224 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1225 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1226 	ctxt->modrm_seg = VCPU_SREG_DS;
1227 
1228 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1229 		op->type = OP_REG;
1230 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1231 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1232 				ctxt->d & ByteOp);
1233 		if (ctxt->d & Sse) {
1234 			op->type = OP_XMM;
1235 			op->bytes = 16;
1236 			op->addr.xmm = ctxt->modrm_rm;
1237 			read_sse_reg(ctxt, &op->vec_val, ctxt->modrm_rm);
1238 			return rc;
1239 		}
1240 		if (ctxt->d & Mmx) {
1241 			op->type = OP_MM;
1242 			op->bytes = 8;
1243 			op->addr.mm = ctxt->modrm_rm & 7;
1244 			return rc;
1245 		}
1246 		fetch_register_operand(op);
1247 		return rc;
1248 	}
1249 
1250 	op->type = OP_MEM;
1251 
1252 	if (ctxt->ad_bytes == 2) {
1253 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1254 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1255 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1256 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1257 
1258 		/* 16-bit ModR/M decode. */
1259 		switch (ctxt->modrm_mod) {
1260 		case 0:
1261 			if (ctxt->modrm_rm == 6)
1262 				modrm_ea += insn_fetch(u16, ctxt);
1263 			break;
1264 		case 1:
1265 			modrm_ea += insn_fetch(s8, ctxt);
1266 			break;
1267 		case 2:
1268 			modrm_ea += insn_fetch(u16, ctxt);
1269 			break;
1270 		}
1271 		switch (ctxt->modrm_rm) {
1272 		case 0:
1273 			modrm_ea += bx + si;
1274 			break;
1275 		case 1:
1276 			modrm_ea += bx + di;
1277 			break;
1278 		case 2:
1279 			modrm_ea += bp + si;
1280 			break;
1281 		case 3:
1282 			modrm_ea += bp + di;
1283 			break;
1284 		case 4:
1285 			modrm_ea += si;
1286 			break;
1287 		case 5:
1288 			modrm_ea += di;
1289 			break;
1290 		case 6:
1291 			if (ctxt->modrm_mod != 0)
1292 				modrm_ea += bp;
1293 			break;
1294 		case 7:
1295 			modrm_ea += bx;
1296 			break;
1297 		}
1298 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1299 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1300 			ctxt->modrm_seg = VCPU_SREG_SS;
1301 		modrm_ea = (u16)modrm_ea;
1302 	} else {
1303 		/* 32/64-bit ModR/M decode. */
1304 		if ((ctxt->modrm_rm & 7) == 4) {
1305 			sib = insn_fetch(u8, ctxt);
1306 			index_reg |= (sib >> 3) & 7;
1307 			base_reg |= sib & 7;
1308 			scale = sib >> 6;
1309 
1310 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1311 				modrm_ea += insn_fetch(s32, ctxt);
1312 			else {
1313 				modrm_ea += reg_read(ctxt, base_reg);
1314 				adjust_modrm_seg(ctxt, base_reg);
1315 				/* Increment ESP on POP [ESP] */
1316 				if ((ctxt->d & IncSP) &&
1317 				    base_reg == VCPU_REGS_RSP)
1318 					modrm_ea += ctxt->op_bytes;
1319 			}
1320 			if (index_reg != 4)
1321 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1322 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1323 			modrm_ea += insn_fetch(s32, ctxt);
1324 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1325 				ctxt->rip_relative = 1;
1326 		} else {
1327 			base_reg = ctxt->modrm_rm;
1328 			modrm_ea += reg_read(ctxt, base_reg);
1329 			adjust_modrm_seg(ctxt, base_reg);
1330 		}
1331 		switch (ctxt->modrm_mod) {
1332 		case 1:
1333 			modrm_ea += insn_fetch(s8, ctxt);
1334 			break;
1335 		case 2:
1336 			modrm_ea += insn_fetch(s32, ctxt);
1337 			break;
1338 		}
1339 	}
1340 	op->addr.mem.ea = modrm_ea;
1341 	if (ctxt->ad_bytes != 8)
1342 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1343 
1344 done:
1345 	return rc;
1346 }
1347 
1348 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1349 		      struct operand *op)
1350 {
1351 	int rc = X86EMUL_CONTINUE;
1352 
1353 	op->type = OP_MEM;
1354 	switch (ctxt->ad_bytes) {
1355 	case 2:
1356 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1357 		break;
1358 	case 4:
1359 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1360 		break;
1361 	case 8:
1362 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1363 		break;
1364 	}
1365 done:
1366 	return rc;
1367 }
1368 
1369 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1370 {
1371 	long sv = 0, mask;
1372 
1373 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1374 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1375 
1376 		if (ctxt->src.bytes == 2)
1377 			sv = (s16)ctxt->src.val & (s16)mask;
1378 		else if (ctxt->src.bytes == 4)
1379 			sv = (s32)ctxt->src.val & (s32)mask;
1380 		else
1381 			sv = (s64)ctxt->src.val & (s64)mask;
1382 
1383 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1384 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1385 	}
1386 
1387 	/* only subword offset */
1388 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1389 }
1390 
1391 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1392 			 unsigned long addr, void *dest, unsigned size)
1393 {
1394 	int rc;
1395 	struct read_cache *mc = &ctxt->mem_read;
1396 
1397 	if (mc->pos < mc->end)
1398 		goto read_cached;
1399 
1400 	WARN_ON((mc->end + size) >= sizeof(mc->data));
1401 
1402 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1403 				      &ctxt->exception);
1404 	if (rc != X86EMUL_CONTINUE)
1405 		return rc;
1406 
1407 	mc->end += size;
1408 
1409 read_cached:
1410 	memcpy(dest, mc->data + mc->pos, size);
1411 	mc->pos += size;
1412 	return X86EMUL_CONTINUE;
1413 }
1414 
1415 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1416 			  struct segmented_address addr,
1417 			  void *data,
1418 			  unsigned size)
1419 {
1420 	int rc;
1421 	ulong linear;
1422 
1423 	rc = linearize(ctxt, addr, size, false, &linear);
1424 	if (rc != X86EMUL_CONTINUE)
1425 		return rc;
1426 	return read_emulated(ctxt, linear, data, size);
1427 }
1428 
1429 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1430 			   struct segmented_address addr,
1431 			   const void *data,
1432 			   unsigned size)
1433 {
1434 	int rc;
1435 	ulong linear;
1436 
1437 	rc = linearize(ctxt, addr, size, true, &linear);
1438 	if (rc != X86EMUL_CONTINUE)
1439 		return rc;
1440 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1441 					 &ctxt->exception);
1442 }
1443 
1444 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1445 			     struct segmented_address addr,
1446 			     const void *orig_data, const void *data,
1447 			     unsigned size)
1448 {
1449 	int rc;
1450 	ulong linear;
1451 
1452 	rc = linearize(ctxt, addr, size, true, &linear);
1453 	if (rc != X86EMUL_CONTINUE)
1454 		return rc;
1455 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1456 					   size, &ctxt->exception);
1457 }
1458 
1459 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1460 			   unsigned int size, unsigned short port,
1461 			   void *dest)
1462 {
1463 	struct read_cache *rc = &ctxt->io_read;
1464 
1465 	if (rc->pos == rc->end) { /* refill pio read ahead */
1466 		unsigned int in_page, n;
1467 		unsigned int count = ctxt->rep_prefix ?
1468 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1469 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1470 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1471 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1472 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1473 		if (n == 0)
1474 			n = 1;
1475 		rc->pos = rc->end = 0;
1476 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1477 			return 0;
1478 		rc->end = n * size;
1479 	}
1480 
1481 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1482 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1483 		ctxt->dst.data = rc->data + rc->pos;
1484 		ctxt->dst.type = OP_MEM_STR;
1485 		ctxt->dst.count = (rc->end - rc->pos) / size;
1486 		rc->pos = rc->end;
1487 	} else {
1488 		memcpy(dest, rc->data + rc->pos, size);
1489 		rc->pos += size;
1490 	}
1491 	return 1;
1492 }
1493 
1494 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1495 				     u16 index, struct desc_struct *desc)
1496 {
1497 	struct desc_ptr dt;
1498 	ulong addr;
1499 
1500 	ctxt->ops->get_idt(ctxt, &dt);
1501 
1502 	if (dt.size < index * 8 + 7)
1503 		return emulate_gp(ctxt, index << 3 | 0x2);
1504 
1505 	addr = dt.address + index * 8;
1506 	return ctxt->ops->read_std(ctxt, addr, desc, sizeof *desc,
1507 				   &ctxt->exception);
1508 }
1509 
1510 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1511 				     u16 selector, struct desc_ptr *dt)
1512 {
1513 	const struct x86_emulate_ops *ops = ctxt->ops;
1514 	u32 base3 = 0;
1515 
1516 	if (selector & 1 << 2) {
1517 		struct desc_struct desc;
1518 		u16 sel;
1519 
1520 		memset (dt, 0, sizeof *dt);
1521 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1522 				      VCPU_SREG_LDTR))
1523 			return;
1524 
1525 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1526 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1527 	} else
1528 		ops->get_gdt(ctxt, dt);
1529 }
1530 
1531 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1532 			      u16 selector, ulong *desc_addr_p)
1533 {
1534 	struct desc_ptr dt;
1535 	u16 index = selector >> 3;
1536 	ulong addr;
1537 
1538 	get_descriptor_table_ptr(ctxt, selector, &dt);
1539 
1540 	if (dt.size < index * 8 + 7)
1541 		return emulate_gp(ctxt, selector & 0xfffc);
1542 
1543 	addr = dt.address + index * 8;
1544 
1545 #ifdef CONFIG_X86_64
1546 	if (addr >> 32 != 0) {
1547 		u64 efer = 0;
1548 
1549 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1550 		if (!(efer & EFER_LMA))
1551 			addr &= (u32)-1;
1552 	}
1553 #endif
1554 
1555 	*desc_addr_p = addr;
1556 	return X86EMUL_CONTINUE;
1557 }
1558 
1559 /* allowed just for 8 bytes segments */
1560 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1561 				   u16 selector, struct desc_struct *desc,
1562 				   ulong *desc_addr_p)
1563 {
1564 	int rc;
1565 
1566 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1567 	if (rc != X86EMUL_CONTINUE)
1568 		return rc;
1569 
1570 	return ctxt->ops->read_std(ctxt, *desc_addr_p, desc, sizeof(*desc),
1571 				   &ctxt->exception);
1572 }
1573 
1574 /* allowed just for 8 bytes segments */
1575 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1576 				    u16 selector, struct desc_struct *desc)
1577 {
1578 	int rc;
1579 	ulong addr;
1580 
1581 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1582 	if (rc != X86EMUL_CONTINUE)
1583 		return rc;
1584 
1585 	return ctxt->ops->write_std(ctxt, addr, desc, sizeof *desc,
1586 				    &ctxt->exception);
1587 }
1588 
1589 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1590 				     u16 selector, int seg, u8 cpl,
1591 				     enum x86_transfer_type transfer,
1592 				     struct desc_struct *desc)
1593 {
1594 	struct desc_struct seg_desc, old_desc;
1595 	u8 dpl, rpl;
1596 	unsigned err_vec = GP_VECTOR;
1597 	u32 err_code = 0;
1598 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1599 	ulong desc_addr;
1600 	int ret;
1601 	u16 dummy;
1602 	u32 base3 = 0;
1603 
1604 	memset(&seg_desc, 0, sizeof seg_desc);
1605 
1606 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1607 		/* set real mode segment descriptor (keep limit etc. for
1608 		 * unreal mode) */
1609 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1610 		set_desc_base(&seg_desc, selector << 4);
1611 		goto load;
1612 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1613 		/* VM86 needs a clean new segment descriptor */
1614 		set_desc_base(&seg_desc, selector << 4);
1615 		set_desc_limit(&seg_desc, 0xffff);
1616 		seg_desc.type = 3;
1617 		seg_desc.p = 1;
1618 		seg_desc.s = 1;
1619 		seg_desc.dpl = 3;
1620 		goto load;
1621 	}
1622 
1623 	rpl = selector & 3;
1624 
1625 	/* TR should be in GDT only */
1626 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1627 		goto exception;
1628 
1629 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1630 	if (null_selector) {
1631 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1632 			goto exception;
1633 
1634 		if (seg == VCPU_SREG_SS) {
1635 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1636 				goto exception;
1637 
1638 			/*
1639 			 * ctxt->ops->set_segment expects the CPL to be in
1640 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1641 			 */
1642 			seg_desc.type = 3;
1643 			seg_desc.p = 1;
1644 			seg_desc.s = 1;
1645 			seg_desc.dpl = cpl;
1646 			seg_desc.d = 1;
1647 			seg_desc.g = 1;
1648 		}
1649 
1650 		/* Skip all following checks */
1651 		goto load;
1652 	}
1653 
1654 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1655 	if (ret != X86EMUL_CONTINUE)
1656 		return ret;
1657 
1658 	err_code = selector & 0xfffc;
1659 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1660 							   GP_VECTOR;
1661 
1662 	/* can't load system descriptor into segment selector */
1663 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1664 		if (transfer == X86_TRANSFER_CALL_JMP)
1665 			return X86EMUL_UNHANDLEABLE;
1666 		goto exception;
1667 	}
1668 
1669 	if (!seg_desc.p) {
1670 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1671 		goto exception;
1672 	}
1673 
1674 	dpl = seg_desc.dpl;
1675 
1676 	switch (seg) {
1677 	case VCPU_SREG_SS:
1678 		/*
1679 		 * segment is not a writable data segment or segment
1680 		 * selector's RPL != CPL or segment selector's RPL != CPL
1681 		 */
1682 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1683 			goto exception;
1684 		break;
1685 	case VCPU_SREG_CS:
1686 		if (!(seg_desc.type & 8))
1687 			goto exception;
1688 
1689 		if (seg_desc.type & 4) {
1690 			/* conforming */
1691 			if (dpl > cpl)
1692 				goto exception;
1693 		} else {
1694 			/* nonconforming */
1695 			if (rpl > cpl || dpl != cpl)
1696 				goto exception;
1697 		}
1698 		/* in long-mode d/b must be clear if l is set */
1699 		if (seg_desc.d && seg_desc.l) {
1700 			u64 efer = 0;
1701 
1702 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1703 			if (efer & EFER_LMA)
1704 				goto exception;
1705 		}
1706 
1707 		/* CS(RPL) <- CPL */
1708 		selector = (selector & 0xfffc) | cpl;
1709 		break;
1710 	case VCPU_SREG_TR:
1711 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1712 			goto exception;
1713 		old_desc = seg_desc;
1714 		seg_desc.type |= 2; /* busy */
1715 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1716 						  sizeof(seg_desc), &ctxt->exception);
1717 		if (ret != X86EMUL_CONTINUE)
1718 			return ret;
1719 		break;
1720 	case VCPU_SREG_LDTR:
1721 		if (seg_desc.s || seg_desc.type != 2)
1722 			goto exception;
1723 		break;
1724 	default: /*  DS, ES, FS, or GS */
1725 		/*
1726 		 * segment is not a data or readable code segment or
1727 		 * ((segment is a data or nonconforming code segment)
1728 		 * and (both RPL and CPL > DPL))
1729 		 */
1730 		if ((seg_desc.type & 0xa) == 0x8 ||
1731 		    (((seg_desc.type & 0xc) != 0xc) &&
1732 		     (rpl > dpl && cpl > dpl)))
1733 			goto exception;
1734 		break;
1735 	}
1736 
1737 	if (seg_desc.s) {
1738 		/* mark segment as accessed */
1739 		if (!(seg_desc.type & 1)) {
1740 			seg_desc.type |= 1;
1741 			ret = write_segment_descriptor(ctxt, selector,
1742 						       &seg_desc);
1743 			if (ret != X86EMUL_CONTINUE)
1744 				return ret;
1745 		}
1746 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1747 		ret = ctxt->ops->read_std(ctxt, desc_addr+8, &base3,
1748 				sizeof(base3), &ctxt->exception);
1749 		if (ret != X86EMUL_CONTINUE)
1750 			return ret;
1751 		if (is_noncanonical_address(get_desc_base(&seg_desc) |
1752 					     ((u64)base3 << 32)))
1753 			return emulate_gp(ctxt, 0);
1754 	}
1755 load:
1756 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1757 	if (desc)
1758 		*desc = seg_desc;
1759 	return X86EMUL_CONTINUE;
1760 exception:
1761 	return emulate_exception(ctxt, err_vec, err_code, true);
1762 }
1763 
1764 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1765 				   u16 selector, int seg)
1766 {
1767 	u8 cpl = ctxt->ops->cpl(ctxt);
1768 
1769 	/*
1770 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1771 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1772 	 * but it's wrong).
1773 	 *
1774 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1775 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1776 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1777 	 * and only forbid it here.
1778 	 */
1779 	if (seg == VCPU_SREG_SS && selector == 3 &&
1780 	    ctxt->mode == X86EMUL_MODE_PROT64)
1781 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1782 
1783 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1784 					 X86_TRANSFER_NONE, NULL);
1785 }
1786 
1787 static void write_register_operand(struct operand *op)
1788 {
1789 	return assign_register(op->addr.reg, op->val, op->bytes);
1790 }
1791 
1792 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1793 {
1794 	switch (op->type) {
1795 	case OP_REG:
1796 		write_register_operand(op);
1797 		break;
1798 	case OP_MEM:
1799 		if (ctxt->lock_prefix)
1800 			return segmented_cmpxchg(ctxt,
1801 						 op->addr.mem,
1802 						 &op->orig_val,
1803 						 &op->val,
1804 						 op->bytes);
1805 		else
1806 			return segmented_write(ctxt,
1807 					       op->addr.mem,
1808 					       &op->val,
1809 					       op->bytes);
1810 		break;
1811 	case OP_MEM_STR:
1812 		return segmented_write(ctxt,
1813 				       op->addr.mem,
1814 				       op->data,
1815 				       op->bytes * op->count);
1816 		break;
1817 	case OP_XMM:
1818 		write_sse_reg(ctxt, &op->vec_val, op->addr.xmm);
1819 		break;
1820 	case OP_MM:
1821 		write_mmx_reg(ctxt, &op->mm_val, op->addr.mm);
1822 		break;
1823 	case OP_NONE:
1824 		/* no writeback */
1825 		break;
1826 	default:
1827 		break;
1828 	}
1829 	return X86EMUL_CONTINUE;
1830 }
1831 
1832 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1833 {
1834 	struct segmented_address addr;
1835 
1836 	rsp_increment(ctxt, -bytes);
1837 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1838 	addr.seg = VCPU_SREG_SS;
1839 
1840 	return segmented_write(ctxt, addr, data, bytes);
1841 }
1842 
1843 static int em_push(struct x86_emulate_ctxt *ctxt)
1844 {
1845 	/* Disable writeback. */
1846 	ctxt->dst.type = OP_NONE;
1847 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1848 }
1849 
1850 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1851 		       void *dest, int len)
1852 {
1853 	int rc;
1854 	struct segmented_address addr;
1855 
1856 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1857 	addr.seg = VCPU_SREG_SS;
1858 	rc = segmented_read(ctxt, addr, dest, len);
1859 	if (rc != X86EMUL_CONTINUE)
1860 		return rc;
1861 
1862 	rsp_increment(ctxt, len);
1863 	return rc;
1864 }
1865 
1866 static int em_pop(struct x86_emulate_ctxt *ctxt)
1867 {
1868 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1869 }
1870 
1871 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1872 			void *dest, int len)
1873 {
1874 	int rc;
1875 	unsigned long val, change_mask;
1876 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1877 	int cpl = ctxt->ops->cpl(ctxt);
1878 
1879 	rc = emulate_pop(ctxt, &val, len);
1880 	if (rc != X86EMUL_CONTINUE)
1881 		return rc;
1882 
1883 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1884 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1885 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1886 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1887 
1888 	switch(ctxt->mode) {
1889 	case X86EMUL_MODE_PROT64:
1890 	case X86EMUL_MODE_PROT32:
1891 	case X86EMUL_MODE_PROT16:
1892 		if (cpl == 0)
1893 			change_mask |= X86_EFLAGS_IOPL;
1894 		if (cpl <= iopl)
1895 			change_mask |= X86_EFLAGS_IF;
1896 		break;
1897 	case X86EMUL_MODE_VM86:
1898 		if (iopl < 3)
1899 			return emulate_gp(ctxt, 0);
1900 		change_mask |= X86_EFLAGS_IF;
1901 		break;
1902 	default: /* real mode */
1903 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1904 		break;
1905 	}
1906 
1907 	*(unsigned long *)dest =
1908 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1909 
1910 	return rc;
1911 }
1912 
1913 static int em_popf(struct x86_emulate_ctxt *ctxt)
1914 {
1915 	ctxt->dst.type = OP_REG;
1916 	ctxt->dst.addr.reg = &ctxt->eflags;
1917 	ctxt->dst.bytes = ctxt->op_bytes;
1918 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1919 }
1920 
1921 static int em_enter(struct x86_emulate_ctxt *ctxt)
1922 {
1923 	int rc;
1924 	unsigned frame_size = ctxt->src.val;
1925 	unsigned nesting_level = ctxt->src2.val & 31;
1926 	ulong rbp;
1927 
1928 	if (nesting_level)
1929 		return X86EMUL_UNHANDLEABLE;
1930 
1931 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1932 	rc = push(ctxt, &rbp, stack_size(ctxt));
1933 	if (rc != X86EMUL_CONTINUE)
1934 		return rc;
1935 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1936 		      stack_mask(ctxt));
1937 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1938 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1939 		      stack_mask(ctxt));
1940 	return X86EMUL_CONTINUE;
1941 }
1942 
1943 static int em_leave(struct x86_emulate_ctxt *ctxt)
1944 {
1945 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1946 		      stack_mask(ctxt));
1947 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1948 }
1949 
1950 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1951 {
1952 	int seg = ctxt->src2.val;
1953 
1954 	ctxt->src.val = get_segment_selector(ctxt, seg);
1955 	if (ctxt->op_bytes == 4) {
1956 		rsp_increment(ctxt, -2);
1957 		ctxt->op_bytes = 2;
1958 	}
1959 
1960 	return em_push(ctxt);
1961 }
1962 
1963 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1964 {
1965 	int seg = ctxt->src2.val;
1966 	unsigned long selector;
1967 	int rc;
1968 
1969 	rc = emulate_pop(ctxt, &selector, 2);
1970 	if (rc != X86EMUL_CONTINUE)
1971 		return rc;
1972 
1973 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1974 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1975 	if (ctxt->op_bytes > 2)
1976 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1977 
1978 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1979 	return rc;
1980 }
1981 
1982 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1983 {
1984 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1985 	int rc = X86EMUL_CONTINUE;
1986 	int reg = VCPU_REGS_RAX;
1987 
1988 	while (reg <= VCPU_REGS_RDI) {
1989 		(reg == VCPU_REGS_RSP) ?
1990 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1991 
1992 		rc = em_push(ctxt);
1993 		if (rc != X86EMUL_CONTINUE)
1994 			return rc;
1995 
1996 		++reg;
1997 	}
1998 
1999 	return rc;
2000 }
2001 
2002 static int em_pushf(struct x86_emulate_ctxt *ctxt)
2003 {
2004 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
2005 	return em_push(ctxt);
2006 }
2007 
2008 static int em_popa(struct x86_emulate_ctxt *ctxt)
2009 {
2010 	int rc = X86EMUL_CONTINUE;
2011 	int reg = VCPU_REGS_RDI;
2012 	u32 val;
2013 
2014 	while (reg >= VCPU_REGS_RAX) {
2015 		if (reg == VCPU_REGS_RSP) {
2016 			rsp_increment(ctxt, ctxt->op_bytes);
2017 			--reg;
2018 		}
2019 
2020 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2021 		if (rc != X86EMUL_CONTINUE)
2022 			break;
2023 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2024 		--reg;
2025 	}
2026 	return rc;
2027 }
2028 
2029 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2030 {
2031 	const struct x86_emulate_ops *ops = ctxt->ops;
2032 	int rc;
2033 	struct desc_ptr dt;
2034 	gva_t cs_addr;
2035 	gva_t eip_addr;
2036 	u16 cs, eip;
2037 
2038 	/* TODO: Add limit checks */
2039 	ctxt->src.val = ctxt->eflags;
2040 	rc = em_push(ctxt);
2041 	if (rc != X86EMUL_CONTINUE)
2042 		return rc;
2043 
2044 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2045 
2046 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2047 	rc = em_push(ctxt);
2048 	if (rc != X86EMUL_CONTINUE)
2049 		return rc;
2050 
2051 	ctxt->src.val = ctxt->_eip;
2052 	rc = em_push(ctxt);
2053 	if (rc != X86EMUL_CONTINUE)
2054 		return rc;
2055 
2056 	ops->get_idt(ctxt, &dt);
2057 
2058 	eip_addr = dt.address + (irq << 2);
2059 	cs_addr = dt.address + (irq << 2) + 2;
2060 
2061 	rc = ops->read_std(ctxt, cs_addr, &cs, 2, &ctxt->exception);
2062 	if (rc != X86EMUL_CONTINUE)
2063 		return rc;
2064 
2065 	rc = ops->read_std(ctxt, eip_addr, &eip, 2, &ctxt->exception);
2066 	if (rc != X86EMUL_CONTINUE)
2067 		return rc;
2068 
2069 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2070 	if (rc != X86EMUL_CONTINUE)
2071 		return rc;
2072 
2073 	ctxt->_eip = eip;
2074 
2075 	return rc;
2076 }
2077 
2078 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2079 {
2080 	int rc;
2081 
2082 	invalidate_registers(ctxt);
2083 	rc = __emulate_int_real(ctxt, irq);
2084 	if (rc == X86EMUL_CONTINUE)
2085 		writeback_registers(ctxt);
2086 	return rc;
2087 }
2088 
2089 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2090 {
2091 	switch(ctxt->mode) {
2092 	case X86EMUL_MODE_REAL:
2093 		return __emulate_int_real(ctxt, irq);
2094 	case X86EMUL_MODE_VM86:
2095 	case X86EMUL_MODE_PROT16:
2096 	case X86EMUL_MODE_PROT32:
2097 	case X86EMUL_MODE_PROT64:
2098 	default:
2099 		/* Protected mode interrupts unimplemented yet */
2100 		return X86EMUL_UNHANDLEABLE;
2101 	}
2102 }
2103 
2104 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2105 {
2106 	int rc = X86EMUL_CONTINUE;
2107 	unsigned long temp_eip = 0;
2108 	unsigned long temp_eflags = 0;
2109 	unsigned long cs = 0;
2110 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2111 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2112 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2113 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2114 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2115 			     X86_EFLAGS_FIXED;
2116 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2117 				  X86_EFLAGS_VIP;
2118 
2119 	/* TODO: Add stack limit check */
2120 
2121 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2122 
2123 	if (rc != X86EMUL_CONTINUE)
2124 		return rc;
2125 
2126 	if (temp_eip & ~0xffff)
2127 		return emulate_gp(ctxt, 0);
2128 
2129 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2130 
2131 	if (rc != X86EMUL_CONTINUE)
2132 		return rc;
2133 
2134 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2135 
2136 	if (rc != X86EMUL_CONTINUE)
2137 		return rc;
2138 
2139 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2140 
2141 	if (rc != X86EMUL_CONTINUE)
2142 		return rc;
2143 
2144 	ctxt->_eip = temp_eip;
2145 
2146 	if (ctxt->op_bytes == 4)
2147 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2148 	else if (ctxt->op_bytes == 2) {
2149 		ctxt->eflags &= ~0xffff;
2150 		ctxt->eflags |= temp_eflags;
2151 	}
2152 
2153 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2154 	ctxt->eflags |= X86_EFLAGS_FIXED;
2155 	ctxt->ops->set_nmi_mask(ctxt, false);
2156 
2157 	return rc;
2158 }
2159 
2160 static int em_iret(struct x86_emulate_ctxt *ctxt)
2161 {
2162 	switch(ctxt->mode) {
2163 	case X86EMUL_MODE_REAL:
2164 		return emulate_iret_real(ctxt);
2165 	case X86EMUL_MODE_VM86:
2166 	case X86EMUL_MODE_PROT16:
2167 	case X86EMUL_MODE_PROT32:
2168 	case X86EMUL_MODE_PROT64:
2169 	default:
2170 		/* iret from protected mode unimplemented yet */
2171 		return X86EMUL_UNHANDLEABLE;
2172 	}
2173 }
2174 
2175 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2176 {
2177 	int rc;
2178 	unsigned short sel;
2179 	struct desc_struct new_desc;
2180 	u8 cpl = ctxt->ops->cpl(ctxt);
2181 
2182 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2183 
2184 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2185 				       X86_TRANSFER_CALL_JMP,
2186 				       &new_desc);
2187 	if (rc != X86EMUL_CONTINUE)
2188 		return rc;
2189 
2190 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2191 	/* Error handling is not implemented. */
2192 	if (rc != X86EMUL_CONTINUE)
2193 		return X86EMUL_UNHANDLEABLE;
2194 
2195 	return rc;
2196 }
2197 
2198 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2199 {
2200 	return assign_eip_near(ctxt, ctxt->src.val);
2201 }
2202 
2203 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2204 {
2205 	int rc;
2206 	long int old_eip;
2207 
2208 	old_eip = ctxt->_eip;
2209 	rc = assign_eip_near(ctxt, ctxt->src.val);
2210 	if (rc != X86EMUL_CONTINUE)
2211 		return rc;
2212 	ctxt->src.val = old_eip;
2213 	rc = em_push(ctxt);
2214 	return rc;
2215 }
2216 
2217 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2218 {
2219 	u64 old = ctxt->dst.orig_val64;
2220 
2221 	if (ctxt->dst.bytes == 16)
2222 		return X86EMUL_UNHANDLEABLE;
2223 
2224 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2225 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2226 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2227 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2228 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2229 	} else {
2230 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2231 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2232 
2233 		ctxt->eflags |= X86_EFLAGS_ZF;
2234 	}
2235 	return X86EMUL_CONTINUE;
2236 }
2237 
2238 static int em_ret(struct x86_emulate_ctxt *ctxt)
2239 {
2240 	int rc;
2241 	unsigned long eip;
2242 
2243 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2244 	if (rc != X86EMUL_CONTINUE)
2245 		return rc;
2246 
2247 	return assign_eip_near(ctxt, eip);
2248 }
2249 
2250 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2251 {
2252 	int rc;
2253 	unsigned long eip, cs;
2254 	int cpl = ctxt->ops->cpl(ctxt);
2255 	struct desc_struct new_desc;
2256 
2257 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2258 	if (rc != X86EMUL_CONTINUE)
2259 		return rc;
2260 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2261 	if (rc != X86EMUL_CONTINUE)
2262 		return rc;
2263 	/* Outer-privilege level return is not implemented */
2264 	if (ctxt->mode >= X86EMUL_MODE_PROT16 && (cs & 3) > cpl)
2265 		return X86EMUL_UNHANDLEABLE;
2266 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2267 				       X86_TRANSFER_RET,
2268 				       &new_desc);
2269 	if (rc != X86EMUL_CONTINUE)
2270 		return rc;
2271 	rc = assign_eip_far(ctxt, eip, &new_desc);
2272 	/* Error handling is not implemented. */
2273 	if (rc != X86EMUL_CONTINUE)
2274 		return X86EMUL_UNHANDLEABLE;
2275 
2276 	return rc;
2277 }
2278 
2279 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2280 {
2281         int rc;
2282 
2283         rc = em_ret_far(ctxt);
2284         if (rc != X86EMUL_CONTINUE)
2285                 return rc;
2286         rsp_increment(ctxt, ctxt->src.val);
2287         return X86EMUL_CONTINUE;
2288 }
2289 
2290 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2291 {
2292 	/* Save real source value, then compare EAX against destination. */
2293 	ctxt->dst.orig_val = ctxt->dst.val;
2294 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2295 	ctxt->src.orig_val = ctxt->src.val;
2296 	ctxt->src.val = ctxt->dst.orig_val;
2297 	fastop(ctxt, em_cmp);
2298 
2299 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2300 		/* Success: write back to memory; no update of EAX */
2301 		ctxt->src.type = OP_NONE;
2302 		ctxt->dst.val = ctxt->src.orig_val;
2303 	} else {
2304 		/* Failure: write the value we saw to EAX. */
2305 		ctxt->src.type = OP_REG;
2306 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2307 		ctxt->src.val = ctxt->dst.orig_val;
2308 		/* Create write-cycle to dest by writing the same value */
2309 		ctxt->dst.val = ctxt->dst.orig_val;
2310 	}
2311 	return X86EMUL_CONTINUE;
2312 }
2313 
2314 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2315 {
2316 	int seg = ctxt->src2.val;
2317 	unsigned short sel;
2318 	int rc;
2319 
2320 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2321 
2322 	rc = load_segment_descriptor(ctxt, sel, seg);
2323 	if (rc != X86EMUL_CONTINUE)
2324 		return rc;
2325 
2326 	ctxt->dst.val = ctxt->src.val;
2327 	return rc;
2328 }
2329 
2330 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2331 {
2332 	u32 eax, ebx, ecx, edx;
2333 
2334 	eax = 0x80000001;
2335 	ecx = 0;
2336 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
2337 	return edx & bit(X86_FEATURE_LM);
2338 }
2339 
2340 #define GET_SMSTATE(type, smbase, offset)				  \
2341 	({								  \
2342 	 type __val;							  \
2343 	 int r = ctxt->ops->read_phys(ctxt, smbase + offset, &__val,      \
2344 				      sizeof(__val));			  \
2345 	 if (r != X86EMUL_CONTINUE)					  \
2346 		 return X86EMUL_UNHANDLEABLE;				  \
2347 	 __val;								  \
2348 	})
2349 
2350 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2351 {
2352 	desc->g    = (flags >> 23) & 1;
2353 	desc->d    = (flags >> 22) & 1;
2354 	desc->l    = (flags >> 21) & 1;
2355 	desc->avl  = (flags >> 20) & 1;
2356 	desc->p    = (flags >> 15) & 1;
2357 	desc->dpl  = (flags >> 13) & 3;
2358 	desc->s    = (flags >> 12) & 1;
2359 	desc->type = (flags >>  8) & 15;
2360 }
2361 
2362 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, u64 smbase, int n)
2363 {
2364 	struct desc_struct desc;
2365 	int offset;
2366 	u16 selector;
2367 
2368 	selector = GET_SMSTATE(u32, smbase, 0x7fa8 + n * 4);
2369 
2370 	if (n < 3)
2371 		offset = 0x7f84 + n * 12;
2372 	else
2373 		offset = 0x7f2c + (n - 3) * 12;
2374 
2375 	set_desc_base(&desc,      GET_SMSTATE(u32, smbase, offset + 8));
2376 	set_desc_limit(&desc,     GET_SMSTATE(u32, smbase, offset + 4));
2377 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smbase, offset));
2378 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2379 	return X86EMUL_CONTINUE;
2380 }
2381 
2382 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, u64 smbase, int n)
2383 {
2384 	struct desc_struct desc;
2385 	int offset;
2386 	u16 selector;
2387 	u32 base3;
2388 
2389 	offset = 0x7e00 + n * 16;
2390 
2391 	selector =                GET_SMSTATE(u16, smbase, offset);
2392 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smbase, offset + 2) << 8);
2393 	set_desc_limit(&desc,     GET_SMSTATE(u32, smbase, offset + 4));
2394 	set_desc_base(&desc,      GET_SMSTATE(u32, smbase, offset + 8));
2395 	base3 =                   GET_SMSTATE(u32, smbase, offset + 12);
2396 
2397 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2398 	return X86EMUL_CONTINUE;
2399 }
2400 
2401 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2402 				     u64 cr0, u64 cr4)
2403 {
2404 	int bad;
2405 
2406 	/*
2407 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2408 	 * Then enable protected mode.	However, PCID cannot be enabled
2409 	 * if EFER.LMA=0, so set it separately.
2410 	 */
2411 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2412 	if (bad)
2413 		return X86EMUL_UNHANDLEABLE;
2414 
2415 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2416 	if (bad)
2417 		return X86EMUL_UNHANDLEABLE;
2418 
2419 	if (cr4 & X86_CR4_PCIDE) {
2420 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2421 		if (bad)
2422 			return X86EMUL_UNHANDLEABLE;
2423 	}
2424 
2425 	return X86EMUL_CONTINUE;
2426 }
2427 
2428 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt, u64 smbase)
2429 {
2430 	struct desc_struct desc;
2431 	struct desc_ptr dt;
2432 	u16 selector;
2433 	u32 val, cr0, cr4;
2434 	int i;
2435 
2436 	cr0 =                      GET_SMSTATE(u32, smbase, 0x7ffc);
2437 	ctxt->ops->set_cr(ctxt, 3, GET_SMSTATE(u32, smbase, 0x7ff8));
2438 	ctxt->eflags =             GET_SMSTATE(u32, smbase, 0x7ff4) | X86_EFLAGS_FIXED;
2439 	ctxt->_eip =               GET_SMSTATE(u32, smbase, 0x7ff0);
2440 
2441 	for (i = 0; i < 8; i++)
2442 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smbase, 0x7fd0 + i * 4);
2443 
2444 	val = GET_SMSTATE(u32, smbase, 0x7fcc);
2445 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2446 	val = GET_SMSTATE(u32, smbase, 0x7fc8);
2447 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2448 
2449 	selector =                 GET_SMSTATE(u32, smbase, 0x7fc4);
2450 	set_desc_base(&desc,       GET_SMSTATE(u32, smbase, 0x7f64));
2451 	set_desc_limit(&desc,      GET_SMSTATE(u32, smbase, 0x7f60));
2452 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smbase, 0x7f5c));
2453 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2454 
2455 	selector =                 GET_SMSTATE(u32, smbase, 0x7fc0);
2456 	set_desc_base(&desc,       GET_SMSTATE(u32, smbase, 0x7f80));
2457 	set_desc_limit(&desc,      GET_SMSTATE(u32, smbase, 0x7f7c));
2458 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smbase, 0x7f78));
2459 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2460 
2461 	dt.address =               GET_SMSTATE(u32, smbase, 0x7f74);
2462 	dt.size =                  GET_SMSTATE(u32, smbase, 0x7f70);
2463 	ctxt->ops->set_gdt(ctxt, &dt);
2464 
2465 	dt.address =               GET_SMSTATE(u32, smbase, 0x7f58);
2466 	dt.size =                  GET_SMSTATE(u32, smbase, 0x7f54);
2467 	ctxt->ops->set_idt(ctxt, &dt);
2468 
2469 	for (i = 0; i < 6; i++) {
2470 		int r = rsm_load_seg_32(ctxt, smbase, i);
2471 		if (r != X86EMUL_CONTINUE)
2472 			return r;
2473 	}
2474 
2475 	cr4 = GET_SMSTATE(u32, smbase, 0x7f14);
2476 
2477 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smbase, 0x7ef8));
2478 
2479 	return rsm_enter_protected_mode(ctxt, cr0, cr4);
2480 }
2481 
2482 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt, u64 smbase)
2483 {
2484 	struct desc_struct desc;
2485 	struct desc_ptr dt;
2486 	u64 val, cr0, cr4;
2487 	u32 base3;
2488 	u16 selector;
2489 	int i, r;
2490 
2491 	for (i = 0; i < 16; i++)
2492 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smbase, 0x7ff8 - i * 8);
2493 
2494 	ctxt->_eip   = GET_SMSTATE(u64, smbase, 0x7f78);
2495 	ctxt->eflags = GET_SMSTATE(u32, smbase, 0x7f70) | X86_EFLAGS_FIXED;
2496 
2497 	val = GET_SMSTATE(u32, smbase, 0x7f68);
2498 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2499 	val = GET_SMSTATE(u32, smbase, 0x7f60);
2500 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2501 
2502 	cr0 =                       GET_SMSTATE(u64, smbase, 0x7f58);
2503 	ctxt->ops->set_cr(ctxt, 3,  GET_SMSTATE(u64, smbase, 0x7f50));
2504 	cr4 =                       GET_SMSTATE(u64, smbase, 0x7f48);
2505 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smbase, 0x7f00));
2506 	val =                       GET_SMSTATE(u64, smbase, 0x7ed0);
2507 	ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA);
2508 
2509 	selector =                  GET_SMSTATE(u32, smbase, 0x7e90);
2510 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smbase, 0x7e92) << 8);
2511 	set_desc_limit(&desc,       GET_SMSTATE(u32, smbase, 0x7e94));
2512 	set_desc_base(&desc,        GET_SMSTATE(u32, smbase, 0x7e98));
2513 	base3 =                     GET_SMSTATE(u32, smbase, 0x7e9c);
2514 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2515 
2516 	dt.size =                   GET_SMSTATE(u32, smbase, 0x7e84);
2517 	dt.address =                GET_SMSTATE(u64, smbase, 0x7e88);
2518 	ctxt->ops->set_idt(ctxt, &dt);
2519 
2520 	selector =                  GET_SMSTATE(u32, smbase, 0x7e70);
2521 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smbase, 0x7e72) << 8);
2522 	set_desc_limit(&desc,       GET_SMSTATE(u32, smbase, 0x7e74));
2523 	set_desc_base(&desc,        GET_SMSTATE(u32, smbase, 0x7e78));
2524 	base3 =                     GET_SMSTATE(u32, smbase, 0x7e7c);
2525 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2526 
2527 	dt.size =                   GET_SMSTATE(u32, smbase, 0x7e64);
2528 	dt.address =                GET_SMSTATE(u64, smbase, 0x7e68);
2529 	ctxt->ops->set_gdt(ctxt, &dt);
2530 
2531 	r = rsm_enter_protected_mode(ctxt, cr0, cr4);
2532 	if (r != X86EMUL_CONTINUE)
2533 		return r;
2534 
2535 	for (i = 0; i < 6; i++) {
2536 		r = rsm_load_seg_64(ctxt, smbase, i);
2537 		if (r != X86EMUL_CONTINUE)
2538 			return r;
2539 	}
2540 
2541 	return X86EMUL_CONTINUE;
2542 }
2543 
2544 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2545 {
2546 	unsigned long cr0, cr4, efer;
2547 	u64 smbase;
2548 	int ret;
2549 
2550 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2551 		return emulate_ud(ctxt);
2552 
2553 	/*
2554 	 * Get back to real mode, to prepare a safe state in which to load
2555 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2556 	 * supports long mode.
2557 	 */
2558 	cr4 = ctxt->ops->get_cr(ctxt, 4);
2559 	if (emulator_has_longmode(ctxt)) {
2560 		struct desc_struct cs_desc;
2561 
2562 		/* Zero CR4.PCIDE before CR0.PG.  */
2563 		if (cr4 & X86_CR4_PCIDE) {
2564 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2565 			cr4 &= ~X86_CR4_PCIDE;
2566 		}
2567 
2568 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2569 		memset(&cs_desc, 0, sizeof(cs_desc));
2570 		cs_desc.type = 0xb;
2571 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2572 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2573 	}
2574 
2575 	/* For the 64-bit case, this will clear EFER.LMA.  */
2576 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2577 	if (cr0 & X86_CR0_PE)
2578 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2579 
2580 	/* Now clear CR4.PAE (which must be done before clearing EFER.LME).  */
2581 	if (cr4 & X86_CR4_PAE)
2582 		ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2583 
2584 	/* And finally go back to 32-bit mode.  */
2585 	efer = 0;
2586 	ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2587 
2588 	smbase = ctxt->ops->get_smbase(ctxt);
2589 	if (emulator_has_longmode(ctxt))
2590 		ret = rsm_load_state_64(ctxt, smbase + 0x8000);
2591 	else
2592 		ret = rsm_load_state_32(ctxt, smbase + 0x8000);
2593 
2594 	if (ret != X86EMUL_CONTINUE) {
2595 		/* FIXME: should triple fault */
2596 		return X86EMUL_UNHANDLEABLE;
2597 	}
2598 
2599 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2600 		ctxt->ops->set_nmi_mask(ctxt, false);
2601 
2602 	ctxt->ops->set_hflags(ctxt, ctxt->ops->get_hflags(ctxt) &
2603 		~(X86EMUL_SMM_INSIDE_NMI_MASK | X86EMUL_SMM_MASK));
2604 	return X86EMUL_CONTINUE;
2605 }
2606 
2607 static void
2608 setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
2609 			struct desc_struct *cs, struct desc_struct *ss)
2610 {
2611 	cs->l = 0;		/* will be adjusted later */
2612 	set_desc_base(cs, 0);	/* flat segment */
2613 	cs->g = 1;		/* 4kb granularity */
2614 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2615 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2616 	cs->s = 1;
2617 	cs->dpl = 0;		/* will be adjusted later */
2618 	cs->p = 1;
2619 	cs->d = 1;
2620 	cs->avl = 0;
2621 
2622 	set_desc_base(ss, 0);	/* flat segment */
2623 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2624 	ss->g = 1;		/* 4kb granularity */
2625 	ss->s = 1;
2626 	ss->type = 0x03;	/* Read/Write, Accessed */
2627 	ss->d = 1;		/* 32bit stack segment */
2628 	ss->dpl = 0;
2629 	ss->p = 1;
2630 	ss->l = 0;
2631 	ss->avl = 0;
2632 }
2633 
2634 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2635 {
2636 	u32 eax, ebx, ecx, edx;
2637 
2638 	eax = ecx = 0;
2639 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
2640 	return ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx
2641 		&& ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx
2642 		&& edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx;
2643 }
2644 
2645 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2646 {
2647 	const struct x86_emulate_ops *ops = ctxt->ops;
2648 	u32 eax, ebx, ecx, edx;
2649 
2650 	/*
2651 	 * syscall should always be enabled in longmode - so only become
2652 	 * vendor specific (cpuid) if other modes are active...
2653 	 */
2654 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2655 		return true;
2656 
2657 	eax = 0x00000000;
2658 	ecx = 0x00000000;
2659 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
2660 	/*
2661 	 * Intel ("GenuineIntel")
2662 	 * remark: Intel CPUs only support "syscall" in 64bit
2663 	 * longmode. Also an 64bit guest with a
2664 	 * 32bit compat-app running will #UD !! While this
2665 	 * behaviour can be fixed (by emulating) into AMD
2666 	 * response - CPUs of AMD can't behave like Intel.
2667 	 */
2668 	if (ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx &&
2669 	    ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx &&
2670 	    edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx)
2671 		return false;
2672 
2673 	/* AMD ("AuthenticAMD") */
2674 	if (ebx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ebx &&
2675 	    ecx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ecx &&
2676 	    edx == X86EMUL_CPUID_VENDOR_AuthenticAMD_edx)
2677 		return true;
2678 
2679 	/* AMD ("AMDisbetter!") */
2680 	if (ebx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ebx &&
2681 	    ecx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ecx &&
2682 	    edx == X86EMUL_CPUID_VENDOR_AMDisbetterI_edx)
2683 		return true;
2684 
2685 	/* default: (not Intel, not AMD), apply Intel's stricter rules... */
2686 	return false;
2687 }
2688 
2689 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2690 {
2691 	const struct x86_emulate_ops *ops = ctxt->ops;
2692 	struct desc_struct cs, ss;
2693 	u64 msr_data;
2694 	u16 cs_sel, ss_sel;
2695 	u64 efer = 0;
2696 
2697 	/* syscall is not available in real mode */
2698 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2699 	    ctxt->mode == X86EMUL_MODE_VM86)
2700 		return emulate_ud(ctxt);
2701 
2702 	if (!(em_syscall_is_enabled(ctxt)))
2703 		return emulate_ud(ctxt);
2704 
2705 	ops->get_msr(ctxt, MSR_EFER, &efer);
2706 	setup_syscalls_segments(ctxt, &cs, &ss);
2707 
2708 	if (!(efer & EFER_SCE))
2709 		return emulate_ud(ctxt);
2710 
2711 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2712 	msr_data >>= 32;
2713 	cs_sel = (u16)(msr_data & 0xfffc);
2714 	ss_sel = (u16)(msr_data + 8);
2715 
2716 	if (efer & EFER_LMA) {
2717 		cs.d = 0;
2718 		cs.l = 1;
2719 	}
2720 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2721 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2722 
2723 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2724 	if (efer & EFER_LMA) {
2725 #ifdef CONFIG_X86_64
2726 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2727 
2728 		ops->get_msr(ctxt,
2729 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2730 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2731 		ctxt->_eip = msr_data;
2732 
2733 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2734 		ctxt->eflags &= ~msr_data;
2735 		ctxt->eflags |= X86_EFLAGS_FIXED;
2736 #endif
2737 	} else {
2738 		/* legacy mode */
2739 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2740 		ctxt->_eip = (u32)msr_data;
2741 
2742 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2743 	}
2744 
2745 	return X86EMUL_CONTINUE;
2746 }
2747 
2748 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2749 {
2750 	const struct x86_emulate_ops *ops = ctxt->ops;
2751 	struct desc_struct cs, ss;
2752 	u64 msr_data;
2753 	u16 cs_sel, ss_sel;
2754 	u64 efer = 0;
2755 
2756 	ops->get_msr(ctxt, MSR_EFER, &efer);
2757 	/* inject #GP if in real mode */
2758 	if (ctxt->mode == X86EMUL_MODE_REAL)
2759 		return emulate_gp(ctxt, 0);
2760 
2761 	/*
2762 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2763 	 * mode).
2764 	 */
2765 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2766 	    && !vendor_intel(ctxt))
2767 		return emulate_ud(ctxt);
2768 
2769 	/* sysenter/sysexit have not been tested in 64bit mode. */
2770 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2771 		return X86EMUL_UNHANDLEABLE;
2772 
2773 	setup_syscalls_segments(ctxt, &cs, &ss);
2774 
2775 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2776 	if ((msr_data & 0xfffc) == 0x0)
2777 		return emulate_gp(ctxt, 0);
2778 
2779 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2780 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2781 	ss_sel = cs_sel + 8;
2782 	if (efer & EFER_LMA) {
2783 		cs.d = 0;
2784 		cs.l = 1;
2785 	}
2786 
2787 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2788 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2789 
2790 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2791 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2792 
2793 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2794 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2795 							      (u32)msr_data;
2796 
2797 	return X86EMUL_CONTINUE;
2798 }
2799 
2800 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2801 {
2802 	const struct x86_emulate_ops *ops = ctxt->ops;
2803 	struct desc_struct cs, ss;
2804 	u64 msr_data, rcx, rdx;
2805 	int usermode;
2806 	u16 cs_sel = 0, ss_sel = 0;
2807 
2808 	/* inject #GP if in real mode or Virtual 8086 mode */
2809 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2810 	    ctxt->mode == X86EMUL_MODE_VM86)
2811 		return emulate_gp(ctxt, 0);
2812 
2813 	setup_syscalls_segments(ctxt, &cs, &ss);
2814 
2815 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2816 		usermode = X86EMUL_MODE_PROT64;
2817 	else
2818 		usermode = X86EMUL_MODE_PROT32;
2819 
2820 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2821 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2822 
2823 	cs.dpl = 3;
2824 	ss.dpl = 3;
2825 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2826 	switch (usermode) {
2827 	case X86EMUL_MODE_PROT32:
2828 		cs_sel = (u16)(msr_data + 16);
2829 		if ((msr_data & 0xfffc) == 0x0)
2830 			return emulate_gp(ctxt, 0);
2831 		ss_sel = (u16)(msr_data + 24);
2832 		rcx = (u32)rcx;
2833 		rdx = (u32)rdx;
2834 		break;
2835 	case X86EMUL_MODE_PROT64:
2836 		cs_sel = (u16)(msr_data + 32);
2837 		if (msr_data == 0x0)
2838 			return emulate_gp(ctxt, 0);
2839 		ss_sel = cs_sel + 8;
2840 		cs.d = 0;
2841 		cs.l = 1;
2842 		if (is_noncanonical_address(rcx) ||
2843 		    is_noncanonical_address(rdx))
2844 			return emulate_gp(ctxt, 0);
2845 		break;
2846 	}
2847 	cs_sel |= SEGMENT_RPL_MASK;
2848 	ss_sel |= SEGMENT_RPL_MASK;
2849 
2850 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2851 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2852 
2853 	ctxt->_eip = rdx;
2854 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2855 
2856 	return X86EMUL_CONTINUE;
2857 }
2858 
2859 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2860 {
2861 	int iopl;
2862 	if (ctxt->mode == X86EMUL_MODE_REAL)
2863 		return false;
2864 	if (ctxt->mode == X86EMUL_MODE_VM86)
2865 		return true;
2866 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2867 	return ctxt->ops->cpl(ctxt) > iopl;
2868 }
2869 
2870 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2871 					    u16 port, u16 len)
2872 {
2873 	const struct x86_emulate_ops *ops = ctxt->ops;
2874 	struct desc_struct tr_seg;
2875 	u32 base3;
2876 	int r;
2877 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2878 	unsigned mask = (1 << len) - 1;
2879 	unsigned long base;
2880 
2881 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2882 	if (!tr_seg.p)
2883 		return false;
2884 	if (desc_limit_scaled(&tr_seg) < 103)
2885 		return false;
2886 	base = get_desc_base(&tr_seg);
2887 #ifdef CONFIG_X86_64
2888 	base |= ((u64)base3) << 32;
2889 #endif
2890 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL);
2891 	if (r != X86EMUL_CONTINUE)
2892 		return false;
2893 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2894 		return false;
2895 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL);
2896 	if (r != X86EMUL_CONTINUE)
2897 		return false;
2898 	if ((perm >> bit_idx) & mask)
2899 		return false;
2900 	return true;
2901 }
2902 
2903 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2904 				 u16 port, u16 len)
2905 {
2906 	if (ctxt->perm_ok)
2907 		return true;
2908 
2909 	if (emulator_bad_iopl(ctxt))
2910 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2911 			return false;
2912 
2913 	ctxt->perm_ok = true;
2914 
2915 	return true;
2916 }
2917 
2918 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2919 {
2920 	/*
2921 	 * Intel CPUs mask the counter and pointers in quite strange
2922 	 * manner when ECX is zero due to REP-string optimizations.
2923 	 */
2924 #ifdef CONFIG_X86_64
2925 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2926 		return;
2927 
2928 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2929 
2930 	switch (ctxt->b) {
2931 	case 0xa4:	/* movsb */
2932 	case 0xa5:	/* movsd/w */
2933 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2934 		/* fall through */
2935 	case 0xaa:	/* stosb */
2936 	case 0xab:	/* stosd/w */
2937 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2938 	}
2939 #endif
2940 }
2941 
2942 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2943 				struct tss_segment_16 *tss)
2944 {
2945 	tss->ip = ctxt->_eip;
2946 	tss->flag = ctxt->eflags;
2947 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2948 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2949 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2950 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2951 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2952 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2953 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2954 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2955 
2956 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2957 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2958 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2959 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2960 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2961 }
2962 
2963 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2964 				 struct tss_segment_16 *tss)
2965 {
2966 	int ret;
2967 	u8 cpl;
2968 
2969 	ctxt->_eip = tss->ip;
2970 	ctxt->eflags = tss->flag | 2;
2971 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2972 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2973 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2974 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2975 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2976 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2977 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2978 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2979 
2980 	/*
2981 	 * SDM says that segment selectors are loaded before segment
2982 	 * descriptors
2983 	 */
2984 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
2985 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2986 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2987 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2988 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2989 
2990 	cpl = tss->cs & 3;
2991 
2992 	/*
2993 	 * Now load segment descriptors. If fault happens at this stage
2994 	 * it is handled in a context of new task
2995 	 */
2996 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
2997 					X86_TRANSFER_TASK_SWITCH, NULL);
2998 	if (ret != X86EMUL_CONTINUE)
2999 		return ret;
3000 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3001 					X86_TRANSFER_TASK_SWITCH, NULL);
3002 	if (ret != X86EMUL_CONTINUE)
3003 		return ret;
3004 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3005 					X86_TRANSFER_TASK_SWITCH, NULL);
3006 	if (ret != X86EMUL_CONTINUE)
3007 		return ret;
3008 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3009 					X86_TRANSFER_TASK_SWITCH, NULL);
3010 	if (ret != X86EMUL_CONTINUE)
3011 		return ret;
3012 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3013 					X86_TRANSFER_TASK_SWITCH, NULL);
3014 	if (ret != X86EMUL_CONTINUE)
3015 		return ret;
3016 
3017 	return X86EMUL_CONTINUE;
3018 }
3019 
3020 static int task_switch_16(struct x86_emulate_ctxt *ctxt,
3021 			  u16 tss_selector, u16 old_tss_sel,
3022 			  ulong old_tss_base, struct desc_struct *new_desc)
3023 {
3024 	const struct x86_emulate_ops *ops = ctxt->ops;
3025 	struct tss_segment_16 tss_seg;
3026 	int ret;
3027 	u32 new_tss_base = get_desc_base(new_desc);
3028 
3029 	ret = ops->read_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3030 			    &ctxt->exception);
3031 	if (ret != X86EMUL_CONTINUE)
3032 		return ret;
3033 
3034 	save_state_to_tss16(ctxt, &tss_seg);
3035 
3036 	ret = ops->write_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3037 			     &ctxt->exception);
3038 	if (ret != X86EMUL_CONTINUE)
3039 		return ret;
3040 
3041 	ret = ops->read_std(ctxt, new_tss_base, &tss_seg, sizeof tss_seg,
3042 			    &ctxt->exception);
3043 	if (ret != X86EMUL_CONTINUE)
3044 		return ret;
3045 
3046 	if (old_tss_sel != 0xffff) {
3047 		tss_seg.prev_task_link = old_tss_sel;
3048 
3049 		ret = ops->write_std(ctxt, new_tss_base,
3050 				     &tss_seg.prev_task_link,
3051 				     sizeof tss_seg.prev_task_link,
3052 				     &ctxt->exception);
3053 		if (ret != X86EMUL_CONTINUE)
3054 			return ret;
3055 	}
3056 
3057 	return load_state_from_tss16(ctxt, &tss_seg);
3058 }
3059 
3060 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3061 				struct tss_segment_32 *tss)
3062 {
3063 	/* CR3 and ldt selector are not saved intentionally */
3064 	tss->eip = ctxt->_eip;
3065 	tss->eflags = ctxt->eflags;
3066 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3067 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3068 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3069 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3070 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3071 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3072 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3073 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3074 
3075 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3076 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3077 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3078 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3079 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3080 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3081 }
3082 
3083 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3084 				 struct tss_segment_32 *tss)
3085 {
3086 	int ret;
3087 	u8 cpl;
3088 
3089 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3090 		return emulate_gp(ctxt, 0);
3091 	ctxt->_eip = tss->eip;
3092 	ctxt->eflags = tss->eflags | 2;
3093 
3094 	/* General purpose registers */
3095 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3096 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3097 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3098 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3099 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3100 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3101 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3102 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3103 
3104 	/*
3105 	 * SDM says that segment selectors are loaded before segment
3106 	 * descriptors.  This is important because CPL checks will
3107 	 * use CS.RPL.
3108 	 */
3109 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3110 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3111 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3112 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3113 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3114 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3115 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3116 
3117 	/*
3118 	 * If we're switching between Protected Mode and VM86, we need to make
3119 	 * sure to update the mode before loading the segment descriptors so
3120 	 * that the selectors are interpreted correctly.
3121 	 */
3122 	if (ctxt->eflags & X86_EFLAGS_VM) {
3123 		ctxt->mode = X86EMUL_MODE_VM86;
3124 		cpl = 3;
3125 	} else {
3126 		ctxt->mode = X86EMUL_MODE_PROT32;
3127 		cpl = tss->cs & 3;
3128 	}
3129 
3130 	/*
3131 	 * Now load segment descriptors. If fault happenes at this stage
3132 	 * it is handled in a context of new task
3133 	 */
3134 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3135 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3136 	if (ret != X86EMUL_CONTINUE)
3137 		return ret;
3138 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3139 					X86_TRANSFER_TASK_SWITCH, NULL);
3140 	if (ret != X86EMUL_CONTINUE)
3141 		return ret;
3142 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3143 					X86_TRANSFER_TASK_SWITCH, NULL);
3144 	if (ret != X86EMUL_CONTINUE)
3145 		return ret;
3146 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3147 					X86_TRANSFER_TASK_SWITCH, NULL);
3148 	if (ret != X86EMUL_CONTINUE)
3149 		return ret;
3150 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3151 					X86_TRANSFER_TASK_SWITCH, NULL);
3152 	if (ret != X86EMUL_CONTINUE)
3153 		return ret;
3154 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3155 					X86_TRANSFER_TASK_SWITCH, NULL);
3156 	if (ret != X86EMUL_CONTINUE)
3157 		return ret;
3158 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3159 					X86_TRANSFER_TASK_SWITCH, NULL);
3160 
3161 	return ret;
3162 }
3163 
3164 static int task_switch_32(struct x86_emulate_ctxt *ctxt,
3165 			  u16 tss_selector, u16 old_tss_sel,
3166 			  ulong old_tss_base, struct desc_struct *new_desc)
3167 {
3168 	const struct x86_emulate_ops *ops = ctxt->ops;
3169 	struct tss_segment_32 tss_seg;
3170 	int ret;
3171 	u32 new_tss_base = get_desc_base(new_desc);
3172 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3173 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3174 
3175 	ret = ops->read_std(ctxt, old_tss_base, &tss_seg, sizeof tss_seg,
3176 			    &ctxt->exception);
3177 	if (ret != X86EMUL_CONTINUE)
3178 		return ret;
3179 
3180 	save_state_to_tss32(ctxt, &tss_seg);
3181 
3182 	/* Only GP registers and segment selectors are saved */
3183 	ret = ops->write_std(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3184 			     ldt_sel_offset - eip_offset, &ctxt->exception);
3185 	if (ret != X86EMUL_CONTINUE)
3186 		return ret;
3187 
3188 	ret = ops->read_std(ctxt, new_tss_base, &tss_seg, sizeof tss_seg,
3189 			    &ctxt->exception);
3190 	if (ret != X86EMUL_CONTINUE)
3191 		return ret;
3192 
3193 	if (old_tss_sel != 0xffff) {
3194 		tss_seg.prev_task_link = old_tss_sel;
3195 
3196 		ret = ops->write_std(ctxt, new_tss_base,
3197 				     &tss_seg.prev_task_link,
3198 				     sizeof tss_seg.prev_task_link,
3199 				     &ctxt->exception);
3200 		if (ret != X86EMUL_CONTINUE)
3201 			return ret;
3202 	}
3203 
3204 	return load_state_from_tss32(ctxt, &tss_seg);
3205 }
3206 
3207 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3208 				   u16 tss_selector, int idt_index, int reason,
3209 				   bool has_error_code, u32 error_code)
3210 {
3211 	const struct x86_emulate_ops *ops = ctxt->ops;
3212 	struct desc_struct curr_tss_desc, next_tss_desc;
3213 	int ret;
3214 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3215 	ulong old_tss_base =
3216 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3217 	u32 desc_limit;
3218 	ulong desc_addr, dr7;
3219 
3220 	/* FIXME: old_tss_base == ~0 ? */
3221 
3222 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3223 	if (ret != X86EMUL_CONTINUE)
3224 		return ret;
3225 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3226 	if (ret != X86EMUL_CONTINUE)
3227 		return ret;
3228 
3229 	/* FIXME: check that next_tss_desc is tss */
3230 
3231 	/*
3232 	 * Check privileges. The three cases are task switch caused by...
3233 	 *
3234 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3235 	 * 2. Exception/IRQ/iret: No check is performed
3236 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3237 	 *    hardware checks it before exiting.
3238 	 */
3239 	if (reason == TASK_SWITCH_GATE) {
3240 		if (idt_index != -1) {
3241 			/* Software interrupts */
3242 			struct desc_struct task_gate_desc;
3243 			int dpl;
3244 
3245 			ret = read_interrupt_descriptor(ctxt, idt_index,
3246 							&task_gate_desc);
3247 			if (ret != X86EMUL_CONTINUE)
3248 				return ret;
3249 
3250 			dpl = task_gate_desc.dpl;
3251 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3252 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3253 		}
3254 	}
3255 
3256 	desc_limit = desc_limit_scaled(&next_tss_desc);
3257 	if (!next_tss_desc.p ||
3258 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3259 	     desc_limit < 0x2b)) {
3260 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3261 	}
3262 
3263 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3264 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3265 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3266 	}
3267 
3268 	if (reason == TASK_SWITCH_IRET)
3269 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3270 
3271 	/* set back link to prev task only if NT bit is set in eflags
3272 	   note that old_tss_sel is not used after this point */
3273 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3274 		old_tss_sel = 0xffff;
3275 
3276 	if (next_tss_desc.type & 8)
3277 		ret = task_switch_32(ctxt, tss_selector, old_tss_sel,
3278 				     old_tss_base, &next_tss_desc);
3279 	else
3280 		ret = task_switch_16(ctxt, tss_selector, old_tss_sel,
3281 				     old_tss_base, &next_tss_desc);
3282 	if (ret != X86EMUL_CONTINUE)
3283 		return ret;
3284 
3285 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3286 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3287 
3288 	if (reason != TASK_SWITCH_IRET) {
3289 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3290 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3291 	}
3292 
3293 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3294 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3295 
3296 	if (has_error_code) {
3297 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3298 		ctxt->lock_prefix = 0;
3299 		ctxt->src.val = (unsigned long) error_code;
3300 		ret = em_push(ctxt);
3301 	}
3302 
3303 	ops->get_dr(ctxt, 7, &dr7);
3304 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3305 
3306 	return ret;
3307 }
3308 
3309 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3310 			 u16 tss_selector, int idt_index, int reason,
3311 			 bool has_error_code, u32 error_code)
3312 {
3313 	int rc;
3314 
3315 	invalidate_registers(ctxt);
3316 	ctxt->_eip = ctxt->eip;
3317 	ctxt->dst.type = OP_NONE;
3318 
3319 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3320 				     has_error_code, error_code);
3321 
3322 	if (rc == X86EMUL_CONTINUE) {
3323 		ctxt->eip = ctxt->_eip;
3324 		writeback_registers(ctxt);
3325 	}
3326 
3327 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3328 }
3329 
3330 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3331 		struct operand *op)
3332 {
3333 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3334 
3335 	register_address_increment(ctxt, reg, df * op->bytes);
3336 	op->addr.mem.ea = register_address(ctxt, reg);
3337 }
3338 
3339 static int em_das(struct x86_emulate_ctxt *ctxt)
3340 {
3341 	u8 al, old_al;
3342 	bool af, cf, old_cf;
3343 
3344 	cf = ctxt->eflags & X86_EFLAGS_CF;
3345 	al = ctxt->dst.val;
3346 
3347 	old_al = al;
3348 	old_cf = cf;
3349 	cf = false;
3350 	af = ctxt->eflags & X86_EFLAGS_AF;
3351 	if ((al & 0x0f) > 9 || af) {
3352 		al -= 6;
3353 		cf = old_cf | (al >= 250);
3354 		af = true;
3355 	} else {
3356 		af = false;
3357 	}
3358 	if (old_al > 0x99 || old_cf) {
3359 		al -= 0x60;
3360 		cf = true;
3361 	}
3362 
3363 	ctxt->dst.val = al;
3364 	/* Set PF, ZF, SF */
3365 	ctxt->src.type = OP_IMM;
3366 	ctxt->src.val = 0;
3367 	ctxt->src.bytes = 1;
3368 	fastop(ctxt, em_or);
3369 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3370 	if (cf)
3371 		ctxt->eflags |= X86_EFLAGS_CF;
3372 	if (af)
3373 		ctxt->eflags |= X86_EFLAGS_AF;
3374 	return X86EMUL_CONTINUE;
3375 }
3376 
3377 static int em_aam(struct x86_emulate_ctxt *ctxt)
3378 {
3379 	u8 al, ah;
3380 
3381 	if (ctxt->src.val == 0)
3382 		return emulate_de(ctxt);
3383 
3384 	al = ctxt->dst.val & 0xff;
3385 	ah = al / ctxt->src.val;
3386 	al %= ctxt->src.val;
3387 
3388 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3389 
3390 	/* Set PF, ZF, SF */
3391 	ctxt->src.type = OP_IMM;
3392 	ctxt->src.val = 0;
3393 	ctxt->src.bytes = 1;
3394 	fastop(ctxt, em_or);
3395 
3396 	return X86EMUL_CONTINUE;
3397 }
3398 
3399 static int em_aad(struct x86_emulate_ctxt *ctxt)
3400 {
3401 	u8 al = ctxt->dst.val & 0xff;
3402 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3403 
3404 	al = (al + (ah * ctxt->src.val)) & 0xff;
3405 
3406 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3407 
3408 	/* Set PF, ZF, SF */
3409 	ctxt->src.type = OP_IMM;
3410 	ctxt->src.val = 0;
3411 	ctxt->src.bytes = 1;
3412 	fastop(ctxt, em_or);
3413 
3414 	return X86EMUL_CONTINUE;
3415 }
3416 
3417 static int em_call(struct x86_emulate_ctxt *ctxt)
3418 {
3419 	int rc;
3420 	long rel = ctxt->src.val;
3421 
3422 	ctxt->src.val = (unsigned long)ctxt->_eip;
3423 	rc = jmp_rel(ctxt, rel);
3424 	if (rc != X86EMUL_CONTINUE)
3425 		return rc;
3426 	return em_push(ctxt);
3427 }
3428 
3429 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3430 {
3431 	u16 sel, old_cs;
3432 	ulong old_eip;
3433 	int rc;
3434 	struct desc_struct old_desc, new_desc;
3435 	const struct x86_emulate_ops *ops = ctxt->ops;
3436 	int cpl = ctxt->ops->cpl(ctxt);
3437 	enum x86emul_mode prev_mode = ctxt->mode;
3438 
3439 	old_eip = ctxt->_eip;
3440 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3441 
3442 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3443 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3444 				       X86_TRANSFER_CALL_JMP, &new_desc);
3445 	if (rc != X86EMUL_CONTINUE)
3446 		return rc;
3447 
3448 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3449 	if (rc != X86EMUL_CONTINUE)
3450 		goto fail;
3451 
3452 	ctxt->src.val = old_cs;
3453 	rc = em_push(ctxt);
3454 	if (rc != X86EMUL_CONTINUE)
3455 		goto fail;
3456 
3457 	ctxt->src.val = old_eip;
3458 	rc = em_push(ctxt);
3459 	/* If we failed, we tainted the memory, but the very least we should
3460 	   restore cs */
3461 	if (rc != X86EMUL_CONTINUE) {
3462 		pr_warn_once("faulting far call emulation tainted memory\n");
3463 		goto fail;
3464 	}
3465 	return rc;
3466 fail:
3467 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3468 	ctxt->mode = prev_mode;
3469 	return rc;
3470 
3471 }
3472 
3473 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3474 {
3475 	int rc;
3476 	unsigned long eip;
3477 
3478 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3479 	if (rc != X86EMUL_CONTINUE)
3480 		return rc;
3481 	rc = assign_eip_near(ctxt, eip);
3482 	if (rc != X86EMUL_CONTINUE)
3483 		return rc;
3484 	rsp_increment(ctxt, ctxt->src.val);
3485 	return X86EMUL_CONTINUE;
3486 }
3487 
3488 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3489 {
3490 	/* Write back the register source. */
3491 	ctxt->src.val = ctxt->dst.val;
3492 	write_register_operand(&ctxt->src);
3493 
3494 	/* Write back the memory destination with implicit LOCK prefix. */
3495 	ctxt->dst.val = ctxt->src.orig_val;
3496 	ctxt->lock_prefix = 1;
3497 	return X86EMUL_CONTINUE;
3498 }
3499 
3500 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3501 {
3502 	ctxt->dst.val = ctxt->src2.val;
3503 	return fastop(ctxt, em_imul);
3504 }
3505 
3506 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3507 {
3508 	ctxt->dst.type = OP_REG;
3509 	ctxt->dst.bytes = ctxt->src.bytes;
3510 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3511 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3512 
3513 	return X86EMUL_CONTINUE;
3514 }
3515 
3516 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3517 {
3518 	u64 tsc = 0;
3519 
3520 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3521 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3522 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3523 	return X86EMUL_CONTINUE;
3524 }
3525 
3526 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3527 {
3528 	u64 pmc;
3529 
3530 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3531 		return emulate_gp(ctxt, 0);
3532 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3533 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3534 	return X86EMUL_CONTINUE;
3535 }
3536 
3537 static int em_mov(struct x86_emulate_ctxt *ctxt)
3538 {
3539 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3540 	return X86EMUL_CONTINUE;
3541 }
3542 
3543 #define FFL(x) bit(X86_FEATURE_##x)
3544 
3545 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3546 {
3547 	u32 ebx, ecx, edx, eax = 1;
3548 	u16 tmp;
3549 
3550 	/*
3551 	 * Check MOVBE is set in the guest-visible CPUID leaf.
3552 	 */
3553 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
3554 	if (!(ecx & FFL(MOVBE)))
3555 		return emulate_ud(ctxt);
3556 
3557 	switch (ctxt->op_bytes) {
3558 	case 2:
3559 		/*
3560 		 * From MOVBE definition: "...When the operand size is 16 bits,
3561 		 * the upper word of the destination register remains unchanged
3562 		 * ..."
3563 		 *
3564 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3565 		 * rules so we have to do the operation almost per hand.
3566 		 */
3567 		tmp = (u16)ctxt->src.val;
3568 		ctxt->dst.val &= ~0xffffUL;
3569 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3570 		break;
3571 	case 4:
3572 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3573 		break;
3574 	case 8:
3575 		ctxt->dst.val = swab64(ctxt->src.val);
3576 		break;
3577 	default:
3578 		BUG();
3579 	}
3580 	return X86EMUL_CONTINUE;
3581 }
3582 
3583 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3584 {
3585 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3586 		return emulate_gp(ctxt, 0);
3587 
3588 	/* Disable writeback. */
3589 	ctxt->dst.type = OP_NONE;
3590 	return X86EMUL_CONTINUE;
3591 }
3592 
3593 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3594 {
3595 	unsigned long val;
3596 
3597 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3598 		val = ctxt->src.val & ~0ULL;
3599 	else
3600 		val = ctxt->src.val & ~0U;
3601 
3602 	/* #UD condition is already handled. */
3603 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3604 		return emulate_gp(ctxt, 0);
3605 
3606 	/* Disable writeback. */
3607 	ctxt->dst.type = OP_NONE;
3608 	return X86EMUL_CONTINUE;
3609 }
3610 
3611 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3612 {
3613 	u64 msr_data;
3614 
3615 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3616 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3617 	if (ctxt->ops->set_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), msr_data))
3618 		return emulate_gp(ctxt, 0);
3619 
3620 	return X86EMUL_CONTINUE;
3621 }
3622 
3623 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3624 {
3625 	u64 msr_data;
3626 
3627 	if (ctxt->ops->get_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &msr_data))
3628 		return emulate_gp(ctxt, 0);
3629 
3630 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3631 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3632 	return X86EMUL_CONTINUE;
3633 }
3634 
3635 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3636 {
3637 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3638 		return emulate_ud(ctxt);
3639 
3640 	ctxt->dst.val = get_segment_selector(ctxt, ctxt->modrm_reg);
3641 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3642 		ctxt->dst.bytes = 2;
3643 	return X86EMUL_CONTINUE;
3644 }
3645 
3646 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3647 {
3648 	u16 sel = ctxt->src.val;
3649 
3650 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3651 		return emulate_ud(ctxt);
3652 
3653 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3654 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3655 
3656 	/* Disable writeback. */
3657 	ctxt->dst.type = OP_NONE;
3658 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3659 }
3660 
3661 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3662 {
3663 	u16 sel = ctxt->src.val;
3664 
3665 	/* Disable writeback. */
3666 	ctxt->dst.type = OP_NONE;
3667 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3668 }
3669 
3670 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3671 {
3672 	u16 sel = ctxt->src.val;
3673 
3674 	/* Disable writeback. */
3675 	ctxt->dst.type = OP_NONE;
3676 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3677 }
3678 
3679 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3680 {
3681 	int rc;
3682 	ulong linear;
3683 
3684 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3685 	if (rc == X86EMUL_CONTINUE)
3686 		ctxt->ops->invlpg(ctxt, linear);
3687 	/* Disable writeback. */
3688 	ctxt->dst.type = OP_NONE;
3689 	return X86EMUL_CONTINUE;
3690 }
3691 
3692 static int em_clts(struct x86_emulate_ctxt *ctxt)
3693 {
3694 	ulong cr0;
3695 
3696 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3697 	cr0 &= ~X86_CR0_TS;
3698 	ctxt->ops->set_cr(ctxt, 0, cr0);
3699 	return X86EMUL_CONTINUE;
3700 }
3701 
3702 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3703 {
3704 	int rc = ctxt->ops->fix_hypercall(ctxt);
3705 
3706 	if (rc != X86EMUL_CONTINUE)
3707 		return rc;
3708 
3709 	/* Let the processor re-execute the fixed hypercall */
3710 	ctxt->_eip = ctxt->eip;
3711 	/* Disable writeback. */
3712 	ctxt->dst.type = OP_NONE;
3713 	return X86EMUL_CONTINUE;
3714 }
3715 
3716 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3717 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3718 					      struct desc_ptr *ptr))
3719 {
3720 	struct desc_ptr desc_ptr;
3721 
3722 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3723 		ctxt->op_bytes = 8;
3724 	get(ctxt, &desc_ptr);
3725 	if (ctxt->op_bytes == 2) {
3726 		ctxt->op_bytes = 4;
3727 		desc_ptr.address &= 0x00ffffff;
3728 	}
3729 	/* Disable writeback. */
3730 	ctxt->dst.type = OP_NONE;
3731 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3732 				   &desc_ptr, 2 + ctxt->op_bytes);
3733 }
3734 
3735 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3736 {
3737 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3738 }
3739 
3740 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3741 {
3742 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3743 }
3744 
3745 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3746 {
3747 	struct desc_ptr desc_ptr;
3748 	int rc;
3749 
3750 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3751 		ctxt->op_bytes = 8;
3752 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3753 			     &desc_ptr.size, &desc_ptr.address,
3754 			     ctxt->op_bytes);
3755 	if (rc != X86EMUL_CONTINUE)
3756 		return rc;
3757 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3758 	    is_noncanonical_address(desc_ptr.address))
3759 		return emulate_gp(ctxt, 0);
3760 	if (lgdt)
3761 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3762 	else
3763 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3764 	/* Disable writeback. */
3765 	ctxt->dst.type = OP_NONE;
3766 	return X86EMUL_CONTINUE;
3767 }
3768 
3769 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3770 {
3771 	return em_lgdt_lidt(ctxt, true);
3772 }
3773 
3774 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3775 {
3776 	return em_lgdt_lidt(ctxt, false);
3777 }
3778 
3779 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3780 {
3781 	if (ctxt->dst.type == OP_MEM)
3782 		ctxt->dst.bytes = 2;
3783 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3784 	return X86EMUL_CONTINUE;
3785 }
3786 
3787 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3788 {
3789 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3790 			  | (ctxt->src.val & 0x0f));
3791 	ctxt->dst.type = OP_NONE;
3792 	return X86EMUL_CONTINUE;
3793 }
3794 
3795 static int em_loop(struct x86_emulate_ctxt *ctxt)
3796 {
3797 	int rc = X86EMUL_CONTINUE;
3798 
3799 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3800 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3801 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3802 		rc = jmp_rel(ctxt, ctxt->src.val);
3803 
3804 	return rc;
3805 }
3806 
3807 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3808 {
3809 	int rc = X86EMUL_CONTINUE;
3810 
3811 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3812 		rc = jmp_rel(ctxt, ctxt->src.val);
3813 
3814 	return rc;
3815 }
3816 
3817 static int em_in(struct x86_emulate_ctxt *ctxt)
3818 {
3819 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3820 			     &ctxt->dst.val))
3821 		return X86EMUL_IO_NEEDED;
3822 
3823 	return X86EMUL_CONTINUE;
3824 }
3825 
3826 static int em_out(struct x86_emulate_ctxt *ctxt)
3827 {
3828 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3829 				    &ctxt->src.val, 1);
3830 	/* Disable writeback. */
3831 	ctxt->dst.type = OP_NONE;
3832 	return X86EMUL_CONTINUE;
3833 }
3834 
3835 static int em_cli(struct x86_emulate_ctxt *ctxt)
3836 {
3837 	if (emulator_bad_iopl(ctxt))
3838 		return emulate_gp(ctxt, 0);
3839 
3840 	ctxt->eflags &= ~X86_EFLAGS_IF;
3841 	return X86EMUL_CONTINUE;
3842 }
3843 
3844 static int em_sti(struct x86_emulate_ctxt *ctxt)
3845 {
3846 	if (emulator_bad_iopl(ctxt))
3847 		return emulate_gp(ctxt, 0);
3848 
3849 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3850 	ctxt->eflags |= X86_EFLAGS_IF;
3851 	return X86EMUL_CONTINUE;
3852 }
3853 
3854 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3855 {
3856 	u32 eax, ebx, ecx, edx;
3857 	u64 msr = 0;
3858 
3859 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3860 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3861 	    ctxt->ops->cpl(ctxt)) {
3862 		return emulate_gp(ctxt, 0);
3863 	}
3864 
3865 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3866 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3867 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
3868 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3869 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3870 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3871 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3872 	return X86EMUL_CONTINUE;
3873 }
3874 
3875 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3876 {
3877 	u32 flags;
3878 
3879 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3880 		X86_EFLAGS_SF;
3881 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3882 
3883 	ctxt->eflags &= ~0xffUL;
3884 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3885 	return X86EMUL_CONTINUE;
3886 }
3887 
3888 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3889 {
3890 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3891 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3892 	return X86EMUL_CONTINUE;
3893 }
3894 
3895 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3896 {
3897 	switch (ctxt->op_bytes) {
3898 #ifdef CONFIG_X86_64
3899 	case 8:
3900 		asm("bswap %0" : "+r"(ctxt->dst.val));
3901 		break;
3902 #endif
3903 	default:
3904 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3905 		break;
3906 	}
3907 	return X86EMUL_CONTINUE;
3908 }
3909 
3910 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3911 {
3912 	/* emulating clflush regardless of cpuid */
3913 	return X86EMUL_CONTINUE;
3914 }
3915 
3916 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3917 {
3918 	ctxt->dst.val = (s32) ctxt->src.val;
3919 	return X86EMUL_CONTINUE;
3920 }
3921 
3922 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3923 {
3924 	u32 eax = 1, ebx, ecx = 0, edx;
3925 
3926 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx);
3927 	if (!(edx & FFL(FXSR)))
3928 		return emulate_ud(ctxt);
3929 
3930 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3931 		return emulate_nm(ctxt);
3932 
3933 	/*
3934 	 * Don't emulate a case that should never be hit, instead of working
3935 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3936 	 */
3937 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3938 		return X86EMUL_UNHANDLEABLE;
3939 
3940 	return X86EMUL_CONTINUE;
3941 }
3942 
3943 /*
3944  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
3945  *  1) 16 bit mode
3946  *  2) 32 bit mode
3947  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
3948  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
3949  *       save and restore
3950  *  3) 64-bit mode with REX.W prefix
3951  *     - like (2), but XMM 8-15 are being saved and restored
3952  *  4) 64-bit mode without REX.W prefix
3953  *     - like (3), but FIP and FDP are 64 bit
3954  *
3955  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
3956  * desired result.  (4) is not emulated.
3957  *
3958  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
3959  * and FPU DS) should match.
3960  */
3961 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
3962 {
3963 	struct fxregs_state fx_state;
3964 	size_t size;
3965 	int rc;
3966 
3967 	rc = check_fxsr(ctxt);
3968 	if (rc != X86EMUL_CONTINUE)
3969 		return rc;
3970 
3971 	ctxt->ops->get_fpu(ctxt);
3972 
3973 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
3974 
3975 	ctxt->ops->put_fpu(ctxt);
3976 
3977 	if (rc != X86EMUL_CONTINUE)
3978 		return rc;
3979 
3980 	if (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR)
3981 		size = offsetof(struct fxregs_state, xmm_space[8 * 16/4]);
3982 	else
3983 		size = offsetof(struct fxregs_state, xmm_space[0]);
3984 
3985 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
3986 }
3987 
3988 static int fxrstor_fixup(struct x86_emulate_ctxt *ctxt,
3989 		struct fxregs_state *new)
3990 {
3991 	int rc = X86EMUL_CONTINUE;
3992 	struct fxregs_state old;
3993 
3994 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(old));
3995 	if (rc != X86EMUL_CONTINUE)
3996 		return rc;
3997 
3998 	/*
3999 	 * 64 bit host will restore XMM 8-15, which is not correct on non-64
4000 	 * bit guests.  Load the current values in order to preserve 64 bit
4001 	 * XMMs after fxrstor.
4002 	 */
4003 #ifdef CONFIG_X86_64
4004 	/* XXX: accessing XMM 8-15 very awkwardly */
4005 	memcpy(&new->xmm_space[8 * 16/4], &old.xmm_space[8 * 16/4], 8 * 16);
4006 #endif
4007 
4008 	/*
4009 	 * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but
4010 	 * does save and restore MXCSR.
4011 	 */
4012 	if (!(ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))
4013 		memcpy(new->xmm_space, old.xmm_space, 8 * 16);
4014 
4015 	return rc;
4016 }
4017 
4018 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4019 {
4020 	struct fxregs_state fx_state;
4021 	int rc;
4022 
4023 	rc = check_fxsr(ctxt);
4024 	if (rc != X86EMUL_CONTINUE)
4025 		return rc;
4026 
4027 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, 512);
4028 	if (rc != X86EMUL_CONTINUE)
4029 		return rc;
4030 
4031 	if (fx_state.mxcsr >> 16)
4032 		return emulate_gp(ctxt, 0);
4033 
4034 	ctxt->ops->get_fpu(ctxt);
4035 
4036 	if (ctxt->mode < X86EMUL_MODE_PROT64)
4037 		rc = fxrstor_fixup(ctxt, &fx_state);
4038 
4039 	if (rc == X86EMUL_CONTINUE)
4040 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4041 
4042 	ctxt->ops->put_fpu(ctxt);
4043 
4044 	return rc;
4045 }
4046 
4047 static bool valid_cr(int nr)
4048 {
4049 	switch (nr) {
4050 	case 0:
4051 	case 2 ... 4:
4052 	case 8:
4053 		return true;
4054 	default:
4055 		return false;
4056 	}
4057 }
4058 
4059 static int check_cr_read(struct x86_emulate_ctxt *ctxt)
4060 {
4061 	if (!valid_cr(ctxt->modrm_reg))
4062 		return emulate_ud(ctxt);
4063 
4064 	return X86EMUL_CONTINUE;
4065 }
4066 
4067 static int check_cr_write(struct x86_emulate_ctxt *ctxt)
4068 {
4069 	u64 new_val = ctxt->src.val64;
4070 	int cr = ctxt->modrm_reg;
4071 	u64 efer = 0;
4072 
4073 	static u64 cr_reserved_bits[] = {
4074 		0xffffffff00000000ULL,
4075 		0, 0, 0, /* CR3 checked later */
4076 		CR4_RESERVED_BITS,
4077 		0, 0, 0,
4078 		CR8_RESERVED_BITS,
4079 	};
4080 
4081 	if (!valid_cr(cr))
4082 		return emulate_ud(ctxt);
4083 
4084 	if (new_val & cr_reserved_bits[cr])
4085 		return emulate_gp(ctxt, 0);
4086 
4087 	switch (cr) {
4088 	case 0: {
4089 		u64 cr4;
4090 		if (((new_val & X86_CR0_PG) && !(new_val & X86_CR0_PE)) ||
4091 		    ((new_val & X86_CR0_NW) && !(new_val & X86_CR0_CD)))
4092 			return emulate_gp(ctxt, 0);
4093 
4094 		cr4 = ctxt->ops->get_cr(ctxt, 4);
4095 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4096 
4097 		if ((new_val & X86_CR0_PG) && (efer & EFER_LME) &&
4098 		    !(cr4 & X86_CR4_PAE))
4099 			return emulate_gp(ctxt, 0);
4100 
4101 		break;
4102 		}
4103 	case 3: {
4104 		u64 rsvd = 0;
4105 
4106 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4107 		if (efer & EFER_LMA)
4108 			rsvd = CR3_L_MODE_RESERVED_BITS & ~CR3_PCID_INVD;
4109 
4110 		if (new_val & rsvd)
4111 			return emulate_gp(ctxt, 0);
4112 
4113 		break;
4114 		}
4115 	case 4: {
4116 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4117 
4118 		if ((efer & EFER_LMA) && !(new_val & X86_CR4_PAE))
4119 			return emulate_gp(ctxt, 0);
4120 
4121 		break;
4122 		}
4123 	}
4124 
4125 	return X86EMUL_CONTINUE;
4126 }
4127 
4128 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4129 {
4130 	unsigned long dr7;
4131 
4132 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4133 
4134 	/* Check if DR7.Global_Enable is set */
4135 	return dr7 & (1 << 13);
4136 }
4137 
4138 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4139 {
4140 	int dr = ctxt->modrm_reg;
4141 	u64 cr4;
4142 
4143 	if (dr > 7)
4144 		return emulate_ud(ctxt);
4145 
4146 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4147 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4148 		return emulate_ud(ctxt);
4149 
4150 	if (check_dr7_gd(ctxt)) {
4151 		ulong dr6;
4152 
4153 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4154 		dr6 &= ~15;
4155 		dr6 |= DR6_BD | DR6_RTM;
4156 		ctxt->ops->set_dr(ctxt, 6, dr6);
4157 		return emulate_db(ctxt);
4158 	}
4159 
4160 	return X86EMUL_CONTINUE;
4161 }
4162 
4163 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4164 {
4165 	u64 new_val = ctxt->src.val64;
4166 	int dr = ctxt->modrm_reg;
4167 
4168 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4169 		return emulate_gp(ctxt, 0);
4170 
4171 	return check_dr_read(ctxt);
4172 }
4173 
4174 static int check_svme(struct x86_emulate_ctxt *ctxt)
4175 {
4176 	u64 efer = 0;
4177 
4178 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4179 
4180 	if (!(efer & EFER_SVME))
4181 		return emulate_ud(ctxt);
4182 
4183 	return X86EMUL_CONTINUE;
4184 }
4185 
4186 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4187 {
4188 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4189 
4190 	/* Valid physical address? */
4191 	if (rax & 0xffff000000000000ULL)
4192 		return emulate_gp(ctxt, 0);
4193 
4194 	return check_svme(ctxt);
4195 }
4196 
4197 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4198 {
4199 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4200 
4201 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4202 		return emulate_ud(ctxt);
4203 
4204 	return X86EMUL_CONTINUE;
4205 }
4206 
4207 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4208 {
4209 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4210 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4211 
4212 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4213 	    ctxt->ops->check_pmc(ctxt, rcx))
4214 		return emulate_gp(ctxt, 0);
4215 
4216 	return X86EMUL_CONTINUE;
4217 }
4218 
4219 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4220 {
4221 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4222 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4223 		return emulate_gp(ctxt, 0);
4224 
4225 	return X86EMUL_CONTINUE;
4226 }
4227 
4228 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4229 {
4230 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4231 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4232 		return emulate_gp(ctxt, 0);
4233 
4234 	return X86EMUL_CONTINUE;
4235 }
4236 
4237 #define D(_y) { .flags = (_y) }
4238 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4239 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4240 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4241 #define N    D(NotImpl)
4242 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4243 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4244 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4245 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4246 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4247 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4248 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4249 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4250 #define II(_f, _e, _i) \
4251 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4252 #define IIP(_f, _e, _i, _p) \
4253 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4254 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4255 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4256 
4257 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4258 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4259 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4260 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4261 #define I2bvIP(_f, _e, _i, _p) \
4262 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4263 
4264 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4265 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4266 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4267 
4268 static const struct opcode group7_rm0[] = {
4269 	N,
4270 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4271 	N, N, N, N, N, N,
4272 };
4273 
4274 static const struct opcode group7_rm1[] = {
4275 	DI(SrcNone | Priv, monitor),
4276 	DI(SrcNone | Priv, mwait),
4277 	N, N, N, N, N, N,
4278 };
4279 
4280 static const struct opcode group7_rm3[] = {
4281 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4282 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4283 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4284 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4285 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4286 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4287 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4288 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4289 };
4290 
4291 static const struct opcode group7_rm7[] = {
4292 	N,
4293 	DIP(SrcNone, rdtscp, check_rdtsc),
4294 	N, N, N, N, N, N,
4295 };
4296 
4297 static const struct opcode group1[] = {
4298 	F(Lock, em_add),
4299 	F(Lock | PageTable, em_or),
4300 	F(Lock, em_adc),
4301 	F(Lock, em_sbb),
4302 	F(Lock | PageTable, em_and),
4303 	F(Lock, em_sub),
4304 	F(Lock, em_xor),
4305 	F(NoWrite, em_cmp),
4306 };
4307 
4308 static const struct opcode group1A[] = {
4309 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4310 };
4311 
4312 static const struct opcode group2[] = {
4313 	F(DstMem | ModRM, em_rol),
4314 	F(DstMem | ModRM, em_ror),
4315 	F(DstMem | ModRM, em_rcl),
4316 	F(DstMem | ModRM, em_rcr),
4317 	F(DstMem | ModRM, em_shl),
4318 	F(DstMem | ModRM, em_shr),
4319 	F(DstMem | ModRM, em_shl),
4320 	F(DstMem | ModRM, em_sar),
4321 };
4322 
4323 static const struct opcode group3[] = {
4324 	F(DstMem | SrcImm | NoWrite, em_test),
4325 	F(DstMem | SrcImm | NoWrite, em_test),
4326 	F(DstMem | SrcNone | Lock, em_not),
4327 	F(DstMem | SrcNone | Lock, em_neg),
4328 	F(DstXacc | Src2Mem, em_mul_ex),
4329 	F(DstXacc | Src2Mem, em_imul_ex),
4330 	F(DstXacc | Src2Mem, em_div_ex),
4331 	F(DstXacc | Src2Mem, em_idiv_ex),
4332 };
4333 
4334 static const struct opcode group4[] = {
4335 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4336 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4337 	N, N, N, N, N, N,
4338 };
4339 
4340 static const struct opcode group5[] = {
4341 	F(DstMem | SrcNone | Lock,		em_inc),
4342 	F(DstMem | SrcNone | Lock,		em_dec),
4343 	I(SrcMem | NearBranch,			em_call_near_abs),
4344 	I(SrcMemFAddr | ImplicitOps,		em_call_far),
4345 	I(SrcMem | NearBranch,			em_jmp_abs),
4346 	I(SrcMemFAddr | ImplicitOps,		em_jmp_far),
4347 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4348 };
4349 
4350 static const struct opcode group6[] = {
4351 	DI(Prot | DstMem,	sldt),
4352 	DI(Prot | DstMem,	str),
4353 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4354 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4355 	N, N, N, N,
4356 };
4357 
4358 static const struct group_dual group7 = { {
4359 	II(Mov | DstMem,			em_sgdt, sgdt),
4360 	II(Mov | DstMem,			em_sidt, sidt),
4361 	II(SrcMem | Priv,			em_lgdt, lgdt),
4362 	II(SrcMem | Priv,			em_lidt, lidt),
4363 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4364 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4365 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4366 }, {
4367 	EXT(0, group7_rm0),
4368 	EXT(0, group7_rm1),
4369 	N, EXT(0, group7_rm3),
4370 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4371 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4372 	EXT(0, group7_rm7),
4373 } };
4374 
4375 static const struct opcode group8[] = {
4376 	N, N, N, N,
4377 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4378 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4379 	F(DstMem | SrcImmByte | Lock,			em_btr),
4380 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4381 };
4382 
4383 static const struct group_dual group9 = { {
4384 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4385 }, {
4386 	N, N, N, N, N, N, N, N,
4387 } };
4388 
4389 static const struct opcode group11[] = {
4390 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4391 	X7(D(Undefined)),
4392 };
4393 
4394 static const struct gprefix pfx_0f_ae_7 = {
4395 	I(SrcMem | ByteOp, em_clflush), N, N, N,
4396 };
4397 
4398 static const struct group_dual group15 = { {
4399 	I(ModRM | Aligned16, em_fxsave),
4400 	I(ModRM | Aligned16, em_fxrstor),
4401 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4402 }, {
4403 	N, N, N, N, N, N, N, N,
4404 } };
4405 
4406 static const struct gprefix pfx_0f_6f_0f_7f = {
4407 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4408 };
4409 
4410 static const struct instr_dual instr_dual_0f_2b = {
4411 	I(0, em_mov), N
4412 };
4413 
4414 static const struct gprefix pfx_0f_2b = {
4415 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4416 };
4417 
4418 static const struct gprefix pfx_0f_28_0f_29 = {
4419 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4420 };
4421 
4422 static const struct gprefix pfx_0f_e7 = {
4423 	N, I(Sse, em_mov), N, N,
4424 };
4425 
4426 static const struct escape escape_d9 = { {
4427 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4428 }, {
4429 	/* 0xC0 - 0xC7 */
4430 	N, N, N, N, N, N, N, N,
4431 	/* 0xC8 - 0xCF */
4432 	N, N, N, N, N, N, N, N,
4433 	/* 0xD0 - 0xC7 */
4434 	N, N, N, N, N, N, N, N,
4435 	/* 0xD8 - 0xDF */
4436 	N, N, N, N, N, N, N, N,
4437 	/* 0xE0 - 0xE7 */
4438 	N, N, N, N, N, N, N, N,
4439 	/* 0xE8 - 0xEF */
4440 	N, N, N, N, N, N, N, N,
4441 	/* 0xF0 - 0xF7 */
4442 	N, N, N, N, N, N, N, N,
4443 	/* 0xF8 - 0xFF */
4444 	N, N, N, N, N, N, N, N,
4445 } };
4446 
4447 static const struct escape escape_db = { {
4448 	N, N, N, N, N, N, N, N,
4449 }, {
4450 	/* 0xC0 - 0xC7 */
4451 	N, N, N, N, N, N, N, N,
4452 	/* 0xC8 - 0xCF */
4453 	N, N, N, N, N, N, N, N,
4454 	/* 0xD0 - 0xC7 */
4455 	N, N, N, N, N, N, N, N,
4456 	/* 0xD8 - 0xDF */
4457 	N, N, N, N, N, N, N, N,
4458 	/* 0xE0 - 0xE7 */
4459 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4460 	/* 0xE8 - 0xEF */
4461 	N, N, N, N, N, N, N, N,
4462 	/* 0xF0 - 0xF7 */
4463 	N, N, N, N, N, N, N, N,
4464 	/* 0xF8 - 0xFF */
4465 	N, N, N, N, N, N, N, N,
4466 } };
4467 
4468 static const struct escape escape_dd = { {
4469 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4470 }, {
4471 	/* 0xC0 - 0xC7 */
4472 	N, N, N, N, N, N, N, N,
4473 	/* 0xC8 - 0xCF */
4474 	N, N, N, N, N, N, N, N,
4475 	/* 0xD0 - 0xC7 */
4476 	N, N, N, N, N, N, N, N,
4477 	/* 0xD8 - 0xDF */
4478 	N, N, N, N, N, N, N, N,
4479 	/* 0xE0 - 0xE7 */
4480 	N, N, N, N, N, N, N, N,
4481 	/* 0xE8 - 0xEF */
4482 	N, N, N, N, N, N, N, N,
4483 	/* 0xF0 - 0xF7 */
4484 	N, N, N, N, N, N, N, N,
4485 	/* 0xF8 - 0xFF */
4486 	N, N, N, N, N, N, N, N,
4487 } };
4488 
4489 static const struct instr_dual instr_dual_0f_c3 = {
4490 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4491 };
4492 
4493 static const struct mode_dual mode_dual_63 = {
4494 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4495 };
4496 
4497 static const struct opcode opcode_table[256] = {
4498 	/* 0x00 - 0x07 */
4499 	F6ALU(Lock, em_add),
4500 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4501 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4502 	/* 0x08 - 0x0F */
4503 	F6ALU(Lock | PageTable, em_or),
4504 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4505 	N,
4506 	/* 0x10 - 0x17 */
4507 	F6ALU(Lock, em_adc),
4508 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4509 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4510 	/* 0x18 - 0x1F */
4511 	F6ALU(Lock, em_sbb),
4512 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4513 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4514 	/* 0x20 - 0x27 */
4515 	F6ALU(Lock | PageTable, em_and), N, N,
4516 	/* 0x28 - 0x2F */
4517 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4518 	/* 0x30 - 0x37 */
4519 	F6ALU(Lock, em_xor), N, N,
4520 	/* 0x38 - 0x3F */
4521 	F6ALU(NoWrite, em_cmp), N, N,
4522 	/* 0x40 - 0x4F */
4523 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4524 	/* 0x50 - 0x57 */
4525 	X8(I(SrcReg | Stack, em_push)),
4526 	/* 0x58 - 0x5F */
4527 	X8(I(DstReg | Stack, em_pop)),
4528 	/* 0x60 - 0x67 */
4529 	I(ImplicitOps | Stack | No64, em_pusha),
4530 	I(ImplicitOps | Stack | No64, em_popa),
4531 	N, MD(ModRM, &mode_dual_63),
4532 	N, N, N, N,
4533 	/* 0x68 - 0x6F */
4534 	I(SrcImm | Mov | Stack, em_push),
4535 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4536 	I(SrcImmByte | Mov | Stack, em_push),
4537 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4538 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4539 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4540 	/* 0x70 - 0x7F */
4541 	X16(D(SrcImmByte | NearBranch)),
4542 	/* 0x80 - 0x87 */
4543 	G(ByteOp | DstMem | SrcImm, group1),
4544 	G(DstMem | SrcImm, group1),
4545 	G(ByteOp | DstMem | SrcImm | No64, group1),
4546 	G(DstMem | SrcImmByte, group1),
4547 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4548 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4549 	/* 0x88 - 0x8F */
4550 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4551 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4552 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4553 	D(ModRM | SrcMem | NoAccess | DstReg),
4554 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4555 	G(0, group1A),
4556 	/* 0x90 - 0x97 */
4557 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4558 	/* 0x98 - 0x9F */
4559 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4560 	I(SrcImmFAddr | No64, em_call_far), N,
4561 	II(ImplicitOps | Stack, em_pushf, pushf),
4562 	II(ImplicitOps | Stack, em_popf, popf),
4563 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4564 	/* 0xA0 - 0xA7 */
4565 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4566 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4567 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4568 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4569 	/* 0xA8 - 0xAF */
4570 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4571 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4572 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4573 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4574 	/* 0xB0 - 0xB7 */
4575 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4576 	/* 0xB8 - 0xBF */
4577 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4578 	/* 0xC0 - 0xC7 */
4579 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4580 	I(ImplicitOps | NearBranch | SrcImmU16, em_ret_near_imm),
4581 	I(ImplicitOps | NearBranch, em_ret),
4582 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4583 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4584 	G(ByteOp, group11), G(0, group11),
4585 	/* 0xC8 - 0xCF */
4586 	I(Stack | SrcImmU16 | Src2ImmByte, em_enter), I(Stack, em_leave),
4587 	I(ImplicitOps | SrcImmU16, em_ret_far_imm),
4588 	I(ImplicitOps, em_ret_far),
4589 	D(ImplicitOps), DI(SrcImmByte, intn),
4590 	D(ImplicitOps | No64), II(ImplicitOps, em_iret, iret),
4591 	/* 0xD0 - 0xD7 */
4592 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4593 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4594 	I(DstAcc | SrcImmUByte | No64, em_aam),
4595 	I(DstAcc | SrcImmUByte | No64, em_aad),
4596 	F(DstAcc | ByteOp | No64, em_salc),
4597 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4598 	/* 0xD8 - 0xDF */
4599 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4600 	/* 0xE0 - 0xE7 */
4601 	X3(I(SrcImmByte | NearBranch, em_loop)),
4602 	I(SrcImmByte | NearBranch, em_jcxz),
4603 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4604 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4605 	/* 0xE8 - 0xEF */
4606 	I(SrcImm | NearBranch, em_call), D(SrcImm | ImplicitOps | NearBranch),
4607 	I(SrcImmFAddr | No64, em_jmp_far),
4608 	D(SrcImmByte | ImplicitOps | NearBranch),
4609 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4610 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4611 	/* 0xF0 - 0xF7 */
4612 	N, DI(ImplicitOps, icebp), N, N,
4613 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4614 	G(ByteOp, group3), G(0, group3),
4615 	/* 0xF8 - 0xFF */
4616 	D(ImplicitOps), D(ImplicitOps),
4617 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4618 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4619 };
4620 
4621 static const struct opcode twobyte_table[256] = {
4622 	/* 0x00 - 0x0F */
4623 	G(0, group6), GD(0, &group7), N, N,
4624 	N, I(ImplicitOps | EmulateOnUD, em_syscall),
4625 	II(ImplicitOps | Priv, em_clts, clts), N,
4626 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4627 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4628 	/* 0x10 - 0x1F */
4629 	N, N, N, N, N, N, N, N,
4630 	D(ImplicitOps | ModRM | SrcMem | NoAccess),
4631 	N, N, N, N, N, N, D(ImplicitOps | ModRM | SrcMem | NoAccess),
4632 	/* 0x20 - 0x2F */
4633 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_read),
4634 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4635 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4636 						check_cr_write),
4637 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4638 						check_dr_write),
4639 	N, N, N, N,
4640 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4641 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4642 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4643 	N, N, N, N,
4644 	/* 0x30 - 0x3F */
4645 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4646 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4647 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4648 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4649 	I(ImplicitOps | EmulateOnUD, em_sysenter),
4650 	I(ImplicitOps | Priv | EmulateOnUD, em_sysexit),
4651 	N, N,
4652 	N, N, N, N, N, N, N, N,
4653 	/* 0x40 - 0x4F */
4654 	X16(D(DstReg | SrcMem | ModRM)),
4655 	/* 0x50 - 0x5F */
4656 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4657 	/* 0x60 - 0x6F */
4658 	N, N, N, N,
4659 	N, N, N, N,
4660 	N, N, N, N,
4661 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4662 	/* 0x70 - 0x7F */
4663 	N, N, N, N,
4664 	N, N, N, N,
4665 	N, N, N, N,
4666 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4667 	/* 0x80 - 0x8F */
4668 	X16(D(SrcImm | NearBranch)),
4669 	/* 0x90 - 0x9F */
4670 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4671 	/* 0xA0 - 0xA7 */
4672 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4673 	II(ImplicitOps, em_cpuid, cpuid),
4674 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4675 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4676 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4677 	/* 0xA8 - 0xAF */
4678 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4679 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4680 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4681 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4682 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4683 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4684 	/* 0xB0 - 0xB7 */
4685 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4686 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4687 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4688 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4689 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4690 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4691 	/* 0xB8 - 0xBF */
4692 	N, N,
4693 	G(BitOp, group8),
4694 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4695 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4696 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4697 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4698 	/* 0xC0 - 0xC7 */
4699 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4700 	N, ID(0, &instr_dual_0f_c3),
4701 	N, N, N, GD(0, &group9),
4702 	/* 0xC8 - 0xCF */
4703 	X8(I(DstReg, em_bswap)),
4704 	/* 0xD0 - 0xDF */
4705 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4706 	/* 0xE0 - 0xEF */
4707 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4708 	N, N, N, N, N, N, N, N,
4709 	/* 0xF0 - 0xFF */
4710 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4711 };
4712 
4713 static const struct instr_dual instr_dual_0f_38_f0 = {
4714 	I(DstReg | SrcMem | Mov, em_movbe), N
4715 };
4716 
4717 static const struct instr_dual instr_dual_0f_38_f1 = {
4718 	I(DstMem | SrcReg | Mov, em_movbe), N
4719 };
4720 
4721 static const struct gprefix three_byte_0f_38_f0 = {
4722 	ID(0, &instr_dual_0f_38_f0), N, N, N
4723 };
4724 
4725 static const struct gprefix three_byte_0f_38_f1 = {
4726 	ID(0, &instr_dual_0f_38_f1), N, N, N
4727 };
4728 
4729 /*
4730  * Insns below are selected by the prefix which indexed by the third opcode
4731  * byte.
4732  */
4733 static const struct opcode opcode_map_0f_38[256] = {
4734 	/* 0x00 - 0x7f */
4735 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4736 	/* 0x80 - 0xef */
4737 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4738 	/* 0xf0 - 0xf1 */
4739 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4740 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4741 	/* 0xf2 - 0xff */
4742 	N, N, X4(N), X8(N)
4743 };
4744 
4745 #undef D
4746 #undef N
4747 #undef G
4748 #undef GD
4749 #undef I
4750 #undef GP
4751 #undef EXT
4752 #undef MD
4753 #undef ID
4754 
4755 #undef D2bv
4756 #undef D2bvIP
4757 #undef I2bv
4758 #undef I2bvIP
4759 #undef I6ALU
4760 
4761 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4762 {
4763 	unsigned size;
4764 
4765 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4766 	if (size == 8)
4767 		size = 4;
4768 	return size;
4769 }
4770 
4771 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4772 		      unsigned size, bool sign_extension)
4773 {
4774 	int rc = X86EMUL_CONTINUE;
4775 
4776 	op->type = OP_IMM;
4777 	op->bytes = size;
4778 	op->addr.mem.ea = ctxt->_eip;
4779 	/* NB. Immediates are sign-extended as necessary. */
4780 	switch (op->bytes) {
4781 	case 1:
4782 		op->val = insn_fetch(s8, ctxt);
4783 		break;
4784 	case 2:
4785 		op->val = insn_fetch(s16, ctxt);
4786 		break;
4787 	case 4:
4788 		op->val = insn_fetch(s32, ctxt);
4789 		break;
4790 	case 8:
4791 		op->val = insn_fetch(s64, ctxt);
4792 		break;
4793 	}
4794 	if (!sign_extension) {
4795 		switch (op->bytes) {
4796 		case 1:
4797 			op->val &= 0xff;
4798 			break;
4799 		case 2:
4800 			op->val &= 0xffff;
4801 			break;
4802 		case 4:
4803 			op->val &= 0xffffffff;
4804 			break;
4805 		}
4806 	}
4807 done:
4808 	return rc;
4809 }
4810 
4811 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4812 			  unsigned d)
4813 {
4814 	int rc = X86EMUL_CONTINUE;
4815 
4816 	switch (d) {
4817 	case OpReg:
4818 		decode_register_operand(ctxt, op);
4819 		break;
4820 	case OpImmUByte:
4821 		rc = decode_imm(ctxt, op, 1, false);
4822 		break;
4823 	case OpMem:
4824 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4825 	mem_common:
4826 		*op = ctxt->memop;
4827 		ctxt->memopp = op;
4828 		if (ctxt->d & BitOp)
4829 			fetch_bit_operand(ctxt);
4830 		op->orig_val = op->val;
4831 		break;
4832 	case OpMem64:
4833 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4834 		goto mem_common;
4835 	case OpAcc:
4836 		op->type = OP_REG;
4837 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4838 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4839 		fetch_register_operand(op);
4840 		op->orig_val = op->val;
4841 		break;
4842 	case OpAccLo:
4843 		op->type = OP_REG;
4844 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4845 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4846 		fetch_register_operand(op);
4847 		op->orig_val = op->val;
4848 		break;
4849 	case OpAccHi:
4850 		if (ctxt->d & ByteOp) {
4851 			op->type = OP_NONE;
4852 			break;
4853 		}
4854 		op->type = OP_REG;
4855 		op->bytes = ctxt->op_bytes;
4856 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4857 		fetch_register_operand(op);
4858 		op->orig_val = op->val;
4859 		break;
4860 	case OpDI:
4861 		op->type = OP_MEM;
4862 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4863 		op->addr.mem.ea =
4864 			register_address(ctxt, VCPU_REGS_RDI);
4865 		op->addr.mem.seg = VCPU_SREG_ES;
4866 		op->val = 0;
4867 		op->count = 1;
4868 		break;
4869 	case OpDX:
4870 		op->type = OP_REG;
4871 		op->bytes = 2;
4872 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4873 		fetch_register_operand(op);
4874 		break;
4875 	case OpCL:
4876 		op->type = OP_IMM;
4877 		op->bytes = 1;
4878 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4879 		break;
4880 	case OpImmByte:
4881 		rc = decode_imm(ctxt, op, 1, true);
4882 		break;
4883 	case OpOne:
4884 		op->type = OP_IMM;
4885 		op->bytes = 1;
4886 		op->val = 1;
4887 		break;
4888 	case OpImm:
4889 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4890 		break;
4891 	case OpImm64:
4892 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4893 		break;
4894 	case OpMem8:
4895 		ctxt->memop.bytes = 1;
4896 		if (ctxt->memop.type == OP_REG) {
4897 			ctxt->memop.addr.reg = decode_register(ctxt,
4898 					ctxt->modrm_rm, true);
4899 			fetch_register_operand(&ctxt->memop);
4900 		}
4901 		goto mem_common;
4902 	case OpMem16:
4903 		ctxt->memop.bytes = 2;
4904 		goto mem_common;
4905 	case OpMem32:
4906 		ctxt->memop.bytes = 4;
4907 		goto mem_common;
4908 	case OpImmU16:
4909 		rc = decode_imm(ctxt, op, 2, false);
4910 		break;
4911 	case OpImmU:
4912 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4913 		break;
4914 	case OpSI:
4915 		op->type = OP_MEM;
4916 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4917 		op->addr.mem.ea =
4918 			register_address(ctxt, VCPU_REGS_RSI);
4919 		op->addr.mem.seg = ctxt->seg_override;
4920 		op->val = 0;
4921 		op->count = 1;
4922 		break;
4923 	case OpXLat:
4924 		op->type = OP_MEM;
4925 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4926 		op->addr.mem.ea =
4927 			address_mask(ctxt,
4928 				reg_read(ctxt, VCPU_REGS_RBX) +
4929 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
4930 		op->addr.mem.seg = ctxt->seg_override;
4931 		op->val = 0;
4932 		break;
4933 	case OpImmFAddr:
4934 		op->type = OP_IMM;
4935 		op->addr.mem.ea = ctxt->_eip;
4936 		op->bytes = ctxt->op_bytes + 2;
4937 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
4938 		break;
4939 	case OpMemFAddr:
4940 		ctxt->memop.bytes = ctxt->op_bytes + 2;
4941 		goto mem_common;
4942 	case OpES:
4943 		op->type = OP_IMM;
4944 		op->val = VCPU_SREG_ES;
4945 		break;
4946 	case OpCS:
4947 		op->type = OP_IMM;
4948 		op->val = VCPU_SREG_CS;
4949 		break;
4950 	case OpSS:
4951 		op->type = OP_IMM;
4952 		op->val = VCPU_SREG_SS;
4953 		break;
4954 	case OpDS:
4955 		op->type = OP_IMM;
4956 		op->val = VCPU_SREG_DS;
4957 		break;
4958 	case OpFS:
4959 		op->type = OP_IMM;
4960 		op->val = VCPU_SREG_FS;
4961 		break;
4962 	case OpGS:
4963 		op->type = OP_IMM;
4964 		op->val = VCPU_SREG_GS;
4965 		break;
4966 	case OpImplicit:
4967 		/* Special instructions do their own operand decoding. */
4968 	default:
4969 		op->type = OP_NONE; /* Disable writeback. */
4970 		break;
4971 	}
4972 
4973 done:
4974 	return rc;
4975 }
4976 
4977 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len)
4978 {
4979 	int rc = X86EMUL_CONTINUE;
4980 	int mode = ctxt->mode;
4981 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
4982 	bool op_prefix = false;
4983 	bool has_seg_override = false;
4984 	struct opcode opcode;
4985 
4986 	ctxt->memop.type = OP_NONE;
4987 	ctxt->memopp = NULL;
4988 	ctxt->_eip = ctxt->eip;
4989 	ctxt->fetch.ptr = ctxt->fetch.data;
4990 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
4991 	ctxt->opcode_len = 1;
4992 	if (insn_len > 0)
4993 		memcpy(ctxt->fetch.data, insn, insn_len);
4994 	else {
4995 		rc = __do_insn_fetch_bytes(ctxt, 1);
4996 		if (rc != X86EMUL_CONTINUE)
4997 			return rc;
4998 	}
4999 
5000 	switch (mode) {
5001 	case X86EMUL_MODE_REAL:
5002 	case X86EMUL_MODE_VM86:
5003 	case X86EMUL_MODE_PROT16:
5004 		def_op_bytes = def_ad_bytes = 2;
5005 		break;
5006 	case X86EMUL_MODE_PROT32:
5007 		def_op_bytes = def_ad_bytes = 4;
5008 		break;
5009 #ifdef CONFIG_X86_64
5010 	case X86EMUL_MODE_PROT64:
5011 		def_op_bytes = 4;
5012 		def_ad_bytes = 8;
5013 		break;
5014 #endif
5015 	default:
5016 		return EMULATION_FAILED;
5017 	}
5018 
5019 	ctxt->op_bytes = def_op_bytes;
5020 	ctxt->ad_bytes = def_ad_bytes;
5021 
5022 	/* Legacy prefixes. */
5023 	for (;;) {
5024 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5025 		case 0x66:	/* operand-size override */
5026 			op_prefix = true;
5027 			/* switch between 2/4 bytes */
5028 			ctxt->op_bytes = def_op_bytes ^ 6;
5029 			break;
5030 		case 0x67:	/* address-size override */
5031 			if (mode == X86EMUL_MODE_PROT64)
5032 				/* switch between 4/8 bytes */
5033 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5034 			else
5035 				/* switch between 2/4 bytes */
5036 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5037 			break;
5038 		case 0x26:	/* ES override */
5039 		case 0x2e:	/* CS override */
5040 		case 0x36:	/* SS override */
5041 		case 0x3e:	/* DS override */
5042 			has_seg_override = true;
5043 			ctxt->seg_override = (ctxt->b >> 3) & 3;
5044 			break;
5045 		case 0x64:	/* FS override */
5046 		case 0x65:	/* GS override */
5047 			has_seg_override = true;
5048 			ctxt->seg_override = ctxt->b & 7;
5049 			break;
5050 		case 0x40 ... 0x4f: /* REX */
5051 			if (mode != X86EMUL_MODE_PROT64)
5052 				goto done_prefixes;
5053 			ctxt->rex_prefix = ctxt->b;
5054 			continue;
5055 		case 0xf0:	/* LOCK */
5056 			ctxt->lock_prefix = 1;
5057 			break;
5058 		case 0xf2:	/* REPNE/REPNZ */
5059 		case 0xf3:	/* REP/REPE/REPZ */
5060 			ctxt->rep_prefix = ctxt->b;
5061 			break;
5062 		default:
5063 			goto done_prefixes;
5064 		}
5065 
5066 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5067 
5068 		ctxt->rex_prefix = 0;
5069 	}
5070 
5071 done_prefixes:
5072 
5073 	/* REX prefix. */
5074 	if (ctxt->rex_prefix & 8)
5075 		ctxt->op_bytes = 8;	/* REX.W */
5076 
5077 	/* Opcode byte(s). */
5078 	opcode = opcode_table[ctxt->b];
5079 	/* Two-byte opcode? */
5080 	if (ctxt->b == 0x0f) {
5081 		ctxt->opcode_len = 2;
5082 		ctxt->b = insn_fetch(u8, ctxt);
5083 		opcode = twobyte_table[ctxt->b];
5084 
5085 		/* 0F_38 opcode map */
5086 		if (ctxt->b == 0x38) {
5087 			ctxt->opcode_len = 3;
5088 			ctxt->b = insn_fetch(u8, ctxt);
5089 			opcode = opcode_map_0f_38[ctxt->b];
5090 		}
5091 	}
5092 	ctxt->d = opcode.flags;
5093 
5094 	if (ctxt->d & ModRM)
5095 		ctxt->modrm = insn_fetch(u8, ctxt);
5096 
5097 	/* vex-prefix instructions are not implemented */
5098 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5099 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5100 		ctxt->d = NotImpl;
5101 	}
5102 
5103 	while (ctxt->d & GroupMask) {
5104 		switch (ctxt->d & GroupMask) {
5105 		case Group:
5106 			goffset = (ctxt->modrm >> 3) & 7;
5107 			opcode = opcode.u.group[goffset];
5108 			break;
5109 		case GroupDual:
5110 			goffset = (ctxt->modrm >> 3) & 7;
5111 			if ((ctxt->modrm >> 6) == 3)
5112 				opcode = opcode.u.gdual->mod3[goffset];
5113 			else
5114 				opcode = opcode.u.gdual->mod012[goffset];
5115 			break;
5116 		case RMExt:
5117 			goffset = ctxt->modrm & 7;
5118 			opcode = opcode.u.group[goffset];
5119 			break;
5120 		case Prefix:
5121 			if (ctxt->rep_prefix && op_prefix)
5122 				return EMULATION_FAILED;
5123 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5124 			switch (simd_prefix) {
5125 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5126 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5127 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5128 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5129 			}
5130 			break;
5131 		case Escape:
5132 			if (ctxt->modrm > 0xbf)
5133 				opcode = opcode.u.esc->high[ctxt->modrm - 0xc0];
5134 			else
5135 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5136 			break;
5137 		case InstrDual:
5138 			if ((ctxt->modrm >> 6) == 3)
5139 				opcode = opcode.u.idual->mod3;
5140 			else
5141 				opcode = opcode.u.idual->mod012;
5142 			break;
5143 		case ModeDual:
5144 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5145 				opcode = opcode.u.mdual->mode64;
5146 			else
5147 				opcode = opcode.u.mdual->mode32;
5148 			break;
5149 		default:
5150 			return EMULATION_FAILED;
5151 		}
5152 
5153 		ctxt->d &= ~(u64)GroupMask;
5154 		ctxt->d |= opcode.flags;
5155 	}
5156 
5157 	/* Unrecognised? */
5158 	if (ctxt->d == 0)
5159 		return EMULATION_FAILED;
5160 
5161 	ctxt->execute = opcode.u.execute;
5162 
5163 	if (unlikely(ctxt->ud) && likely(!(ctxt->d & EmulateOnUD)))
5164 		return EMULATION_FAILED;
5165 
5166 	if (unlikely(ctxt->d &
5167 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5168 	     No16))) {
5169 		/*
5170 		 * These are copied unconditionally here, and checked unconditionally
5171 		 * in x86_emulate_insn.
5172 		 */
5173 		ctxt->check_perm = opcode.check_perm;
5174 		ctxt->intercept = opcode.intercept;
5175 
5176 		if (ctxt->d & NotImpl)
5177 			return EMULATION_FAILED;
5178 
5179 		if (mode == X86EMUL_MODE_PROT64) {
5180 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5181 				ctxt->op_bytes = 8;
5182 			else if (ctxt->d & NearBranch)
5183 				ctxt->op_bytes = 8;
5184 		}
5185 
5186 		if (ctxt->d & Op3264) {
5187 			if (mode == X86EMUL_MODE_PROT64)
5188 				ctxt->op_bytes = 8;
5189 			else
5190 				ctxt->op_bytes = 4;
5191 		}
5192 
5193 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5194 			ctxt->op_bytes = 4;
5195 
5196 		if (ctxt->d & Sse)
5197 			ctxt->op_bytes = 16;
5198 		else if (ctxt->d & Mmx)
5199 			ctxt->op_bytes = 8;
5200 	}
5201 
5202 	/* ModRM and SIB bytes. */
5203 	if (ctxt->d & ModRM) {
5204 		rc = decode_modrm(ctxt, &ctxt->memop);
5205 		if (!has_seg_override) {
5206 			has_seg_override = true;
5207 			ctxt->seg_override = ctxt->modrm_seg;
5208 		}
5209 	} else if (ctxt->d & MemAbs)
5210 		rc = decode_abs(ctxt, &ctxt->memop);
5211 	if (rc != X86EMUL_CONTINUE)
5212 		goto done;
5213 
5214 	if (!has_seg_override)
5215 		ctxt->seg_override = VCPU_SREG_DS;
5216 
5217 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5218 
5219 	/*
5220 	 * Decode and fetch the source operand: register, memory
5221 	 * or immediate.
5222 	 */
5223 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5224 	if (rc != X86EMUL_CONTINUE)
5225 		goto done;
5226 
5227 	/*
5228 	 * Decode and fetch the second source operand: register, memory
5229 	 * or immediate.
5230 	 */
5231 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5232 	if (rc != X86EMUL_CONTINUE)
5233 		goto done;
5234 
5235 	/* Decode and fetch the destination operand: register or memory. */
5236 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5237 
5238 	if (ctxt->rip_relative && likely(ctxt->memopp))
5239 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5240 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5241 
5242 done:
5243 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5244 }
5245 
5246 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5247 {
5248 	return ctxt->d & PageTable;
5249 }
5250 
5251 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5252 {
5253 	/* The second termination condition only applies for REPE
5254 	 * and REPNE. Test if the repeat string operation prefix is
5255 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5256 	 * corresponding termination condition according to:
5257 	 * 	- if REPE/REPZ and ZF = 0 then done
5258 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5259 	 */
5260 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5261 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5262 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5263 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5264 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5265 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5266 		return true;
5267 
5268 	return false;
5269 }
5270 
5271 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5272 {
5273 	int rc;
5274 
5275 	ctxt->ops->get_fpu(ctxt);
5276 	rc = asm_safe("fwait");
5277 	ctxt->ops->put_fpu(ctxt);
5278 
5279 	if (unlikely(rc != X86EMUL_CONTINUE))
5280 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5281 
5282 	return X86EMUL_CONTINUE;
5283 }
5284 
5285 static void fetch_possible_mmx_operand(struct x86_emulate_ctxt *ctxt,
5286 				       struct operand *op)
5287 {
5288 	if (op->type == OP_MM)
5289 		read_mmx_reg(ctxt, &op->mm_val, op->addr.mm);
5290 }
5291 
5292 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *))
5293 {
5294 	register void *__sp asm(_ASM_SP);
5295 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5296 
5297 	if (!(ctxt->d & ByteOp))
5298 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5299 
5300 	asm("push %[flags]; popf; call *%[fastop]; pushf; pop %[flags]\n"
5301 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5302 	      [fastop]"+S"(fop), "+r"(__sp)
5303 	    : "c"(ctxt->src2.val));
5304 
5305 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5306 	if (!fop) /* exception is returned in fop variable */
5307 		return emulate_de(ctxt);
5308 	return X86EMUL_CONTINUE;
5309 }
5310 
5311 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5312 {
5313 	memset(&ctxt->rip_relative, 0,
5314 	       (void *)&ctxt->modrm - (void *)&ctxt->rip_relative);
5315 
5316 	ctxt->io_read.pos = 0;
5317 	ctxt->io_read.end = 0;
5318 	ctxt->mem_read.end = 0;
5319 }
5320 
5321 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5322 {
5323 	const struct x86_emulate_ops *ops = ctxt->ops;
5324 	int rc = X86EMUL_CONTINUE;
5325 	int saved_dst_type = ctxt->dst.type;
5326 	unsigned emul_flags;
5327 
5328 	ctxt->mem_read.pos = 0;
5329 
5330 	/* LOCK prefix is allowed only with some instructions */
5331 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5332 		rc = emulate_ud(ctxt);
5333 		goto done;
5334 	}
5335 
5336 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5337 		rc = emulate_ud(ctxt);
5338 		goto done;
5339 	}
5340 
5341 	emul_flags = ctxt->ops->get_hflags(ctxt);
5342 	if (unlikely(ctxt->d &
5343 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5344 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5345 				(ctxt->d & Undefined)) {
5346 			rc = emulate_ud(ctxt);
5347 			goto done;
5348 		}
5349 
5350 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5351 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5352 			rc = emulate_ud(ctxt);
5353 			goto done;
5354 		}
5355 
5356 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5357 			rc = emulate_nm(ctxt);
5358 			goto done;
5359 		}
5360 
5361 		if (ctxt->d & Mmx) {
5362 			rc = flush_pending_x87_faults(ctxt);
5363 			if (rc != X86EMUL_CONTINUE)
5364 				goto done;
5365 			/*
5366 			 * Now that we know the fpu is exception safe, we can fetch
5367 			 * operands from it.
5368 			 */
5369 			fetch_possible_mmx_operand(ctxt, &ctxt->src);
5370 			fetch_possible_mmx_operand(ctxt, &ctxt->src2);
5371 			if (!(ctxt->d & Mov))
5372 				fetch_possible_mmx_operand(ctxt, &ctxt->dst);
5373 		}
5374 
5375 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5376 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5377 						      X86_ICPT_PRE_EXCEPT);
5378 			if (rc != X86EMUL_CONTINUE)
5379 				goto done;
5380 		}
5381 
5382 		/* Instruction can only be executed in protected mode */
5383 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5384 			rc = emulate_ud(ctxt);
5385 			goto done;
5386 		}
5387 
5388 		/* Privileged instruction can be executed only in CPL=0 */
5389 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5390 			if (ctxt->d & PrivUD)
5391 				rc = emulate_ud(ctxt);
5392 			else
5393 				rc = emulate_gp(ctxt, 0);
5394 			goto done;
5395 		}
5396 
5397 		/* Do instruction specific permission checks */
5398 		if (ctxt->d & CheckPerm) {
5399 			rc = ctxt->check_perm(ctxt);
5400 			if (rc != X86EMUL_CONTINUE)
5401 				goto done;
5402 		}
5403 
5404 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5405 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5406 						      X86_ICPT_POST_EXCEPT);
5407 			if (rc != X86EMUL_CONTINUE)
5408 				goto done;
5409 		}
5410 
5411 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5412 			/* All REP prefixes have the same first termination condition */
5413 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5414 				string_registers_quirk(ctxt);
5415 				ctxt->eip = ctxt->_eip;
5416 				ctxt->eflags &= ~X86_EFLAGS_RF;
5417 				goto done;
5418 			}
5419 		}
5420 	}
5421 
5422 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5423 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5424 				    ctxt->src.valptr, ctxt->src.bytes);
5425 		if (rc != X86EMUL_CONTINUE)
5426 			goto done;
5427 		ctxt->src.orig_val64 = ctxt->src.val64;
5428 	}
5429 
5430 	if (ctxt->src2.type == OP_MEM) {
5431 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5432 				    &ctxt->src2.val, ctxt->src2.bytes);
5433 		if (rc != X86EMUL_CONTINUE)
5434 			goto done;
5435 	}
5436 
5437 	if ((ctxt->d & DstMask) == ImplicitOps)
5438 		goto special_insn;
5439 
5440 
5441 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5442 		/* optimisation - avoid slow emulated read if Mov */
5443 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5444 				   &ctxt->dst.val, ctxt->dst.bytes);
5445 		if (rc != X86EMUL_CONTINUE) {
5446 			if (!(ctxt->d & NoWrite) &&
5447 			    rc == X86EMUL_PROPAGATE_FAULT &&
5448 			    ctxt->exception.vector == PF_VECTOR)
5449 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5450 			goto done;
5451 		}
5452 	}
5453 	/* Copy full 64-bit value for CMPXCHG8B.  */
5454 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5455 
5456 special_insn:
5457 
5458 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5459 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5460 					      X86_ICPT_POST_MEMACCESS);
5461 		if (rc != X86EMUL_CONTINUE)
5462 			goto done;
5463 	}
5464 
5465 	if (ctxt->rep_prefix && (ctxt->d & String))
5466 		ctxt->eflags |= X86_EFLAGS_RF;
5467 	else
5468 		ctxt->eflags &= ~X86_EFLAGS_RF;
5469 
5470 	if (ctxt->execute) {
5471 		if (ctxt->d & Fastop) {
5472 			void (*fop)(struct fastop *) = (void *)ctxt->execute;
5473 			rc = fastop(ctxt, fop);
5474 			if (rc != X86EMUL_CONTINUE)
5475 				goto done;
5476 			goto writeback;
5477 		}
5478 		rc = ctxt->execute(ctxt);
5479 		if (rc != X86EMUL_CONTINUE)
5480 			goto done;
5481 		goto writeback;
5482 	}
5483 
5484 	if (ctxt->opcode_len == 2)
5485 		goto twobyte_insn;
5486 	else if (ctxt->opcode_len == 3)
5487 		goto threebyte_insn;
5488 
5489 	switch (ctxt->b) {
5490 	case 0x70 ... 0x7f: /* jcc (short) */
5491 		if (test_cc(ctxt->b, ctxt->eflags))
5492 			rc = jmp_rel(ctxt, ctxt->src.val);
5493 		break;
5494 	case 0x8d: /* lea r16/r32, m */
5495 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5496 		break;
5497 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5498 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5499 			ctxt->dst.type = OP_NONE;
5500 		else
5501 			rc = em_xchg(ctxt);
5502 		break;
5503 	case 0x98: /* cbw/cwde/cdqe */
5504 		switch (ctxt->op_bytes) {
5505 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5506 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5507 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5508 		}
5509 		break;
5510 	case 0xcc:		/* int3 */
5511 		rc = emulate_int(ctxt, 3);
5512 		break;
5513 	case 0xcd:		/* int n */
5514 		rc = emulate_int(ctxt, ctxt->src.val);
5515 		break;
5516 	case 0xce:		/* into */
5517 		if (ctxt->eflags & X86_EFLAGS_OF)
5518 			rc = emulate_int(ctxt, 4);
5519 		break;
5520 	case 0xe9: /* jmp rel */
5521 	case 0xeb: /* jmp rel short */
5522 		rc = jmp_rel(ctxt, ctxt->src.val);
5523 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5524 		break;
5525 	case 0xf4:              /* hlt */
5526 		ctxt->ops->halt(ctxt);
5527 		break;
5528 	case 0xf5:	/* cmc */
5529 		/* complement carry flag from eflags reg */
5530 		ctxt->eflags ^= X86_EFLAGS_CF;
5531 		break;
5532 	case 0xf8: /* clc */
5533 		ctxt->eflags &= ~X86_EFLAGS_CF;
5534 		break;
5535 	case 0xf9: /* stc */
5536 		ctxt->eflags |= X86_EFLAGS_CF;
5537 		break;
5538 	case 0xfc: /* cld */
5539 		ctxt->eflags &= ~X86_EFLAGS_DF;
5540 		break;
5541 	case 0xfd: /* std */
5542 		ctxt->eflags |= X86_EFLAGS_DF;
5543 		break;
5544 	default:
5545 		goto cannot_emulate;
5546 	}
5547 
5548 	if (rc != X86EMUL_CONTINUE)
5549 		goto done;
5550 
5551 writeback:
5552 	if (ctxt->d & SrcWrite) {
5553 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5554 		rc = writeback(ctxt, &ctxt->src);
5555 		if (rc != X86EMUL_CONTINUE)
5556 			goto done;
5557 	}
5558 	if (!(ctxt->d & NoWrite)) {
5559 		rc = writeback(ctxt, &ctxt->dst);
5560 		if (rc != X86EMUL_CONTINUE)
5561 			goto done;
5562 	}
5563 
5564 	/*
5565 	 * restore dst type in case the decoding will be reused
5566 	 * (happens for string instruction )
5567 	 */
5568 	ctxt->dst.type = saved_dst_type;
5569 
5570 	if ((ctxt->d & SrcMask) == SrcSI)
5571 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5572 
5573 	if ((ctxt->d & DstMask) == DstDI)
5574 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5575 
5576 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5577 		unsigned int count;
5578 		struct read_cache *r = &ctxt->io_read;
5579 		if ((ctxt->d & SrcMask) == SrcSI)
5580 			count = ctxt->src.count;
5581 		else
5582 			count = ctxt->dst.count;
5583 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5584 
5585 		if (!string_insn_completed(ctxt)) {
5586 			/*
5587 			 * Re-enter guest when pio read ahead buffer is empty
5588 			 * or, if it is not used, after each 1024 iteration.
5589 			 */
5590 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5591 			    (r->end == 0 || r->end != r->pos)) {
5592 				/*
5593 				 * Reset read cache. Usually happens before
5594 				 * decode, but since instruction is restarted
5595 				 * we have to do it here.
5596 				 */
5597 				ctxt->mem_read.end = 0;
5598 				writeback_registers(ctxt);
5599 				return EMULATION_RESTART;
5600 			}
5601 			goto done; /* skip rip writeback */
5602 		}
5603 		ctxt->eflags &= ~X86_EFLAGS_RF;
5604 	}
5605 
5606 	ctxt->eip = ctxt->_eip;
5607 
5608 done:
5609 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5610 		WARN_ON(ctxt->exception.vector > 0x1f);
5611 		ctxt->have_exception = true;
5612 	}
5613 	if (rc == X86EMUL_INTERCEPTED)
5614 		return EMULATION_INTERCEPTED;
5615 
5616 	if (rc == X86EMUL_CONTINUE)
5617 		writeback_registers(ctxt);
5618 
5619 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5620 
5621 twobyte_insn:
5622 	switch (ctxt->b) {
5623 	case 0x09:		/* wbinvd */
5624 		(ctxt->ops->wbinvd)(ctxt);
5625 		break;
5626 	case 0x08:		/* invd */
5627 	case 0x0d:		/* GrpP (prefetch) */
5628 	case 0x18:		/* Grp16 (prefetch/nop) */
5629 	case 0x1f:		/* nop */
5630 		break;
5631 	case 0x20: /* mov cr, reg */
5632 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5633 		break;
5634 	case 0x21: /* mov from dr to reg */
5635 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5636 		break;
5637 	case 0x40 ... 0x4f:	/* cmov */
5638 		if (test_cc(ctxt->b, ctxt->eflags))
5639 			ctxt->dst.val = ctxt->src.val;
5640 		else if (ctxt->op_bytes != 4)
5641 			ctxt->dst.type = OP_NONE; /* no writeback */
5642 		break;
5643 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5644 		if (test_cc(ctxt->b, ctxt->eflags))
5645 			rc = jmp_rel(ctxt, ctxt->src.val);
5646 		break;
5647 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5648 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5649 		break;
5650 	case 0xb6 ... 0xb7:	/* movzx */
5651 		ctxt->dst.bytes = ctxt->op_bytes;
5652 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5653 						       : (u16) ctxt->src.val;
5654 		break;
5655 	case 0xbe ... 0xbf:	/* movsx */
5656 		ctxt->dst.bytes = ctxt->op_bytes;
5657 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5658 							(s16) ctxt->src.val;
5659 		break;
5660 	default:
5661 		goto cannot_emulate;
5662 	}
5663 
5664 threebyte_insn:
5665 
5666 	if (rc != X86EMUL_CONTINUE)
5667 		goto done;
5668 
5669 	goto writeback;
5670 
5671 cannot_emulate:
5672 	return EMULATION_FAILED;
5673 }
5674 
5675 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5676 {
5677 	invalidate_registers(ctxt);
5678 }
5679 
5680 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5681 {
5682 	writeback_registers(ctxt);
5683 }
5684 
5685 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5686 {
5687 	if (ctxt->rep_prefix && (ctxt->d & String))
5688 		return false;
5689 
5690 	if (ctxt->d & TwoMemOp)
5691 		return false;
5692 
5693 	return true;
5694 }
5695