xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision 4b4f3acc)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3  * emulate.c
4  *
5  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
6  *
7  * Copyright (c) 2005 Keir Fraser
8  *
9  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
10  * privileged instructions:
11  *
12  * Copyright (C) 2006 Qumranet
13  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
14  *
15  *   Avi Kivity <avi@qumranet.com>
16  *   Yaniv Kamay <yaniv@qumranet.com>
17  *
18  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
19  */
20 
21 #include <linux/kvm_host.h>
22 #include "kvm_cache_regs.h"
23 #include <asm/kvm_emulate.h>
24 #include <linux/stringify.h>
25 #include <asm/debugreg.h>
26 #include <asm/nospec-branch.h>
27 
28 #include "x86.h"
29 #include "tss.h"
30 #include "mmu.h"
31 #include "pmu.h"
32 
33 /*
34  * Operand types
35  */
36 #define OpNone             0ull
37 #define OpImplicit         1ull  /* No generic decode */
38 #define OpReg              2ull  /* Register */
39 #define OpMem              3ull  /* Memory */
40 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
41 #define OpDI               5ull  /* ES:DI/EDI/RDI */
42 #define OpMem64            6ull  /* Memory, 64-bit */
43 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
44 #define OpDX               8ull  /* DX register */
45 #define OpCL               9ull  /* CL register (for shifts) */
46 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
47 #define OpOne             11ull  /* Implied 1 */
48 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
49 #define OpMem16           13ull  /* Memory operand (16-bit). */
50 #define OpMem32           14ull  /* Memory operand (32-bit). */
51 #define OpImmU            15ull  /* Immediate operand, zero extended */
52 #define OpSI              16ull  /* SI/ESI/RSI */
53 #define OpImmFAddr        17ull  /* Immediate far address */
54 #define OpMemFAddr        18ull  /* Far address in memory */
55 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
56 #define OpES              20ull  /* ES */
57 #define OpCS              21ull  /* CS */
58 #define OpSS              22ull  /* SS */
59 #define OpDS              23ull  /* DS */
60 #define OpFS              24ull  /* FS */
61 #define OpGS              25ull  /* GS */
62 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
63 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
64 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
65 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
66 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
67 
68 #define OpBits             5  /* Width of operand field */
69 #define OpMask             ((1ull << OpBits) - 1)
70 
71 /*
72  * Opcode effective-address decode tables.
73  * Note that we only emulate instructions that have at least one memory
74  * operand (excluding implicit stack references). We assume that stack
75  * references and instruction fetches will never occur in special memory
76  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
77  * not be handled.
78  */
79 
80 /* Operand sizes: 8-bit operands or specified/overridden size. */
81 #define ByteOp      (1<<0)	/* 8-bit operands. */
82 /* Destination operand type. */
83 #define DstShift    1
84 #define ImplicitOps (OpImplicit << DstShift)
85 #define DstReg      (OpReg << DstShift)
86 #define DstMem      (OpMem << DstShift)
87 #define DstAcc      (OpAcc << DstShift)
88 #define DstDI       (OpDI << DstShift)
89 #define DstMem64    (OpMem64 << DstShift)
90 #define DstMem16    (OpMem16 << DstShift)
91 #define DstImmUByte (OpImmUByte << DstShift)
92 #define DstDX       (OpDX << DstShift)
93 #define DstAccLo    (OpAccLo << DstShift)
94 #define DstMask     (OpMask << DstShift)
95 /* Source operand type. */
96 #define SrcShift    6
97 #define SrcNone     (OpNone << SrcShift)
98 #define SrcReg      (OpReg << SrcShift)
99 #define SrcMem      (OpMem << SrcShift)
100 #define SrcMem16    (OpMem16 << SrcShift)
101 #define SrcMem32    (OpMem32 << SrcShift)
102 #define SrcImm      (OpImm << SrcShift)
103 #define SrcImmByte  (OpImmByte << SrcShift)
104 #define SrcOne      (OpOne << SrcShift)
105 #define SrcImmUByte (OpImmUByte << SrcShift)
106 #define SrcImmU     (OpImmU << SrcShift)
107 #define SrcSI       (OpSI << SrcShift)
108 #define SrcXLat     (OpXLat << SrcShift)
109 #define SrcImmFAddr (OpImmFAddr << SrcShift)
110 #define SrcMemFAddr (OpMemFAddr << SrcShift)
111 #define SrcAcc      (OpAcc << SrcShift)
112 #define SrcImmU16   (OpImmU16 << SrcShift)
113 #define SrcImm64    (OpImm64 << SrcShift)
114 #define SrcDX       (OpDX << SrcShift)
115 #define SrcMem8     (OpMem8 << SrcShift)
116 #define SrcAccHi    (OpAccHi << SrcShift)
117 #define SrcMask     (OpMask << SrcShift)
118 #define BitOp       (1<<11)
119 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
120 #define String      (1<<13)     /* String instruction (rep capable) */
121 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
122 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
123 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
124 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
125 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
126 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
127 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
128 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
129 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
130 #define Sse         (1<<18)     /* SSE Vector instruction */
131 /* Generic ModRM decode. */
132 #define ModRM       (1<<19)
133 /* Destination is only written; never read. */
134 #define Mov         (1<<20)
135 /* Misc flags */
136 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
137 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
138 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
139 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
140 #define Undefined   (1<<25) /* No Such Instruction */
141 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
142 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
143 #define No64	    (1<<28)
144 #define PageTable   (1 << 29)   /* instruction used to write page table */
145 #define NotImpl     (1 << 30)   /* instruction is not implemented */
146 /* Source 2 operand type */
147 #define Src2Shift   (31)
148 #define Src2None    (OpNone << Src2Shift)
149 #define Src2Mem     (OpMem << Src2Shift)
150 #define Src2CL      (OpCL << Src2Shift)
151 #define Src2ImmByte (OpImmByte << Src2Shift)
152 #define Src2One     (OpOne << Src2Shift)
153 #define Src2Imm     (OpImm << Src2Shift)
154 #define Src2ES      (OpES << Src2Shift)
155 #define Src2CS      (OpCS << Src2Shift)
156 #define Src2SS      (OpSS << Src2Shift)
157 #define Src2DS      (OpDS << Src2Shift)
158 #define Src2FS      (OpFS << Src2Shift)
159 #define Src2GS      (OpGS << Src2Shift)
160 #define Src2Mask    (OpMask << Src2Shift)
161 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
162 #define AlignMask   ((u64)7 << 41)
163 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
164 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
165 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
166 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
167 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
168 #define NoWrite     ((u64)1 << 45)  /* No writeback */
169 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
170 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
171 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
172 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
173 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
174 #define NearBranch  ((u64)1 << 52)  /* Near branches */
175 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
176 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
177 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
178 
179 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
180 
181 #define X2(x...) x, x
182 #define X3(x...) X2(x), x
183 #define X4(x...) X2(x), X2(x)
184 #define X5(x...) X4(x), x
185 #define X6(x...) X4(x), X2(x)
186 #define X7(x...) X4(x), X3(x)
187 #define X8(x...) X4(x), X4(x)
188 #define X16(x...) X8(x), X8(x)
189 
190 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1)
191 #define FASTOP_SIZE 8
192 
193 /*
194  * fastop functions have a special calling convention:
195  *
196  * dst:    rax        (in/out)
197  * src:    rdx        (in/out)
198  * src2:   rcx        (in)
199  * flags:  rflags     (in/out)
200  * ex:     rsi        (in:fastop pointer, out:zero if exception)
201  *
202  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
203  * different operand sizes can be reached by calculation, rather than a jump
204  * table (which would be bigger than the code).
205  *
206  * fastop functions are declared as taking a never-defined fastop parameter,
207  * so they can't be called from C directly.
208  */
209 
210 struct fastop;
211 
212 struct opcode {
213 	u64 flags : 56;
214 	u64 intercept : 8;
215 	union {
216 		int (*execute)(struct x86_emulate_ctxt *ctxt);
217 		const struct opcode *group;
218 		const struct group_dual *gdual;
219 		const struct gprefix *gprefix;
220 		const struct escape *esc;
221 		const struct instr_dual *idual;
222 		const struct mode_dual *mdual;
223 		void (*fastop)(struct fastop *fake);
224 	} u;
225 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
226 };
227 
228 struct group_dual {
229 	struct opcode mod012[8];
230 	struct opcode mod3[8];
231 };
232 
233 struct gprefix {
234 	struct opcode pfx_no;
235 	struct opcode pfx_66;
236 	struct opcode pfx_f2;
237 	struct opcode pfx_f3;
238 };
239 
240 struct escape {
241 	struct opcode op[8];
242 	struct opcode high[64];
243 };
244 
245 struct instr_dual {
246 	struct opcode mod012;
247 	struct opcode mod3;
248 };
249 
250 struct mode_dual {
251 	struct opcode mode32;
252 	struct opcode mode64;
253 };
254 
255 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
256 
257 enum x86_transfer_type {
258 	X86_TRANSFER_NONE,
259 	X86_TRANSFER_CALL_JMP,
260 	X86_TRANSFER_RET,
261 	X86_TRANSFER_TASK_SWITCH,
262 };
263 
264 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
265 {
266 	if (!(ctxt->regs_valid & (1 << nr))) {
267 		ctxt->regs_valid |= 1 << nr;
268 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
269 	}
270 	return ctxt->_regs[nr];
271 }
272 
273 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
274 {
275 	ctxt->regs_valid |= 1 << nr;
276 	ctxt->regs_dirty |= 1 << nr;
277 	return &ctxt->_regs[nr];
278 }
279 
280 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
281 {
282 	reg_read(ctxt, nr);
283 	return reg_write(ctxt, nr);
284 }
285 
286 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
287 {
288 	unsigned reg;
289 
290 	for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16)
291 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
292 }
293 
294 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
295 {
296 	ctxt->regs_dirty = 0;
297 	ctxt->regs_valid = 0;
298 }
299 
300 /*
301  * These EFLAGS bits are restored from saved value during emulation, and
302  * any changes are written back to the saved value after emulation.
303  */
304 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
305 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
306 
307 #ifdef CONFIG_X86_64
308 #define ON64(x) x
309 #else
310 #define ON64(x)
311 #endif
312 
313 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *));
314 
315 #define FOP_FUNC(name) \
316 	".align " __stringify(FASTOP_SIZE) " \n\t" \
317 	".type " name ", @function \n\t" \
318 	name ":\n\t"
319 
320 #define FOP_RET   "ret \n\t"
321 
322 #define FOP_START(op) \
323 	extern void em_##op(struct fastop *fake); \
324 	asm(".pushsection .text, \"ax\" \n\t" \
325 	    ".global em_" #op " \n\t" \
326 	    FOP_FUNC("em_" #op)
327 
328 #define FOP_END \
329 	    ".popsection")
330 
331 #define FOPNOP() \
332 	FOP_FUNC(__stringify(__UNIQUE_ID(nop))) \
333 	FOP_RET
334 
335 #define FOP1E(op,  dst) \
336 	FOP_FUNC(#op "_" #dst) \
337 	"10: " #op " %" #dst " \n\t" FOP_RET
338 
339 #define FOP1EEX(op,  dst) \
340 	FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception)
341 
342 #define FASTOP1(op) \
343 	FOP_START(op) \
344 	FOP1E(op##b, al) \
345 	FOP1E(op##w, ax) \
346 	FOP1E(op##l, eax) \
347 	ON64(FOP1E(op##q, rax))	\
348 	FOP_END
349 
350 /* 1-operand, using src2 (for MUL/DIV r/m) */
351 #define FASTOP1SRC2(op, name) \
352 	FOP_START(name) \
353 	FOP1E(op, cl) \
354 	FOP1E(op, cx) \
355 	FOP1E(op, ecx) \
356 	ON64(FOP1E(op, rcx)) \
357 	FOP_END
358 
359 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
360 #define FASTOP1SRC2EX(op, name) \
361 	FOP_START(name) \
362 	FOP1EEX(op, cl) \
363 	FOP1EEX(op, cx) \
364 	FOP1EEX(op, ecx) \
365 	ON64(FOP1EEX(op, rcx)) \
366 	FOP_END
367 
368 #define FOP2E(op,  dst, src)	   \
369 	FOP_FUNC(#op "_" #dst "_" #src) \
370 	#op " %" #src ", %" #dst " \n\t" FOP_RET
371 
372 #define FASTOP2(op) \
373 	FOP_START(op) \
374 	FOP2E(op##b, al, dl) \
375 	FOP2E(op##w, ax, dx) \
376 	FOP2E(op##l, eax, edx) \
377 	ON64(FOP2E(op##q, rax, rdx)) \
378 	FOP_END
379 
380 /* 2 operand, word only */
381 #define FASTOP2W(op) \
382 	FOP_START(op) \
383 	FOPNOP() \
384 	FOP2E(op##w, ax, dx) \
385 	FOP2E(op##l, eax, edx) \
386 	ON64(FOP2E(op##q, rax, rdx)) \
387 	FOP_END
388 
389 /* 2 operand, src is CL */
390 #define FASTOP2CL(op) \
391 	FOP_START(op) \
392 	FOP2E(op##b, al, cl) \
393 	FOP2E(op##w, ax, cl) \
394 	FOP2E(op##l, eax, cl) \
395 	ON64(FOP2E(op##q, rax, cl)) \
396 	FOP_END
397 
398 /* 2 operand, src and dest are reversed */
399 #define FASTOP2R(op, name) \
400 	FOP_START(name) \
401 	FOP2E(op##b, dl, al) \
402 	FOP2E(op##w, dx, ax) \
403 	FOP2E(op##l, edx, eax) \
404 	ON64(FOP2E(op##q, rdx, rax)) \
405 	FOP_END
406 
407 #define FOP3E(op,  dst, src, src2) \
408 	FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
409 	#op " %" #src2 ", %" #src ", %" #dst " \n\t" FOP_RET
410 
411 /* 3-operand, word-only, src2=cl */
412 #define FASTOP3WCL(op) \
413 	FOP_START(op) \
414 	FOPNOP() \
415 	FOP3E(op##w, ax, dx, cl) \
416 	FOP3E(op##l, eax, edx, cl) \
417 	ON64(FOP3E(op##q, rax, rdx, cl)) \
418 	FOP_END
419 
420 /* Special case for SETcc - 1 instruction per cc */
421 #define FOP_SETCC(op) \
422 	".align 4 \n\t" \
423 	".type " #op ", @function \n\t" \
424 	#op ": \n\t" \
425 	#op " %al \n\t" \
426 	FOP_RET
427 
428 asm(".pushsection .fixup, \"ax\"\n"
429     ".global kvm_fastop_exception \n"
430     "kvm_fastop_exception: xor %esi, %esi; ret\n"
431     ".popsection");
432 
433 FOP_START(setcc)
434 FOP_SETCC(seto)
435 FOP_SETCC(setno)
436 FOP_SETCC(setc)
437 FOP_SETCC(setnc)
438 FOP_SETCC(setz)
439 FOP_SETCC(setnz)
440 FOP_SETCC(setbe)
441 FOP_SETCC(setnbe)
442 FOP_SETCC(sets)
443 FOP_SETCC(setns)
444 FOP_SETCC(setp)
445 FOP_SETCC(setnp)
446 FOP_SETCC(setl)
447 FOP_SETCC(setnl)
448 FOP_SETCC(setle)
449 FOP_SETCC(setnle)
450 FOP_END;
451 
452 FOP_START(salc) "pushf; sbb %al, %al; popf \n\t" FOP_RET
453 FOP_END;
454 
455 /*
456  * XXX: inoutclob user must know where the argument is being expanded.
457  *      Relying on CONFIG_CC_HAS_ASM_GOTO would allow us to remove _fault.
458  */
459 #define asm_safe(insn, inoutclob...) \
460 ({ \
461 	int _fault = 0; \
462  \
463 	asm volatile("1:" insn "\n" \
464 	             "2:\n" \
465 	             ".pushsection .fixup, \"ax\"\n" \
466 	             "3: movl $1, %[_fault]\n" \
467 	             "   jmp  2b\n" \
468 	             ".popsection\n" \
469 	             _ASM_EXTABLE(1b, 3b) \
470 	             : [_fault] "+qm"(_fault) inoutclob ); \
471  \
472 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
473 })
474 
475 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
476 				    enum x86_intercept intercept,
477 				    enum x86_intercept_stage stage)
478 {
479 	struct x86_instruction_info info = {
480 		.intercept  = intercept,
481 		.rep_prefix = ctxt->rep_prefix,
482 		.modrm_mod  = ctxt->modrm_mod,
483 		.modrm_reg  = ctxt->modrm_reg,
484 		.modrm_rm   = ctxt->modrm_rm,
485 		.src_val    = ctxt->src.val64,
486 		.dst_val    = ctxt->dst.val64,
487 		.src_bytes  = ctxt->src.bytes,
488 		.dst_bytes  = ctxt->dst.bytes,
489 		.ad_bytes   = ctxt->ad_bytes,
490 		.next_rip   = ctxt->eip,
491 	};
492 
493 	return ctxt->ops->intercept(ctxt, &info, stage);
494 }
495 
496 static void assign_masked(ulong *dest, ulong src, ulong mask)
497 {
498 	*dest = (*dest & ~mask) | (src & mask);
499 }
500 
501 static void assign_register(unsigned long *reg, u64 val, int bytes)
502 {
503 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
504 	switch (bytes) {
505 	case 1:
506 		*(u8 *)reg = (u8)val;
507 		break;
508 	case 2:
509 		*(u16 *)reg = (u16)val;
510 		break;
511 	case 4:
512 		*reg = (u32)val;
513 		break;	/* 64b: zero-extend */
514 	case 8:
515 		*reg = val;
516 		break;
517 	}
518 }
519 
520 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
521 {
522 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
523 }
524 
525 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
526 {
527 	u16 sel;
528 	struct desc_struct ss;
529 
530 	if (ctxt->mode == X86EMUL_MODE_PROT64)
531 		return ~0UL;
532 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
533 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
534 }
535 
536 static int stack_size(struct x86_emulate_ctxt *ctxt)
537 {
538 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
539 }
540 
541 /* Access/update address held in a register, based on addressing mode. */
542 static inline unsigned long
543 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
544 {
545 	if (ctxt->ad_bytes == sizeof(unsigned long))
546 		return reg;
547 	else
548 		return reg & ad_mask(ctxt);
549 }
550 
551 static inline unsigned long
552 register_address(struct x86_emulate_ctxt *ctxt, int reg)
553 {
554 	return address_mask(ctxt, reg_read(ctxt, reg));
555 }
556 
557 static void masked_increment(ulong *reg, ulong mask, int inc)
558 {
559 	assign_masked(reg, *reg + inc, mask);
560 }
561 
562 static inline void
563 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
564 {
565 	ulong *preg = reg_rmw(ctxt, reg);
566 
567 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
568 }
569 
570 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
571 {
572 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
573 }
574 
575 static u32 desc_limit_scaled(struct desc_struct *desc)
576 {
577 	u32 limit = get_desc_limit(desc);
578 
579 	return desc->g ? (limit << 12) | 0xfff : limit;
580 }
581 
582 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
583 {
584 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
585 		return 0;
586 
587 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
588 }
589 
590 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
591 			     u32 error, bool valid)
592 {
593 	WARN_ON(vec > 0x1f);
594 	ctxt->exception.vector = vec;
595 	ctxt->exception.error_code = error;
596 	ctxt->exception.error_code_valid = valid;
597 	return X86EMUL_PROPAGATE_FAULT;
598 }
599 
600 static int emulate_db(struct x86_emulate_ctxt *ctxt)
601 {
602 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
603 }
604 
605 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
606 {
607 	return emulate_exception(ctxt, GP_VECTOR, err, true);
608 }
609 
610 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
611 {
612 	return emulate_exception(ctxt, SS_VECTOR, err, true);
613 }
614 
615 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
616 {
617 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
618 }
619 
620 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
621 {
622 	return emulate_exception(ctxt, TS_VECTOR, err, true);
623 }
624 
625 static int emulate_de(struct x86_emulate_ctxt *ctxt)
626 {
627 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
628 }
629 
630 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
631 {
632 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
633 }
634 
635 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
636 {
637 	u16 selector;
638 	struct desc_struct desc;
639 
640 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
641 	return selector;
642 }
643 
644 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
645 				 unsigned seg)
646 {
647 	u16 dummy;
648 	u32 base3;
649 	struct desc_struct desc;
650 
651 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
652 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
653 }
654 
655 /*
656  * x86 defines three classes of vector instructions: explicitly
657  * aligned, explicitly unaligned, and the rest, which change behaviour
658  * depending on whether they're AVX encoded or not.
659  *
660  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
661  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
662  * 512 bytes of data must be aligned to a 16 byte boundary.
663  */
664 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
665 {
666 	u64 alignment = ctxt->d & AlignMask;
667 
668 	if (likely(size < 16))
669 		return 1;
670 
671 	switch (alignment) {
672 	case Unaligned:
673 	case Avx:
674 		return 1;
675 	case Aligned16:
676 		return 16;
677 	case Aligned:
678 	default:
679 		return size;
680 	}
681 }
682 
683 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
684 				       struct segmented_address addr,
685 				       unsigned *max_size, unsigned size,
686 				       bool write, bool fetch,
687 				       enum x86emul_mode mode, ulong *linear)
688 {
689 	struct desc_struct desc;
690 	bool usable;
691 	ulong la;
692 	u32 lim;
693 	u16 sel;
694 	u8  va_bits;
695 
696 	la = seg_base(ctxt, addr.seg) + addr.ea;
697 	*max_size = 0;
698 	switch (mode) {
699 	case X86EMUL_MODE_PROT64:
700 		*linear = la;
701 		va_bits = ctxt_virt_addr_bits(ctxt);
702 		if (get_canonical(la, va_bits) != la)
703 			goto bad;
704 
705 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
706 		if (size > *max_size)
707 			goto bad;
708 		break;
709 	default:
710 		*linear = la = (u32)la;
711 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
712 						addr.seg);
713 		if (!usable)
714 			goto bad;
715 		/* code segment in protected mode or read-only data segment */
716 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
717 					|| !(desc.type & 2)) && write)
718 			goto bad;
719 		/* unreadable code segment */
720 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
721 			goto bad;
722 		lim = desc_limit_scaled(&desc);
723 		if (!(desc.type & 8) && (desc.type & 4)) {
724 			/* expand-down segment */
725 			if (addr.ea <= lim)
726 				goto bad;
727 			lim = desc.d ? 0xffffffff : 0xffff;
728 		}
729 		if (addr.ea > lim)
730 			goto bad;
731 		if (lim == 0xffffffff)
732 			*max_size = ~0u;
733 		else {
734 			*max_size = (u64)lim + 1 - addr.ea;
735 			if (size > *max_size)
736 				goto bad;
737 		}
738 		break;
739 	}
740 	if (la & (insn_alignment(ctxt, size) - 1))
741 		return emulate_gp(ctxt, 0);
742 	return X86EMUL_CONTINUE;
743 bad:
744 	if (addr.seg == VCPU_SREG_SS)
745 		return emulate_ss(ctxt, 0);
746 	else
747 		return emulate_gp(ctxt, 0);
748 }
749 
750 static int linearize(struct x86_emulate_ctxt *ctxt,
751 		     struct segmented_address addr,
752 		     unsigned size, bool write,
753 		     ulong *linear)
754 {
755 	unsigned max_size;
756 	return __linearize(ctxt, addr, &max_size, size, write, false,
757 			   ctxt->mode, linear);
758 }
759 
760 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
761 			     enum x86emul_mode mode)
762 {
763 	ulong linear;
764 	int rc;
765 	unsigned max_size;
766 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
767 					   .ea = dst };
768 
769 	if (ctxt->op_bytes != sizeof(unsigned long))
770 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
771 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
772 	if (rc == X86EMUL_CONTINUE)
773 		ctxt->_eip = addr.ea;
774 	return rc;
775 }
776 
777 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
778 {
779 	return assign_eip(ctxt, dst, ctxt->mode);
780 }
781 
782 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
783 			  const struct desc_struct *cs_desc)
784 {
785 	enum x86emul_mode mode = ctxt->mode;
786 	int rc;
787 
788 #ifdef CONFIG_X86_64
789 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
790 		if (cs_desc->l) {
791 			u64 efer = 0;
792 
793 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
794 			if (efer & EFER_LMA)
795 				mode = X86EMUL_MODE_PROT64;
796 		} else
797 			mode = X86EMUL_MODE_PROT32; /* temporary value */
798 	}
799 #endif
800 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
801 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
802 	rc = assign_eip(ctxt, dst, mode);
803 	if (rc == X86EMUL_CONTINUE)
804 		ctxt->mode = mode;
805 	return rc;
806 }
807 
808 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
809 {
810 	return assign_eip_near(ctxt, ctxt->_eip + rel);
811 }
812 
813 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear,
814 			      void *data, unsigned size)
815 {
816 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true);
817 }
818 
819 static int linear_write_system(struct x86_emulate_ctxt *ctxt,
820 			       ulong linear, void *data,
821 			       unsigned int size)
822 {
823 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true);
824 }
825 
826 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
827 			      struct segmented_address addr,
828 			      void *data,
829 			      unsigned size)
830 {
831 	int rc;
832 	ulong linear;
833 
834 	rc = linearize(ctxt, addr, size, false, &linear);
835 	if (rc != X86EMUL_CONTINUE)
836 		return rc;
837 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false);
838 }
839 
840 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
841 			       struct segmented_address addr,
842 			       void *data,
843 			       unsigned int size)
844 {
845 	int rc;
846 	ulong linear;
847 
848 	rc = linearize(ctxt, addr, size, true, &linear);
849 	if (rc != X86EMUL_CONTINUE)
850 		return rc;
851 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false);
852 }
853 
854 /*
855  * Prefetch the remaining bytes of the instruction without crossing page
856  * boundary if they are not in fetch_cache yet.
857  */
858 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
859 {
860 	int rc;
861 	unsigned size, max_size;
862 	unsigned long linear;
863 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
864 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
865 					   .ea = ctxt->eip + cur_size };
866 
867 	/*
868 	 * We do not know exactly how many bytes will be needed, and
869 	 * __linearize is expensive, so fetch as much as possible.  We
870 	 * just have to avoid going beyond the 15 byte limit, the end
871 	 * of the segment, or the end of the page.
872 	 *
873 	 * __linearize is called with size 0 so that it does not do any
874 	 * boundary check itself.  Instead, we use max_size to check
875 	 * against op_size.
876 	 */
877 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
878 			 &linear);
879 	if (unlikely(rc != X86EMUL_CONTINUE))
880 		return rc;
881 
882 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
883 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
884 
885 	/*
886 	 * One instruction can only straddle two pages,
887 	 * and one has been loaded at the beginning of
888 	 * x86_decode_insn.  So, if not enough bytes
889 	 * still, we must have hit the 15-byte boundary.
890 	 */
891 	if (unlikely(size < op_size))
892 		return emulate_gp(ctxt, 0);
893 
894 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
895 			      size, &ctxt->exception);
896 	if (unlikely(rc != X86EMUL_CONTINUE))
897 		return rc;
898 	ctxt->fetch.end += size;
899 	return X86EMUL_CONTINUE;
900 }
901 
902 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
903 					       unsigned size)
904 {
905 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
906 
907 	if (unlikely(done_size < size))
908 		return __do_insn_fetch_bytes(ctxt, size - done_size);
909 	else
910 		return X86EMUL_CONTINUE;
911 }
912 
913 /* Fetch next part of the instruction being emulated. */
914 #define insn_fetch(_type, _ctxt)					\
915 ({	_type _x;							\
916 									\
917 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
918 	if (rc != X86EMUL_CONTINUE)					\
919 		goto done;						\
920 	ctxt->_eip += sizeof(_type);					\
921 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
922 	ctxt->fetch.ptr += sizeof(_type);				\
923 	_x;								\
924 })
925 
926 #define insn_fetch_arr(_arr, _size, _ctxt)				\
927 ({									\
928 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
929 	if (rc != X86EMUL_CONTINUE)					\
930 		goto done;						\
931 	ctxt->_eip += (_size);						\
932 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
933 	ctxt->fetch.ptr += (_size);					\
934 })
935 
936 /*
937  * Given the 'reg' portion of a ModRM byte, and a register block, return a
938  * pointer into the block that addresses the relevant register.
939  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
940  */
941 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
942 			     int byteop)
943 {
944 	void *p;
945 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
946 
947 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
948 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
949 	else
950 		p = reg_rmw(ctxt, modrm_reg);
951 	return p;
952 }
953 
954 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
955 			   struct segmented_address addr,
956 			   u16 *size, unsigned long *address, int op_bytes)
957 {
958 	int rc;
959 
960 	if (op_bytes == 2)
961 		op_bytes = 3;
962 	*address = 0;
963 	rc = segmented_read_std(ctxt, addr, size, 2);
964 	if (rc != X86EMUL_CONTINUE)
965 		return rc;
966 	addr.ea += 2;
967 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
968 	return rc;
969 }
970 
971 FASTOP2(add);
972 FASTOP2(or);
973 FASTOP2(adc);
974 FASTOP2(sbb);
975 FASTOP2(and);
976 FASTOP2(sub);
977 FASTOP2(xor);
978 FASTOP2(cmp);
979 FASTOP2(test);
980 
981 FASTOP1SRC2(mul, mul_ex);
982 FASTOP1SRC2(imul, imul_ex);
983 FASTOP1SRC2EX(div, div_ex);
984 FASTOP1SRC2EX(idiv, idiv_ex);
985 
986 FASTOP3WCL(shld);
987 FASTOP3WCL(shrd);
988 
989 FASTOP2W(imul);
990 
991 FASTOP1(not);
992 FASTOP1(neg);
993 FASTOP1(inc);
994 FASTOP1(dec);
995 
996 FASTOP2CL(rol);
997 FASTOP2CL(ror);
998 FASTOP2CL(rcl);
999 FASTOP2CL(rcr);
1000 FASTOP2CL(shl);
1001 FASTOP2CL(shr);
1002 FASTOP2CL(sar);
1003 
1004 FASTOP2W(bsf);
1005 FASTOP2W(bsr);
1006 FASTOP2W(bt);
1007 FASTOP2W(bts);
1008 FASTOP2W(btr);
1009 FASTOP2W(btc);
1010 
1011 FASTOP2(xadd);
1012 
1013 FASTOP2R(cmp, cmp_r);
1014 
1015 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1016 {
1017 	/* If src is zero, do not writeback, but update flags */
1018 	if (ctxt->src.val == 0)
1019 		ctxt->dst.type = OP_NONE;
1020 	return fastop(ctxt, em_bsf);
1021 }
1022 
1023 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1024 {
1025 	/* If src is zero, do not writeback, but update flags */
1026 	if (ctxt->src.val == 0)
1027 		ctxt->dst.type = OP_NONE;
1028 	return fastop(ctxt, em_bsr);
1029 }
1030 
1031 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1032 {
1033 	u8 rc;
1034 	void (*fop)(void) = (void *)em_setcc + 4 * (condition & 0xf);
1035 
1036 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1037 	asm("push %[flags]; popf; " CALL_NOSPEC
1038 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1039 	return rc;
1040 }
1041 
1042 static void fetch_register_operand(struct operand *op)
1043 {
1044 	switch (op->bytes) {
1045 	case 1:
1046 		op->val = *(u8 *)op->addr.reg;
1047 		break;
1048 	case 2:
1049 		op->val = *(u16 *)op->addr.reg;
1050 		break;
1051 	case 4:
1052 		op->val = *(u32 *)op->addr.reg;
1053 		break;
1054 	case 8:
1055 		op->val = *(u64 *)op->addr.reg;
1056 		break;
1057 	}
1058 }
1059 
1060 static void read_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data, int reg)
1061 {
1062 	switch (reg) {
1063 	case 0: asm("movdqa %%xmm0, %0" : "=m"(*data)); break;
1064 	case 1: asm("movdqa %%xmm1, %0" : "=m"(*data)); break;
1065 	case 2: asm("movdqa %%xmm2, %0" : "=m"(*data)); break;
1066 	case 3: asm("movdqa %%xmm3, %0" : "=m"(*data)); break;
1067 	case 4: asm("movdqa %%xmm4, %0" : "=m"(*data)); break;
1068 	case 5: asm("movdqa %%xmm5, %0" : "=m"(*data)); break;
1069 	case 6: asm("movdqa %%xmm6, %0" : "=m"(*data)); break;
1070 	case 7: asm("movdqa %%xmm7, %0" : "=m"(*data)); break;
1071 #ifdef CONFIG_X86_64
1072 	case 8: asm("movdqa %%xmm8, %0" : "=m"(*data)); break;
1073 	case 9: asm("movdqa %%xmm9, %0" : "=m"(*data)); break;
1074 	case 10: asm("movdqa %%xmm10, %0" : "=m"(*data)); break;
1075 	case 11: asm("movdqa %%xmm11, %0" : "=m"(*data)); break;
1076 	case 12: asm("movdqa %%xmm12, %0" : "=m"(*data)); break;
1077 	case 13: asm("movdqa %%xmm13, %0" : "=m"(*data)); break;
1078 	case 14: asm("movdqa %%xmm14, %0" : "=m"(*data)); break;
1079 	case 15: asm("movdqa %%xmm15, %0" : "=m"(*data)); break;
1080 #endif
1081 	default: BUG();
1082 	}
1083 }
1084 
1085 static void write_sse_reg(struct x86_emulate_ctxt *ctxt, sse128_t *data,
1086 			  int reg)
1087 {
1088 	switch (reg) {
1089 	case 0: asm("movdqa %0, %%xmm0" : : "m"(*data)); break;
1090 	case 1: asm("movdqa %0, %%xmm1" : : "m"(*data)); break;
1091 	case 2: asm("movdqa %0, %%xmm2" : : "m"(*data)); break;
1092 	case 3: asm("movdqa %0, %%xmm3" : : "m"(*data)); break;
1093 	case 4: asm("movdqa %0, %%xmm4" : : "m"(*data)); break;
1094 	case 5: asm("movdqa %0, %%xmm5" : : "m"(*data)); break;
1095 	case 6: asm("movdqa %0, %%xmm6" : : "m"(*data)); break;
1096 	case 7: asm("movdqa %0, %%xmm7" : : "m"(*data)); break;
1097 #ifdef CONFIG_X86_64
1098 	case 8: asm("movdqa %0, %%xmm8" : : "m"(*data)); break;
1099 	case 9: asm("movdqa %0, %%xmm9" : : "m"(*data)); break;
1100 	case 10: asm("movdqa %0, %%xmm10" : : "m"(*data)); break;
1101 	case 11: asm("movdqa %0, %%xmm11" : : "m"(*data)); break;
1102 	case 12: asm("movdqa %0, %%xmm12" : : "m"(*data)); break;
1103 	case 13: asm("movdqa %0, %%xmm13" : : "m"(*data)); break;
1104 	case 14: asm("movdqa %0, %%xmm14" : : "m"(*data)); break;
1105 	case 15: asm("movdqa %0, %%xmm15" : : "m"(*data)); break;
1106 #endif
1107 	default: BUG();
1108 	}
1109 }
1110 
1111 static void read_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1112 {
1113 	switch (reg) {
1114 	case 0: asm("movq %%mm0, %0" : "=m"(*data)); break;
1115 	case 1: asm("movq %%mm1, %0" : "=m"(*data)); break;
1116 	case 2: asm("movq %%mm2, %0" : "=m"(*data)); break;
1117 	case 3: asm("movq %%mm3, %0" : "=m"(*data)); break;
1118 	case 4: asm("movq %%mm4, %0" : "=m"(*data)); break;
1119 	case 5: asm("movq %%mm5, %0" : "=m"(*data)); break;
1120 	case 6: asm("movq %%mm6, %0" : "=m"(*data)); break;
1121 	case 7: asm("movq %%mm7, %0" : "=m"(*data)); break;
1122 	default: BUG();
1123 	}
1124 }
1125 
1126 static void write_mmx_reg(struct x86_emulate_ctxt *ctxt, u64 *data, int reg)
1127 {
1128 	switch (reg) {
1129 	case 0: asm("movq %0, %%mm0" : : "m"(*data)); break;
1130 	case 1: asm("movq %0, %%mm1" : : "m"(*data)); break;
1131 	case 2: asm("movq %0, %%mm2" : : "m"(*data)); break;
1132 	case 3: asm("movq %0, %%mm3" : : "m"(*data)); break;
1133 	case 4: asm("movq %0, %%mm4" : : "m"(*data)); break;
1134 	case 5: asm("movq %0, %%mm5" : : "m"(*data)); break;
1135 	case 6: asm("movq %0, %%mm6" : : "m"(*data)); break;
1136 	case 7: asm("movq %0, %%mm7" : : "m"(*data)); break;
1137 	default: BUG();
1138 	}
1139 }
1140 
1141 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1142 {
1143 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1144 		return emulate_nm(ctxt);
1145 
1146 	asm volatile("fninit");
1147 	return X86EMUL_CONTINUE;
1148 }
1149 
1150 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1151 {
1152 	u16 fcw;
1153 
1154 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1155 		return emulate_nm(ctxt);
1156 
1157 	asm volatile("fnstcw %0": "+m"(fcw));
1158 
1159 	ctxt->dst.val = fcw;
1160 
1161 	return X86EMUL_CONTINUE;
1162 }
1163 
1164 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1165 {
1166 	u16 fsw;
1167 
1168 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1169 		return emulate_nm(ctxt);
1170 
1171 	asm volatile("fnstsw %0": "+m"(fsw));
1172 
1173 	ctxt->dst.val = fsw;
1174 
1175 	return X86EMUL_CONTINUE;
1176 }
1177 
1178 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1179 				    struct operand *op)
1180 {
1181 	unsigned reg = ctxt->modrm_reg;
1182 
1183 	if (!(ctxt->d & ModRM))
1184 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1185 
1186 	if (ctxt->d & Sse) {
1187 		op->type = OP_XMM;
1188 		op->bytes = 16;
1189 		op->addr.xmm = reg;
1190 		read_sse_reg(ctxt, &op->vec_val, reg);
1191 		return;
1192 	}
1193 	if (ctxt->d & Mmx) {
1194 		reg &= 7;
1195 		op->type = OP_MM;
1196 		op->bytes = 8;
1197 		op->addr.mm = reg;
1198 		return;
1199 	}
1200 
1201 	op->type = OP_REG;
1202 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1203 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1204 
1205 	fetch_register_operand(op);
1206 	op->orig_val = op->val;
1207 }
1208 
1209 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1210 {
1211 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1212 		ctxt->modrm_seg = VCPU_SREG_SS;
1213 }
1214 
1215 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1216 			struct operand *op)
1217 {
1218 	u8 sib;
1219 	int index_reg, base_reg, scale;
1220 	int rc = X86EMUL_CONTINUE;
1221 	ulong modrm_ea = 0;
1222 
1223 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1224 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1225 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1226 
1227 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1228 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1229 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1230 	ctxt->modrm_seg = VCPU_SREG_DS;
1231 
1232 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1233 		op->type = OP_REG;
1234 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1235 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1236 				ctxt->d & ByteOp);
1237 		if (ctxt->d & Sse) {
1238 			op->type = OP_XMM;
1239 			op->bytes = 16;
1240 			op->addr.xmm = ctxt->modrm_rm;
1241 			read_sse_reg(ctxt, &op->vec_val, ctxt->modrm_rm);
1242 			return rc;
1243 		}
1244 		if (ctxt->d & Mmx) {
1245 			op->type = OP_MM;
1246 			op->bytes = 8;
1247 			op->addr.mm = ctxt->modrm_rm & 7;
1248 			return rc;
1249 		}
1250 		fetch_register_operand(op);
1251 		return rc;
1252 	}
1253 
1254 	op->type = OP_MEM;
1255 
1256 	if (ctxt->ad_bytes == 2) {
1257 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1258 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1259 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1260 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1261 
1262 		/* 16-bit ModR/M decode. */
1263 		switch (ctxt->modrm_mod) {
1264 		case 0:
1265 			if (ctxt->modrm_rm == 6)
1266 				modrm_ea += insn_fetch(u16, ctxt);
1267 			break;
1268 		case 1:
1269 			modrm_ea += insn_fetch(s8, ctxt);
1270 			break;
1271 		case 2:
1272 			modrm_ea += insn_fetch(u16, ctxt);
1273 			break;
1274 		}
1275 		switch (ctxt->modrm_rm) {
1276 		case 0:
1277 			modrm_ea += bx + si;
1278 			break;
1279 		case 1:
1280 			modrm_ea += bx + di;
1281 			break;
1282 		case 2:
1283 			modrm_ea += bp + si;
1284 			break;
1285 		case 3:
1286 			modrm_ea += bp + di;
1287 			break;
1288 		case 4:
1289 			modrm_ea += si;
1290 			break;
1291 		case 5:
1292 			modrm_ea += di;
1293 			break;
1294 		case 6:
1295 			if (ctxt->modrm_mod != 0)
1296 				modrm_ea += bp;
1297 			break;
1298 		case 7:
1299 			modrm_ea += bx;
1300 			break;
1301 		}
1302 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1303 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1304 			ctxt->modrm_seg = VCPU_SREG_SS;
1305 		modrm_ea = (u16)modrm_ea;
1306 	} else {
1307 		/* 32/64-bit ModR/M decode. */
1308 		if ((ctxt->modrm_rm & 7) == 4) {
1309 			sib = insn_fetch(u8, ctxt);
1310 			index_reg |= (sib >> 3) & 7;
1311 			base_reg |= sib & 7;
1312 			scale = sib >> 6;
1313 
1314 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1315 				modrm_ea += insn_fetch(s32, ctxt);
1316 			else {
1317 				modrm_ea += reg_read(ctxt, base_reg);
1318 				adjust_modrm_seg(ctxt, base_reg);
1319 				/* Increment ESP on POP [ESP] */
1320 				if ((ctxt->d & IncSP) &&
1321 				    base_reg == VCPU_REGS_RSP)
1322 					modrm_ea += ctxt->op_bytes;
1323 			}
1324 			if (index_reg != 4)
1325 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1326 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1327 			modrm_ea += insn_fetch(s32, ctxt);
1328 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1329 				ctxt->rip_relative = 1;
1330 		} else {
1331 			base_reg = ctxt->modrm_rm;
1332 			modrm_ea += reg_read(ctxt, base_reg);
1333 			adjust_modrm_seg(ctxt, base_reg);
1334 		}
1335 		switch (ctxt->modrm_mod) {
1336 		case 1:
1337 			modrm_ea += insn_fetch(s8, ctxt);
1338 			break;
1339 		case 2:
1340 			modrm_ea += insn_fetch(s32, ctxt);
1341 			break;
1342 		}
1343 	}
1344 	op->addr.mem.ea = modrm_ea;
1345 	if (ctxt->ad_bytes != 8)
1346 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1347 
1348 done:
1349 	return rc;
1350 }
1351 
1352 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1353 		      struct operand *op)
1354 {
1355 	int rc = X86EMUL_CONTINUE;
1356 
1357 	op->type = OP_MEM;
1358 	switch (ctxt->ad_bytes) {
1359 	case 2:
1360 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1361 		break;
1362 	case 4:
1363 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1364 		break;
1365 	case 8:
1366 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1367 		break;
1368 	}
1369 done:
1370 	return rc;
1371 }
1372 
1373 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1374 {
1375 	long sv = 0, mask;
1376 
1377 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1378 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1379 
1380 		if (ctxt->src.bytes == 2)
1381 			sv = (s16)ctxt->src.val & (s16)mask;
1382 		else if (ctxt->src.bytes == 4)
1383 			sv = (s32)ctxt->src.val & (s32)mask;
1384 		else
1385 			sv = (s64)ctxt->src.val & (s64)mask;
1386 
1387 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1388 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1389 	}
1390 
1391 	/* only subword offset */
1392 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1393 }
1394 
1395 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1396 			 unsigned long addr, void *dest, unsigned size)
1397 {
1398 	int rc;
1399 	struct read_cache *mc = &ctxt->mem_read;
1400 
1401 	if (mc->pos < mc->end)
1402 		goto read_cached;
1403 
1404 	WARN_ON((mc->end + size) >= sizeof(mc->data));
1405 
1406 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1407 				      &ctxt->exception);
1408 	if (rc != X86EMUL_CONTINUE)
1409 		return rc;
1410 
1411 	mc->end += size;
1412 
1413 read_cached:
1414 	memcpy(dest, mc->data + mc->pos, size);
1415 	mc->pos += size;
1416 	return X86EMUL_CONTINUE;
1417 }
1418 
1419 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1420 			  struct segmented_address addr,
1421 			  void *data,
1422 			  unsigned size)
1423 {
1424 	int rc;
1425 	ulong linear;
1426 
1427 	rc = linearize(ctxt, addr, size, false, &linear);
1428 	if (rc != X86EMUL_CONTINUE)
1429 		return rc;
1430 	return read_emulated(ctxt, linear, data, size);
1431 }
1432 
1433 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1434 			   struct segmented_address addr,
1435 			   const void *data,
1436 			   unsigned size)
1437 {
1438 	int rc;
1439 	ulong linear;
1440 
1441 	rc = linearize(ctxt, addr, size, true, &linear);
1442 	if (rc != X86EMUL_CONTINUE)
1443 		return rc;
1444 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1445 					 &ctxt->exception);
1446 }
1447 
1448 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1449 			     struct segmented_address addr,
1450 			     const void *orig_data, const void *data,
1451 			     unsigned size)
1452 {
1453 	int rc;
1454 	ulong linear;
1455 
1456 	rc = linearize(ctxt, addr, size, true, &linear);
1457 	if (rc != X86EMUL_CONTINUE)
1458 		return rc;
1459 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1460 					   size, &ctxt->exception);
1461 }
1462 
1463 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1464 			   unsigned int size, unsigned short port,
1465 			   void *dest)
1466 {
1467 	struct read_cache *rc = &ctxt->io_read;
1468 
1469 	if (rc->pos == rc->end) { /* refill pio read ahead */
1470 		unsigned int in_page, n;
1471 		unsigned int count = ctxt->rep_prefix ?
1472 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1473 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1474 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1475 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1476 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1477 		if (n == 0)
1478 			n = 1;
1479 		rc->pos = rc->end = 0;
1480 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1481 			return 0;
1482 		rc->end = n * size;
1483 	}
1484 
1485 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1486 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1487 		ctxt->dst.data = rc->data + rc->pos;
1488 		ctxt->dst.type = OP_MEM_STR;
1489 		ctxt->dst.count = (rc->end - rc->pos) / size;
1490 		rc->pos = rc->end;
1491 	} else {
1492 		memcpy(dest, rc->data + rc->pos, size);
1493 		rc->pos += size;
1494 	}
1495 	return 1;
1496 }
1497 
1498 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1499 				     u16 index, struct desc_struct *desc)
1500 {
1501 	struct desc_ptr dt;
1502 	ulong addr;
1503 
1504 	ctxt->ops->get_idt(ctxt, &dt);
1505 
1506 	if (dt.size < index * 8 + 7)
1507 		return emulate_gp(ctxt, index << 3 | 0x2);
1508 
1509 	addr = dt.address + index * 8;
1510 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1511 }
1512 
1513 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1514 				     u16 selector, struct desc_ptr *dt)
1515 {
1516 	const struct x86_emulate_ops *ops = ctxt->ops;
1517 	u32 base3 = 0;
1518 
1519 	if (selector & 1 << 2) {
1520 		struct desc_struct desc;
1521 		u16 sel;
1522 
1523 		memset(dt, 0, sizeof(*dt));
1524 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1525 				      VCPU_SREG_LDTR))
1526 			return;
1527 
1528 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1529 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1530 	} else
1531 		ops->get_gdt(ctxt, dt);
1532 }
1533 
1534 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1535 			      u16 selector, ulong *desc_addr_p)
1536 {
1537 	struct desc_ptr dt;
1538 	u16 index = selector >> 3;
1539 	ulong addr;
1540 
1541 	get_descriptor_table_ptr(ctxt, selector, &dt);
1542 
1543 	if (dt.size < index * 8 + 7)
1544 		return emulate_gp(ctxt, selector & 0xfffc);
1545 
1546 	addr = dt.address + index * 8;
1547 
1548 #ifdef CONFIG_X86_64
1549 	if (addr >> 32 != 0) {
1550 		u64 efer = 0;
1551 
1552 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1553 		if (!(efer & EFER_LMA))
1554 			addr &= (u32)-1;
1555 	}
1556 #endif
1557 
1558 	*desc_addr_p = addr;
1559 	return X86EMUL_CONTINUE;
1560 }
1561 
1562 /* allowed just for 8 bytes segments */
1563 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1564 				   u16 selector, struct desc_struct *desc,
1565 				   ulong *desc_addr_p)
1566 {
1567 	int rc;
1568 
1569 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1570 	if (rc != X86EMUL_CONTINUE)
1571 		return rc;
1572 
1573 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1574 }
1575 
1576 /* allowed just for 8 bytes segments */
1577 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1578 				    u16 selector, struct desc_struct *desc)
1579 {
1580 	int rc;
1581 	ulong addr;
1582 
1583 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1584 	if (rc != X86EMUL_CONTINUE)
1585 		return rc;
1586 
1587 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1588 }
1589 
1590 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1591 				     u16 selector, int seg, u8 cpl,
1592 				     enum x86_transfer_type transfer,
1593 				     struct desc_struct *desc)
1594 {
1595 	struct desc_struct seg_desc, old_desc;
1596 	u8 dpl, rpl;
1597 	unsigned err_vec = GP_VECTOR;
1598 	u32 err_code = 0;
1599 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1600 	ulong desc_addr;
1601 	int ret;
1602 	u16 dummy;
1603 	u32 base3 = 0;
1604 
1605 	memset(&seg_desc, 0, sizeof(seg_desc));
1606 
1607 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1608 		/* set real mode segment descriptor (keep limit etc. for
1609 		 * unreal mode) */
1610 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1611 		set_desc_base(&seg_desc, selector << 4);
1612 		goto load;
1613 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1614 		/* VM86 needs a clean new segment descriptor */
1615 		set_desc_base(&seg_desc, selector << 4);
1616 		set_desc_limit(&seg_desc, 0xffff);
1617 		seg_desc.type = 3;
1618 		seg_desc.p = 1;
1619 		seg_desc.s = 1;
1620 		seg_desc.dpl = 3;
1621 		goto load;
1622 	}
1623 
1624 	rpl = selector & 3;
1625 
1626 	/* TR should be in GDT only */
1627 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1628 		goto exception;
1629 
1630 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1631 	if (null_selector) {
1632 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1633 			goto exception;
1634 
1635 		if (seg == VCPU_SREG_SS) {
1636 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1637 				goto exception;
1638 
1639 			/*
1640 			 * ctxt->ops->set_segment expects the CPL to be in
1641 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1642 			 */
1643 			seg_desc.type = 3;
1644 			seg_desc.p = 1;
1645 			seg_desc.s = 1;
1646 			seg_desc.dpl = cpl;
1647 			seg_desc.d = 1;
1648 			seg_desc.g = 1;
1649 		}
1650 
1651 		/* Skip all following checks */
1652 		goto load;
1653 	}
1654 
1655 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1656 	if (ret != X86EMUL_CONTINUE)
1657 		return ret;
1658 
1659 	err_code = selector & 0xfffc;
1660 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1661 							   GP_VECTOR;
1662 
1663 	/* can't load system descriptor into segment selector */
1664 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1665 		if (transfer == X86_TRANSFER_CALL_JMP)
1666 			return X86EMUL_UNHANDLEABLE;
1667 		goto exception;
1668 	}
1669 
1670 	if (!seg_desc.p) {
1671 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1672 		goto exception;
1673 	}
1674 
1675 	dpl = seg_desc.dpl;
1676 
1677 	switch (seg) {
1678 	case VCPU_SREG_SS:
1679 		/*
1680 		 * segment is not a writable data segment or segment
1681 		 * selector's RPL != CPL or segment selector's RPL != CPL
1682 		 */
1683 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1684 			goto exception;
1685 		break;
1686 	case VCPU_SREG_CS:
1687 		if (!(seg_desc.type & 8))
1688 			goto exception;
1689 
1690 		if (seg_desc.type & 4) {
1691 			/* conforming */
1692 			if (dpl > cpl)
1693 				goto exception;
1694 		} else {
1695 			/* nonconforming */
1696 			if (rpl > cpl || dpl != cpl)
1697 				goto exception;
1698 		}
1699 		/* in long-mode d/b must be clear if l is set */
1700 		if (seg_desc.d && seg_desc.l) {
1701 			u64 efer = 0;
1702 
1703 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1704 			if (efer & EFER_LMA)
1705 				goto exception;
1706 		}
1707 
1708 		/* CS(RPL) <- CPL */
1709 		selector = (selector & 0xfffc) | cpl;
1710 		break;
1711 	case VCPU_SREG_TR:
1712 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1713 			goto exception;
1714 		old_desc = seg_desc;
1715 		seg_desc.type |= 2; /* busy */
1716 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1717 						  sizeof(seg_desc), &ctxt->exception);
1718 		if (ret != X86EMUL_CONTINUE)
1719 			return ret;
1720 		break;
1721 	case VCPU_SREG_LDTR:
1722 		if (seg_desc.s || seg_desc.type != 2)
1723 			goto exception;
1724 		break;
1725 	default: /*  DS, ES, FS, or GS */
1726 		/*
1727 		 * segment is not a data or readable code segment or
1728 		 * ((segment is a data or nonconforming code segment)
1729 		 * and (both RPL and CPL > DPL))
1730 		 */
1731 		if ((seg_desc.type & 0xa) == 0x8 ||
1732 		    (((seg_desc.type & 0xc) != 0xc) &&
1733 		     (rpl > dpl && cpl > dpl)))
1734 			goto exception;
1735 		break;
1736 	}
1737 
1738 	if (seg_desc.s) {
1739 		/* mark segment as accessed */
1740 		if (!(seg_desc.type & 1)) {
1741 			seg_desc.type |= 1;
1742 			ret = write_segment_descriptor(ctxt, selector,
1743 						       &seg_desc);
1744 			if (ret != X86EMUL_CONTINUE)
1745 				return ret;
1746 		}
1747 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1748 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1749 		if (ret != X86EMUL_CONTINUE)
1750 			return ret;
1751 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1752 				((u64)base3 << 32), ctxt))
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 = linear_read_system(ctxt, cs_addr, &cs, 2);
2062 	if (rc != X86EMUL_CONTINUE)
2063 		return rc;
2064 
2065 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
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 #ifdef CONFIG_X86_64
2333 	u32 eax, ebx, ecx, edx;
2334 
2335 	eax = 0x80000001;
2336 	ecx = 0;
2337 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
2338 	return edx & bit(X86_FEATURE_LM);
2339 #else
2340 	return false;
2341 #endif
2342 }
2343 
2344 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2345 {
2346 	desc->g    = (flags >> 23) & 1;
2347 	desc->d    = (flags >> 22) & 1;
2348 	desc->l    = (flags >> 21) & 1;
2349 	desc->avl  = (flags >> 20) & 1;
2350 	desc->p    = (flags >> 15) & 1;
2351 	desc->dpl  = (flags >> 13) & 3;
2352 	desc->s    = (flags >> 12) & 1;
2353 	desc->type = (flags >>  8) & 15;
2354 }
2355 
2356 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, const char *smstate,
2357 			   int n)
2358 {
2359 	struct desc_struct desc;
2360 	int offset;
2361 	u16 selector;
2362 
2363 	selector = GET_SMSTATE(u32, smstate, 0x7fa8 + n * 4);
2364 
2365 	if (n < 3)
2366 		offset = 0x7f84 + n * 12;
2367 	else
2368 		offset = 0x7f2c + (n - 3) * 12;
2369 
2370 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2371 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2372 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, offset));
2373 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2374 	return X86EMUL_CONTINUE;
2375 }
2376 
2377 #ifdef CONFIG_X86_64
2378 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, const char *smstate,
2379 			   int n)
2380 {
2381 	struct desc_struct desc;
2382 	int offset;
2383 	u16 selector;
2384 	u32 base3;
2385 
2386 	offset = 0x7e00 + n * 16;
2387 
2388 	selector =                GET_SMSTATE(u16, smstate, offset);
2389 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smstate, offset + 2) << 8);
2390 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2391 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2392 	base3 =                   GET_SMSTATE(u32, smstate, offset + 12);
2393 
2394 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2395 	return X86EMUL_CONTINUE;
2396 }
2397 #endif
2398 
2399 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2400 				    u64 cr0, u64 cr3, u64 cr4)
2401 {
2402 	int bad;
2403 	u64 pcid;
2404 
2405 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
2406 	pcid = 0;
2407 	if (cr4 & X86_CR4_PCIDE) {
2408 		pcid = cr3 & 0xfff;
2409 		cr3 &= ~0xfff;
2410 	}
2411 
2412 	bad = ctxt->ops->set_cr(ctxt, 3, cr3);
2413 	if (bad)
2414 		return X86EMUL_UNHANDLEABLE;
2415 
2416 	/*
2417 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2418 	 * Then enable protected mode.	However, PCID cannot be enabled
2419 	 * if EFER.LMA=0, so set it separately.
2420 	 */
2421 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2422 	if (bad)
2423 		return X86EMUL_UNHANDLEABLE;
2424 
2425 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2426 	if (bad)
2427 		return X86EMUL_UNHANDLEABLE;
2428 
2429 	if (cr4 & X86_CR4_PCIDE) {
2430 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2431 		if (bad)
2432 			return X86EMUL_UNHANDLEABLE;
2433 		if (pcid) {
2434 			bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid);
2435 			if (bad)
2436 				return X86EMUL_UNHANDLEABLE;
2437 		}
2438 
2439 	}
2440 
2441 	return X86EMUL_CONTINUE;
2442 }
2443 
2444 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
2445 			     const char *smstate)
2446 {
2447 	struct desc_struct desc;
2448 	struct desc_ptr dt;
2449 	u16 selector;
2450 	u32 val, cr0, cr3, cr4;
2451 	int i;
2452 
2453 	cr0 =                      GET_SMSTATE(u32, smstate, 0x7ffc);
2454 	cr3 =                      GET_SMSTATE(u32, smstate, 0x7ff8);
2455 	ctxt->eflags =             GET_SMSTATE(u32, smstate, 0x7ff4) | X86_EFLAGS_FIXED;
2456 	ctxt->_eip =               GET_SMSTATE(u32, smstate, 0x7ff0);
2457 
2458 	for (i = 0; i < 8; i++)
2459 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smstate, 0x7fd0 + i * 4);
2460 
2461 	val = GET_SMSTATE(u32, smstate, 0x7fcc);
2462 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2463 	val = GET_SMSTATE(u32, smstate, 0x7fc8);
2464 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2465 
2466 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc4);
2467 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f64));
2468 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f60));
2469 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f5c));
2470 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2471 
2472 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc0);
2473 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f80));
2474 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f7c));
2475 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f78));
2476 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2477 
2478 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f74);
2479 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f70);
2480 	ctxt->ops->set_gdt(ctxt, &dt);
2481 
2482 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f58);
2483 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f54);
2484 	ctxt->ops->set_idt(ctxt, &dt);
2485 
2486 	for (i = 0; i < 6; i++) {
2487 		int r = rsm_load_seg_32(ctxt, smstate, i);
2488 		if (r != X86EMUL_CONTINUE)
2489 			return r;
2490 	}
2491 
2492 	cr4 = GET_SMSTATE(u32, smstate, 0x7f14);
2493 
2494 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7ef8));
2495 
2496 	return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2497 }
2498 
2499 #ifdef CONFIG_X86_64
2500 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
2501 			     const char *smstate)
2502 {
2503 	struct desc_struct desc;
2504 	struct desc_ptr dt;
2505 	u64 val, cr0, cr3, cr4;
2506 	u32 base3;
2507 	u16 selector;
2508 	int i, r;
2509 
2510 	for (i = 0; i < 16; i++)
2511 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smstate, 0x7ff8 - i * 8);
2512 
2513 	ctxt->_eip   = GET_SMSTATE(u64, smstate, 0x7f78);
2514 	ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7f70) | X86_EFLAGS_FIXED;
2515 
2516 	val = GET_SMSTATE(u32, smstate, 0x7f68);
2517 	ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1);
2518 	val = GET_SMSTATE(u32, smstate, 0x7f60);
2519 	ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1);
2520 
2521 	cr0 =                       GET_SMSTATE(u64, smstate, 0x7f58);
2522 	cr3 =                       GET_SMSTATE(u64, smstate, 0x7f50);
2523 	cr4 =                       GET_SMSTATE(u64, smstate, 0x7f48);
2524 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7f00));
2525 	val =                       GET_SMSTATE(u64, smstate, 0x7ed0);
2526 	ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA);
2527 
2528 	selector =                  GET_SMSTATE(u32, smstate, 0x7e90);
2529 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e92) << 8);
2530 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e94));
2531 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e98));
2532 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e9c);
2533 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2534 
2535 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e84);
2536 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e88);
2537 	ctxt->ops->set_idt(ctxt, &dt);
2538 
2539 	selector =                  GET_SMSTATE(u32, smstate, 0x7e70);
2540 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e72) << 8);
2541 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e74));
2542 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e78));
2543 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e7c);
2544 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2545 
2546 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e64);
2547 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e68);
2548 	ctxt->ops->set_gdt(ctxt, &dt);
2549 
2550 	r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2551 	if (r != X86EMUL_CONTINUE)
2552 		return r;
2553 
2554 	for (i = 0; i < 6; i++) {
2555 		r = rsm_load_seg_64(ctxt, smstate, i);
2556 		if (r != X86EMUL_CONTINUE)
2557 			return r;
2558 	}
2559 
2560 	return X86EMUL_CONTINUE;
2561 }
2562 #endif
2563 
2564 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2565 {
2566 	unsigned long cr0, cr4, efer;
2567 	char buf[512];
2568 	u64 smbase;
2569 	int ret;
2570 
2571 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2572 		return emulate_ud(ctxt);
2573 
2574 	smbase = ctxt->ops->get_smbase(ctxt);
2575 
2576 	ret = ctxt->ops->read_phys(ctxt, smbase + 0xfe00, buf, sizeof(buf));
2577 	if (ret != X86EMUL_CONTINUE)
2578 		return X86EMUL_UNHANDLEABLE;
2579 
2580 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2581 		ctxt->ops->set_nmi_mask(ctxt, false);
2582 
2583 	ctxt->ops->set_hflags(ctxt, ctxt->ops->get_hflags(ctxt) &
2584 		~(X86EMUL_SMM_INSIDE_NMI_MASK | X86EMUL_SMM_MASK));
2585 
2586 	/*
2587 	 * Get back to real mode, to prepare a safe state in which to load
2588 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2589 	 * supports long mode.
2590 	 */
2591 	if (emulator_has_longmode(ctxt)) {
2592 		struct desc_struct cs_desc;
2593 
2594 		/* Zero CR4.PCIDE before CR0.PG.  */
2595 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2596 		if (cr4 & X86_CR4_PCIDE)
2597 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2598 
2599 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2600 		memset(&cs_desc, 0, sizeof(cs_desc));
2601 		cs_desc.type = 0xb;
2602 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2603 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2604 	}
2605 
2606 	/* For the 64-bit case, this will clear EFER.LMA.  */
2607 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2608 	if (cr0 & X86_CR0_PE)
2609 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2610 
2611 	if (emulator_has_longmode(ctxt)) {
2612 		/* Clear CR4.PAE before clearing EFER.LME. */
2613 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2614 		if (cr4 & X86_CR4_PAE)
2615 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2616 
2617 		/* And finally go back to 32-bit mode.  */
2618 		efer = 0;
2619 		ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2620 	}
2621 
2622 	/*
2623 	 * Give pre_leave_smm() a chance to make ISA-specific changes to the
2624 	 * vCPU state (e.g. enter guest mode) before loading state from the SMM
2625 	 * state-save area.
2626 	 */
2627 	if (ctxt->ops->pre_leave_smm(ctxt, buf))
2628 		return X86EMUL_UNHANDLEABLE;
2629 
2630 #ifdef CONFIG_X86_64
2631 	if (emulator_has_longmode(ctxt))
2632 		ret = rsm_load_state_64(ctxt, buf);
2633 	else
2634 #endif
2635 		ret = rsm_load_state_32(ctxt, buf);
2636 
2637 	if (ret != X86EMUL_CONTINUE) {
2638 		/* FIXME: should triple fault */
2639 		return X86EMUL_UNHANDLEABLE;
2640 	}
2641 
2642 	ctxt->ops->post_leave_smm(ctxt);
2643 
2644 	return X86EMUL_CONTINUE;
2645 }
2646 
2647 static void
2648 setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
2649 			struct desc_struct *cs, struct desc_struct *ss)
2650 {
2651 	cs->l = 0;		/* will be adjusted later */
2652 	set_desc_base(cs, 0);	/* flat segment */
2653 	cs->g = 1;		/* 4kb granularity */
2654 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2655 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2656 	cs->s = 1;
2657 	cs->dpl = 0;		/* will be adjusted later */
2658 	cs->p = 1;
2659 	cs->d = 1;
2660 	cs->avl = 0;
2661 
2662 	set_desc_base(ss, 0);	/* flat segment */
2663 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2664 	ss->g = 1;		/* 4kb granularity */
2665 	ss->s = 1;
2666 	ss->type = 0x03;	/* Read/Write, Accessed */
2667 	ss->d = 1;		/* 32bit stack segment */
2668 	ss->dpl = 0;
2669 	ss->p = 1;
2670 	ss->l = 0;
2671 	ss->avl = 0;
2672 }
2673 
2674 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2675 {
2676 	u32 eax, ebx, ecx, edx;
2677 
2678 	eax = ecx = 0;
2679 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
2680 	return ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx
2681 		&& ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx
2682 		&& edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx;
2683 }
2684 
2685 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2686 {
2687 	const struct x86_emulate_ops *ops = ctxt->ops;
2688 	u32 eax, ebx, ecx, edx;
2689 
2690 	/*
2691 	 * syscall should always be enabled in longmode - so only become
2692 	 * vendor specific (cpuid) if other modes are active...
2693 	 */
2694 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2695 		return true;
2696 
2697 	eax = 0x00000000;
2698 	ecx = 0x00000000;
2699 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
2700 	/*
2701 	 * Intel ("GenuineIntel")
2702 	 * remark: Intel CPUs only support "syscall" in 64bit
2703 	 * longmode. Also an 64bit guest with a
2704 	 * 32bit compat-app running will #UD !! While this
2705 	 * behaviour can be fixed (by emulating) into AMD
2706 	 * response - CPUs of AMD can't behave like Intel.
2707 	 */
2708 	if (ebx == X86EMUL_CPUID_VENDOR_GenuineIntel_ebx &&
2709 	    ecx == X86EMUL_CPUID_VENDOR_GenuineIntel_ecx &&
2710 	    edx == X86EMUL_CPUID_VENDOR_GenuineIntel_edx)
2711 		return false;
2712 
2713 	/* AMD ("AuthenticAMD") */
2714 	if (ebx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ebx &&
2715 	    ecx == X86EMUL_CPUID_VENDOR_AuthenticAMD_ecx &&
2716 	    edx == X86EMUL_CPUID_VENDOR_AuthenticAMD_edx)
2717 		return true;
2718 
2719 	/* AMD ("AMDisbetter!") */
2720 	if (ebx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ebx &&
2721 	    ecx == X86EMUL_CPUID_VENDOR_AMDisbetterI_ecx &&
2722 	    edx == X86EMUL_CPUID_VENDOR_AMDisbetterI_edx)
2723 		return true;
2724 
2725 	/* Hygon ("HygonGenuine") */
2726 	if (ebx == X86EMUL_CPUID_VENDOR_HygonGenuine_ebx &&
2727 	    ecx == X86EMUL_CPUID_VENDOR_HygonGenuine_ecx &&
2728 	    edx == X86EMUL_CPUID_VENDOR_HygonGenuine_edx)
2729 		return true;
2730 
2731 	/*
2732 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2733 	 * stricter rules...
2734 	 */
2735 	return false;
2736 }
2737 
2738 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2739 {
2740 	const struct x86_emulate_ops *ops = ctxt->ops;
2741 	struct desc_struct cs, ss;
2742 	u64 msr_data;
2743 	u16 cs_sel, ss_sel;
2744 	u64 efer = 0;
2745 
2746 	/* syscall is not available in real mode */
2747 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2748 	    ctxt->mode == X86EMUL_MODE_VM86)
2749 		return emulate_ud(ctxt);
2750 
2751 	if (!(em_syscall_is_enabled(ctxt)))
2752 		return emulate_ud(ctxt);
2753 
2754 	ops->get_msr(ctxt, MSR_EFER, &efer);
2755 	setup_syscalls_segments(ctxt, &cs, &ss);
2756 
2757 	if (!(efer & EFER_SCE))
2758 		return emulate_ud(ctxt);
2759 
2760 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2761 	msr_data >>= 32;
2762 	cs_sel = (u16)(msr_data & 0xfffc);
2763 	ss_sel = (u16)(msr_data + 8);
2764 
2765 	if (efer & EFER_LMA) {
2766 		cs.d = 0;
2767 		cs.l = 1;
2768 	}
2769 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2770 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2771 
2772 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2773 	if (efer & EFER_LMA) {
2774 #ifdef CONFIG_X86_64
2775 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2776 
2777 		ops->get_msr(ctxt,
2778 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2779 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2780 		ctxt->_eip = msr_data;
2781 
2782 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2783 		ctxt->eflags &= ~msr_data;
2784 		ctxt->eflags |= X86_EFLAGS_FIXED;
2785 #endif
2786 	} else {
2787 		/* legacy mode */
2788 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2789 		ctxt->_eip = (u32)msr_data;
2790 
2791 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2792 	}
2793 
2794 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2795 	return X86EMUL_CONTINUE;
2796 }
2797 
2798 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2799 {
2800 	const struct x86_emulate_ops *ops = ctxt->ops;
2801 	struct desc_struct cs, ss;
2802 	u64 msr_data;
2803 	u16 cs_sel, ss_sel;
2804 	u64 efer = 0;
2805 
2806 	ops->get_msr(ctxt, MSR_EFER, &efer);
2807 	/* inject #GP if in real mode */
2808 	if (ctxt->mode == X86EMUL_MODE_REAL)
2809 		return emulate_gp(ctxt, 0);
2810 
2811 	/*
2812 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2813 	 * mode).
2814 	 */
2815 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2816 	    && !vendor_intel(ctxt))
2817 		return emulate_ud(ctxt);
2818 
2819 	/* sysenter/sysexit have not been tested in 64bit mode. */
2820 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2821 		return X86EMUL_UNHANDLEABLE;
2822 
2823 	setup_syscalls_segments(ctxt, &cs, &ss);
2824 
2825 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2826 	if ((msr_data & 0xfffc) == 0x0)
2827 		return emulate_gp(ctxt, 0);
2828 
2829 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2830 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2831 	ss_sel = cs_sel + 8;
2832 	if (efer & EFER_LMA) {
2833 		cs.d = 0;
2834 		cs.l = 1;
2835 	}
2836 
2837 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2838 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2839 
2840 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2841 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2842 
2843 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2844 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2845 							      (u32)msr_data;
2846 
2847 	return X86EMUL_CONTINUE;
2848 }
2849 
2850 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2851 {
2852 	const struct x86_emulate_ops *ops = ctxt->ops;
2853 	struct desc_struct cs, ss;
2854 	u64 msr_data, rcx, rdx;
2855 	int usermode;
2856 	u16 cs_sel = 0, ss_sel = 0;
2857 
2858 	/* inject #GP if in real mode or Virtual 8086 mode */
2859 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2860 	    ctxt->mode == X86EMUL_MODE_VM86)
2861 		return emulate_gp(ctxt, 0);
2862 
2863 	setup_syscalls_segments(ctxt, &cs, &ss);
2864 
2865 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2866 		usermode = X86EMUL_MODE_PROT64;
2867 	else
2868 		usermode = X86EMUL_MODE_PROT32;
2869 
2870 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2871 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2872 
2873 	cs.dpl = 3;
2874 	ss.dpl = 3;
2875 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2876 	switch (usermode) {
2877 	case X86EMUL_MODE_PROT32:
2878 		cs_sel = (u16)(msr_data + 16);
2879 		if ((msr_data & 0xfffc) == 0x0)
2880 			return emulate_gp(ctxt, 0);
2881 		ss_sel = (u16)(msr_data + 24);
2882 		rcx = (u32)rcx;
2883 		rdx = (u32)rdx;
2884 		break;
2885 	case X86EMUL_MODE_PROT64:
2886 		cs_sel = (u16)(msr_data + 32);
2887 		if (msr_data == 0x0)
2888 			return emulate_gp(ctxt, 0);
2889 		ss_sel = cs_sel + 8;
2890 		cs.d = 0;
2891 		cs.l = 1;
2892 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2893 		    emul_is_noncanonical_address(rdx, ctxt))
2894 			return emulate_gp(ctxt, 0);
2895 		break;
2896 	}
2897 	cs_sel |= SEGMENT_RPL_MASK;
2898 	ss_sel |= SEGMENT_RPL_MASK;
2899 
2900 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2901 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2902 
2903 	ctxt->_eip = rdx;
2904 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2905 
2906 	return X86EMUL_CONTINUE;
2907 }
2908 
2909 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2910 {
2911 	int iopl;
2912 	if (ctxt->mode == X86EMUL_MODE_REAL)
2913 		return false;
2914 	if (ctxt->mode == X86EMUL_MODE_VM86)
2915 		return true;
2916 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2917 	return ctxt->ops->cpl(ctxt) > iopl;
2918 }
2919 
2920 #define VMWARE_PORT_VMPORT	(0x5658)
2921 #define VMWARE_PORT_VMRPC	(0x5659)
2922 
2923 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2924 					    u16 port, u16 len)
2925 {
2926 	const struct x86_emulate_ops *ops = ctxt->ops;
2927 	struct desc_struct tr_seg;
2928 	u32 base3;
2929 	int r;
2930 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2931 	unsigned mask = (1 << len) - 1;
2932 	unsigned long base;
2933 
2934 	/*
2935 	 * VMware allows access to these ports even if denied
2936 	 * by TSS I/O permission bitmap. Mimic behavior.
2937 	 */
2938 	if (enable_vmware_backdoor &&
2939 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2940 		return true;
2941 
2942 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2943 	if (!tr_seg.p)
2944 		return false;
2945 	if (desc_limit_scaled(&tr_seg) < 103)
2946 		return false;
2947 	base = get_desc_base(&tr_seg);
2948 #ifdef CONFIG_X86_64
2949 	base |= ((u64)base3) << 32;
2950 #endif
2951 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2952 	if (r != X86EMUL_CONTINUE)
2953 		return false;
2954 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2955 		return false;
2956 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2957 	if (r != X86EMUL_CONTINUE)
2958 		return false;
2959 	if ((perm >> bit_idx) & mask)
2960 		return false;
2961 	return true;
2962 }
2963 
2964 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2965 				 u16 port, u16 len)
2966 {
2967 	if (ctxt->perm_ok)
2968 		return true;
2969 
2970 	if (emulator_bad_iopl(ctxt))
2971 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2972 			return false;
2973 
2974 	ctxt->perm_ok = true;
2975 
2976 	return true;
2977 }
2978 
2979 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2980 {
2981 	/*
2982 	 * Intel CPUs mask the counter and pointers in quite strange
2983 	 * manner when ECX is zero due to REP-string optimizations.
2984 	 */
2985 #ifdef CONFIG_X86_64
2986 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2987 		return;
2988 
2989 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2990 
2991 	switch (ctxt->b) {
2992 	case 0xa4:	/* movsb */
2993 	case 0xa5:	/* movsd/w */
2994 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2995 		/* fall through */
2996 	case 0xaa:	/* stosb */
2997 	case 0xab:	/* stosd/w */
2998 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2999 	}
3000 #endif
3001 }
3002 
3003 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
3004 				struct tss_segment_16 *tss)
3005 {
3006 	tss->ip = ctxt->_eip;
3007 	tss->flag = ctxt->eflags;
3008 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
3009 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
3010 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
3011 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
3012 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
3013 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
3014 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
3015 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
3016 
3017 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3018 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3019 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3020 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3021 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
3022 }
3023 
3024 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
3025 				 struct tss_segment_16 *tss)
3026 {
3027 	int ret;
3028 	u8 cpl;
3029 
3030 	ctxt->_eip = tss->ip;
3031 	ctxt->eflags = tss->flag | 2;
3032 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
3033 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
3034 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
3035 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
3036 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
3037 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
3038 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
3039 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
3040 
3041 	/*
3042 	 * SDM says that segment selectors are loaded before segment
3043 	 * descriptors
3044 	 */
3045 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
3046 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3047 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3048 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3049 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3050 
3051 	cpl = tss->cs & 3;
3052 
3053 	/*
3054 	 * Now load segment descriptors. If fault happens at this stage
3055 	 * it is handled in a context of new task
3056 	 */
3057 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
3058 					X86_TRANSFER_TASK_SWITCH, NULL);
3059 	if (ret != X86EMUL_CONTINUE)
3060 		return ret;
3061 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3062 					X86_TRANSFER_TASK_SWITCH, NULL);
3063 	if (ret != X86EMUL_CONTINUE)
3064 		return ret;
3065 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3066 					X86_TRANSFER_TASK_SWITCH, NULL);
3067 	if (ret != X86EMUL_CONTINUE)
3068 		return ret;
3069 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3070 					X86_TRANSFER_TASK_SWITCH, NULL);
3071 	if (ret != X86EMUL_CONTINUE)
3072 		return ret;
3073 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3074 					X86_TRANSFER_TASK_SWITCH, NULL);
3075 	if (ret != X86EMUL_CONTINUE)
3076 		return ret;
3077 
3078 	return X86EMUL_CONTINUE;
3079 }
3080 
3081 static int task_switch_16(struct x86_emulate_ctxt *ctxt,
3082 			  u16 tss_selector, u16 old_tss_sel,
3083 			  ulong old_tss_base, struct desc_struct *new_desc)
3084 {
3085 	struct tss_segment_16 tss_seg;
3086 	int ret;
3087 	u32 new_tss_base = get_desc_base(new_desc);
3088 
3089 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3090 	if (ret != X86EMUL_CONTINUE)
3091 		return ret;
3092 
3093 	save_state_to_tss16(ctxt, &tss_seg);
3094 
3095 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3096 	if (ret != X86EMUL_CONTINUE)
3097 		return ret;
3098 
3099 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3100 	if (ret != X86EMUL_CONTINUE)
3101 		return ret;
3102 
3103 	if (old_tss_sel != 0xffff) {
3104 		tss_seg.prev_task_link = old_tss_sel;
3105 
3106 		ret = linear_write_system(ctxt, new_tss_base,
3107 					  &tss_seg.prev_task_link,
3108 					  sizeof(tss_seg.prev_task_link));
3109 		if (ret != X86EMUL_CONTINUE)
3110 			return ret;
3111 	}
3112 
3113 	return load_state_from_tss16(ctxt, &tss_seg);
3114 }
3115 
3116 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3117 				struct tss_segment_32 *tss)
3118 {
3119 	/* CR3 and ldt selector are not saved intentionally */
3120 	tss->eip = ctxt->_eip;
3121 	tss->eflags = ctxt->eflags;
3122 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3123 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3124 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3125 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3126 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3127 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3128 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3129 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3130 
3131 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3132 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3133 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3134 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3135 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3136 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3137 }
3138 
3139 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3140 				 struct tss_segment_32 *tss)
3141 {
3142 	int ret;
3143 	u8 cpl;
3144 
3145 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3146 		return emulate_gp(ctxt, 0);
3147 	ctxt->_eip = tss->eip;
3148 	ctxt->eflags = tss->eflags | 2;
3149 
3150 	/* General purpose registers */
3151 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3152 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3153 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3154 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3155 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3156 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3157 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3158 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3159 
3160 	/*
3161 	 * SDM says that segment selectors are loaded before segment
3162 	 * descriptors.  This is important because CPL checks will
3163 	 * use CS.RPL.
3164 	 */
3165 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3166 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3167 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3168 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3169 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3170 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3171 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3172 
3173 	/*
3174 	 * If we're switching between Protected Mode and VM86, we need to make
3175 	 * sure to update the mode before loading the segment descriptors so
3176 	 * that the selectors are interpreted correctly.
3177 	 */
3178 	if (ctxt->eflags & X86_EFLAGS_VM) {
3179 		ctxt->mode = X86EMUL_MODE_VM86;
3180 		cpl = 3;
3181 	} else {
3182 		ctxt->mode = X86EMUL_MODE_PROT32;
3183 		cpl = tss->cs & 3;
3184 	}
3185 
3186 	/*
3187 	 * Now load segment descriptors. If fault happenes at this stage
3188 	 * it is handled in a context of new task
3189 	 */
3190 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3191 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3192 	if (ret != X86EMUL_CONTINUE)
3193 		return ret;
3194 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3195 					X86_TRANSFER_TASK_SWITCH, NULL);
3196 	if (ret != X86EMUL_CONTINUE)
3197 		return ret;
3198 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3199 					X86_TRANSFER_TASK_SWITCH, NULL);
3200 	if (ret != X86EMUL_CONTINUE)
3201 		return ret;
3202 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3203 					X86_TRANSFER_TASK_SWITCH, NULL);
3204 	if (ret != X86EMUL_CONTINUE)
3205 		return ret;
3206 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3207 					X86_TRANSFER_TASK_SWITCH, NULL);
3208 	if (ret != X86EMUL_CONTINUE)
3209 		return ret;
3210 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3211 					X86_TRANSFER_TASK_SWITCH, NULL);
3212 	if (ret != X86EMUL_CONTINUE)
3213 		return ret;
3214 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3215 					X86_TRANSFER_TASK_SWITCH, NULL);
3216 
3217 	return ret;
3218 }
3219 
3220 static int task_switch_32(struct x86_emulate_ctxt *ctxt,
3221 			  u16 tss_selector, u16 old_tss_sel,
3222 			  ulong old_tss_base, struct desc_struct *new_desc)
3223 {
3224 	struct tss_segment_32 tss_seg;
3225 	int ret;
3226 	u32 new_tss_base = get_desc_base(new_desc);
3227 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3228 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3229 
3230 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3231 	if (ret != X86EMUL_CONTINUE)
3232 		return ret;
3233 
3234 	save_state_to_tss32(ctxt, &tss_seg);
3235 
3236 	/* Only GP registers and segment selectors are saved */
3237 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3238 				  ldt_sel_offset - eip_offset);
3239 	if (ret != X86EMUL_CONTINUE)
3240 		return ret;
3241 
3242 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3243 	if (ret != X86EMUL_CONTINUE)
3244 		return ret;
3245 
3246 	if (old_tss_sel != 0xffff) {
3247 		tss_seg.prev_task_link = old_tss_sel;
3248 
3249 		ret = linear_write_system(ctxt, new_tss_base,
3250 					  &tss_seg.prev_task_link,
3251 					  sizeof(tss_seg.prev_task_link));
3252 		if (ret != X86EMUL_CONTINUE)
3253 			return ret;
3254 	}
3255 
3256 	return load_state_from_tss32(ctxt, &tss_seg);
3257 }
3258 
3259 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3260 				   u16 tss_selector, int idt_index, int reason,
3261 				   bool has_error_code, u32 error_code)
3262 {
3263 	const struct x86_emulate_ops *ops = ctxt->ops;
3264 	struct desc_struct curr_tss_desc, next_tss_desc;
3265 	int ret;
3266 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3267 	ulong old_tss_base =
3268 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3269 	u32 desc_limit;
3270 	ulong desc_addr, dr7;
3271 
3272 	/* FIXME: old_tss_base == ~0 ? */
3273 
3274 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3275 	if (ret != X86EMUL_CONTINUE)
3276 		return ret;
3277 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3278 	if (ret != X86EMUL_CONTINUE)
3279 		return ret;
3280 
3281 	/* FIXME: check that next_tss_desc is tss */
3282 
3283 	/*
3284 	 * Check privileges. The three cases are task switch caused by...
3285 	 *
3286 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3287 	 * 2. Exception/IRQ/iret: No check is performed
3288 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3289 	 *    hardware checks it before exiting.
3290 	 */
3291 	if (reason == TASK_SWITCH_GATE) {
3292 		if (idt_index != -1) {
3293 			/* Software interrupts */
3294 			struct desc_struct task_gate_desc;
3295 			int dpl;
3296 
3297 			ret = read_interrupt_descriptor(ctxt, idt_index,
3298 							&task_gate_desc);
3299 			if (ret != X86EMUL_CONTINUE)
3300 				return ret;
3301 
3302 			dpl = task_gate_desc.dpl;
3303 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3304 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3305 		}
3306 	}
3307 
3308 	desc_limit = desc_limit_scaled(&next_tss_desc);
3309 	if (!next_tss_desc.p ||
3310 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3311 	     desc_limit < 0x2b)) {
3312 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3313 	}
3314 
3315 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3316 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3317 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3318 	}
3319 
3320 	if (reason == TASK_SWITCH_IRET)
3321 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3322 
3323 	/* set back link to prev task only if NT bit is set in eflags
3324 	   note that old_tss_sel is not used after this point */
3325 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3326 		old_tss_sel = 0xffff;
3327 
3328 	if (next_tss_desc.type & 8)
3329 		ret = task_switch_32(ctxt, tss_selector, old_tss_sel,
3330 				     old_tss_base, &next_tss_desc);
3331 	else
3332 		ret = task_switch_16(ctxt, tss_selector, old_tss_sel,
3333 				     old_tss_base, &next_tss_desc);
3334 	if (ret != X86EMUL_CONTINUE)
3335 		return ret;
3336 
3337 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3338 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3339 
3340 	if (reason != TASK_SWITCH_IRET) {
3341 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3342 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3343 	}
3344 
3345 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3346 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3347 
3348 	if (has_error_code) {
3349 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3350 		ctxt->lock_prefix = 0;
3351 		ctxt->src.val = (unsigned long) error_code;
3352 		ret = em_push(ctxt);
3353 	}
3354 
3355 	ops->get_dr(ctxt, 7, &dr7);
3356 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3357 
3358 	return ret;
3359 }
3360 
3361 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3362 			 u16 tss_selector, int idt_index, int reason,
3363 			 bool has_error_code, u32 error_code)
3364 {
3365 	int rc;
3366 
3367 	invalidate_registers(ctxt);
3368 	ctxt->_eip = ctxt->eip;
3369 	ctxt->dst.type = OP_NONE;
3370 
3371 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3372 				     has_error_code, error_code);
3373 
3374 	if (rc == X86EMUL_CONTINUE) {
3375 		ctxt->eip = ctxt->_eip;
3376 		writeback_registers(ctxt);
3377 	}
3378 
3379 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3380 }
3381 
3382 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3383 		struct operand *op)
3384 {
3385 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3386 
3387 	register_address_increment(ctxt, reg, df * op->bytes);
3388 	op->addr.mem.ea = register_address(ctxt, reg);
3389 }
3390 
3391 static int em_das(struct x86_emulate_ctxt *ctxt)
3392 {
3393 	u8 al, old_al;
3394 	bool af, cf, old_cf;
3395 
3396 	cf = ctxt->eflags & X86_EFLAGS_CF;
3397 	al = ctxt->dst.val;
3398 
3399 	old_al = al;
3400 	old_cf = cf;
3401 	cf = false;
3402 	af = ctxt->eflags & X86_EFLAGS_AF;
3403 	if ((al & 0x0f) > 9 || af) {
3404 		al -= 6;
3405 		cf = old_cf | (al >= 250);
3406 		af = true;
3407 	} else {
3408 		af = false;
3409 	}
3410 	if (old_al > 0x99 || old_cf) {
3411 		al -= 0x60;
3412 		cf = true;
3413 	}
3414 
3415 	ctxt->dst.val = al;
3416 	/* Set PF, ZF, SF */
3417 	ctxt->src.type = OP_IMM;
3418 	ctxt->src.val = 0;
3419 	ctxt->src.bytes = 1;
3420 	fastop(ctxt, em_or);
3421 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3422 	if (cf)
3423 		ctxt->eflags |= X86_EFLAGS_CF;
3424 	if (af)
3425 		ctxt->eflags |= X86_EFLAGS_AF;
3426 	return X86EMUL_CONTINUE;
3427 }
3428 
3429 static int em_aam(struct x86_emulate_ctxt *ctxt)
3430 {
3431 	u8 al, ah;
3432 
3433 	if (ctxt->src.val == 0)
3434 		return emulate_de(ctxt);
3435 
3436 	al = ctxt->dst.val & 0xff;
3437 	ah = al / ctxt->src.val;
3438 	al %= ctxt->src.val;
3439 
3440 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3441 
3442 	/* Set PF, ZF, SF */
3443 	ctxt->src.type = OP_IMM;
3444 	ctxt->src.val = 0;
3445 	ctxt->src.bytes = 1;
3446 	fastop(ctxt, em_or);
3447 
3448 	return X86EMUL_CONTINUE;
3449 }
3450 
3451 static int em_aad(struct x86_emulate_ctxt *ctxt)
3452 {
3453 	u8 al = ctxt->dst.val & 0xff;
3454 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3455 
3456 	al = (al + (ah * ctxt->src.val)) & 0xff;
3457 
3458 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3459 
3460 	/* Set PF, ZF, SF */
3461 	ctxt->src.type = OP_IMM;
3462 	ctxt->src.val = 0;
3463 	ctxt->src.bytes = 1;
3464 	fastop(ctxt, em_or);
3465 
3466 	return X86EMUL_CONTINUE;
3467 }
3468 
3469 static int em_call(struct x86_emulate_ctxt *ctxt)
3470 {
3471 	int rc;
3472 	long rel = ctxt->src.val;
3473 
3474 	ctxt->src.val = (unsigned long)ctxt->_eip;
3475 	rc = jmp_rel(ctxt, rel);
3476 	if (rc != X86EMUL_CONTINUE)
3477 		return rc;
3478 	return em_push(ctxt);
3479 }
3480 
3481 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3482 {
3483 	u16 sel, old_cs;
3484 	ulong old_eip;
3485 	int rc;
3486 	struct desc_struct old_desc, new_desc;
3487 	const struct x86_emulate_ops *ops = ctxt->ops;
3488 	int cpl = ctxt->ops->cpl(ctxt);
3489 	enum x86emul_mode prev_mode = ctxt->mode;
3490 
3491 	old_eip = ctxt->_eip;
3492 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3493 
3494 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3495 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3496 				       X86_TRANSFER_CALL_JMP, &new_desc);
3497 	if (rc != X86EMUL_CONTINUE)
3498 		return rc;
3499 
3500 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3501 	if (rc != X86EMUL_CONTINUE)
3502 		goto fail;
3503 
3504 	ctxt->src.val = old_cs;
3505 	rc = em_push(ctxt);
3506 	if (rc != X86EMUL_CONTINUE)
3507 		goto fail;
3508 
3509 	ctxt->src.val = old_eip;
3510 	rc = em_push(ctxt);
3511 	/* If we failed, we tainted the memory, but the very least we should
3512 	   restore cs */
3513 	if (rc != X86EMUL_CONTINUE) {
3514 		pr_warn_once("faulting far call emulation tainted memory\n");
3515 		goto fail;
3516 	}
3517 	return rc;
3518 fail:
3519 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3520 	ctxt->mode = prev_mode;
3521 	return rc;
3522 
3523 }
3524 
3525 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3526 {
3527 	int rc;
3528 	unsigned long eip;
3529 
3530 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3531 	if (rc != X86EMUL_CONTINUE)
3532 		return rc;
3533 	rc = assign_eip_near(ctxt, eip);
3534 	if (rc != X86EMUL_CONTINUE)
3535 		return rc;
3536 	rsp_increment(ctxt, ctxt->src.val);
3537 	return X86EMUL_CONTINUE;
3538 }
3539 
3540 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3541 {
3542 	/* Write back the register source. */
3543 	ctxt->src.val = ctxt->dst.val;
3544 	write_register_operand(&ctxt->src);
3545 
3546 	/* Write back the memory destination with implicit LOCK prefix. */
3547 	ctxt->dst.val = ctxt->src.orig_val;
3548 	ctxt->lock_prefix = 1;
3549 	return X86EMUL_CONTINUE;
3550 }
3551 
3552 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3553 {
3554 	ctxt->dst.val = ctxt->src2.val;
3555 	return fastop(ctxt, em_imul);
3556 }
3557 
3558 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3559 {
3560 	ctxt->dst.type = OP_REG;
3561 	ctxt->dst.bytes = ctxt->src.bytes;
3562 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3563 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3564 
3565 	return X86EMUL_CONTINUE;
3566 }
3567 
3568 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3569 {
3570 	u64 tsc_aux = 0;
3571 
3572 	if (ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux))
3573 		return emulate_gp(ctxt, 0);
3574 	ctxt->dst.val = tsc_aux;
3575 	return X86EMUL_CONTINUE;
3576 }
3577 
3578 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3579 {
3580 	u64 tsc = 0;
3581 
3582 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3583 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3584 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3585 	return X86EMUL_CONTINUE;
3586 }
3587 
3588 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3589 {
3590 	u64 pmc;
3591 
3592 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3593 		return emulate_gp(ctxt, 0);
3594 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3595 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3596 	return X86EMUL_CONTINUE;
3597 }
3598 
3599 static int em_mov(struct x86_emulate_ctxt *ctxt)
3600 {
3601 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3602 	return X86EMUL_CONTINUE;
3603 }
3604 
3605 #define FFL(x) bit(X86_FEATURE_##x)
3606 
3607 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3608 {
3609 	u32 ebx, ecx, edx, eax = 1;
3610 	u16 tmp;
3611 
3612 	/*
3613 	 * Check MOVBE is set in the guest-visible CPUID leaf.
3614 	 */
3615 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3616 	if (!(ecx & FFL(MOVBE)))
3617 		return emulate_ud(ctxt);
3618 
3619 	switch (ctxt->op_bytes) {
3620 	case 2:
3621 		/*
3622 		 * From MOVBE definition: "...When the operand size is 16 bits,
3623 		 * the upper word of the destination register remains unchanged
3624 		 * ..."
3625 		 *
3626 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3627 		 * rules so we have to do the operation almost per hand.
3628 		 */
3629 		tmp = (u16)ctxt->src.val;
3630 		ctxt->dst.val &= ~0xffffUL;
3631 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3632 		break;
3633 	case 4:
3634 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3635 		break;
3636 	case 8:
3637 		ctxt->dst.val = swab64(ctxt->src.val);
3638 		break;
3639 	default:
3640 		BUG();
3641 	}
3642 	return X86EMUL_CONTINUE;
3643 }
3644 
3645 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3646 {
3647 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3648 		return emulate_gp(ctxt, 0);
3649 
3650 	/* Disable writeback. */
3651 	ctxt->dst.type = OP_NONE;
3652 	return X86EMUL_CONTINUE;
3653 }
3654 
3655 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3656 {
3657 	unsigned long val;
3658 
3659 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3660 		val = ctxt->src.val & ~0ULL;
3661 	else
3662 		val = ctxt->src.val & ~0U;
3663 
3664 	/* #UD condition is already handled. */
3665 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3666 		return emulate_gp(ctxt, 0);
3667 
3668 	/* Disable writeback. */
3669 	ctxt->dst.type = OP_NONE;
3670 	return X86EMUL_CONTINUE;
3671 }
3672 
3673 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3674 {
3675 	u64 msr_data;
3676 
3677 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3678 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3679 	if (ctxt->ops->set_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), msr_data))
3680 		return emulate_gp(ctxt, 0);
3681 
3682 	return X86EMUL_CONTINUE;
3683 }
3684 
3685 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3686 {
3687 	u64 msr_data;
3688 
3689 	if (ctxt->ops->get_msr(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &msr_data))
3690 		return emulate_gp(ctxt, 0);
3691 
3692 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3693 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3694 	return X86EMUL_CONTINUE;
3695 }
3696 
3697 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3698 {
3699 	if (segment > VCPU_SREG_GS &&
3700 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3701 	    ctxt->ops->cpl(ctxt) > 0)
3702 		return emulate_gp(ctxt, 0);
3703 
3704 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3705 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3706 		ctxt->dst.bytes = 2;
3707 	return X86EMUL_CONTINUE;
3708 }
3709 
3710 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3711 {
3712 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3713 		return emulate_ud(ctxt);
3714 
3715 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3716 }
3717 
3718 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3719 {
3720 	u16 sel = ctxt->src.val;
3721 
3722 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3723 		return emulate_ud(ctxt);
3724 
3725 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3726 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3727 
3728 	/* Disable writeback. */
3729 	ctxt->dst.type = OP_NONE;
3730 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3731 }
3732 
3733 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3734 {
3735 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3736 }
3737 
3738 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3739 {
3740 	u16 sel = ctxt->src.val;
3741 
3742 	/* Disable writeback. */
3743 	ctxt->dst.type = OP_NONE;
3744 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3745 }
3746 
3747 static int em_str(struct x86_emulate_ctxt *ctxt)
3748 {
3749 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3750 }
3751 
3752 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3753 {
3754 	u16 sel = ctxt->src.val;
3755 
3756 	/* Disable writeback. */
3757 	ctxt->dst.type = OP_NONE;
3758 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3759 }
3760 
3761 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3762 {
3763 	int rc;
3764 	ulong linear;
3765 
3766 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3767 	if (rc == X86EMUL_CONTINUE)
3768 		ctxt->ops->invlpg(ctxt, linear);
3769 	/* Disable writeback. */
3770 	ctxt->dst.type = OP_NONE;
3771 	return X86EMUL_CONTINUE;
3772 }
3773 
3774 static int em_clts(struct x86_emulate_ctxt *ctxt)
3775 {
3776 	ulong cr0;
3777 
3778 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3779 	cr0 &= ~X86_CR0_TS;
3780 	ctxt->ops->set_cr(ctxt, 0, cr0);
3781 	return X86EMUL_CONTINUE;
3782 }
3783 
3784 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3785 {
3786 	int rc = ctxt->ops->fix_hypercall(ctxt);
3787 
3788 	if (rc != X86EMUL_CONTINUE)
3789 		return rc;
3790 
3791 	/* Let the processor re-execute the fixed hypercall */
3792 	ctxt->_eip = ctxt->eip;
3793 	/* Disable writeback. */
3794 	ctxt->dst.type = OP_NONE;
3795 	return X86EMUL_CONTINUE;
3796 }
3797 
3798 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3799 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3800 					      struct desc_ptr *ptr))
3801 {
3802 	struct desc_ptr desc_ptr;
3803 
3804 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3805 	    ctxt->ops->cpl(ctxt) > 0)
3806 		return emulate_gp(ctxt, 0);
3807 
3808 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3809 		ctxt->op_bytes = 8;
3810 	get(ctxt, &desc_ptr);
3811 	if (ctxt->op_bytes == 2) {
3812 		ctxt->op_bytes = 4;
3813 		desc_ptr.address &= 0x00ffffff;
3814 	}
3815 	/* Disable writeback. */
3816 	ctxt->dst.type = OP_NONE;
3817 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3818 				   &desc_ptr, 2 + ctxt->op_bytes);
3819 }
3820 
3821 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3822 {
3823 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3824 }
3825 
3826 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3827 {
3828 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3829 }
3830 
3831 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3832 {
3833 	struct desc_ptr desc_ptr;
3834 	int rc;
3835 
3836 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3837 		ctxt->op_bytes = 8;
3838 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3839 			     &desc_ptr.size, &desc_ptr.address,
3840 			     ctxt->op_bytes);
3841 	if (rc != X86EMUL_CONTINUE)
3842 		return rc;
3843 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3844 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3845 		return emulate_gp(ctxt, 0);
3846 	if (lgdt)
3847 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3848 	else
3849 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3850 	/* Disable writeback. */
3851 	ctxt->dst.type = OP_NONE;
3852 	return X86EMUL_CONTINUE;
3853 }
3854 
3855 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3856 {
3857 	return em_lgdt_lidt(ctxt, true);
3858 }
3859 
3860 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3861 {
3862 	return em_lgdt_lidt(ctxt, false);
3863 }
3864 
3865 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3866 {
3867 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3868 	    ctxt->ops->cpl(ctxt) > 0)
3869 		return emulate_gp(ctxt, 0);
3870 
3871 	if (ctxt->dst.type == OP_MEM)
3872 		ctxt->dst.bytes = 2;
3873 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3874 	return X86EMUL_CONTINUE;
3875 }
3876 
3877 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3878 {
3879 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3880 			  | (ctxt->src.val & 0x0f));
3881 	ctxt->dst.type = OP_NONE;
3882 	return X86EMUL_CONTINUE;
3883 }
3884 
3885 static int em_loop(struct x86_emulate_ctxt *ctxt)
3886 {
3887 	int rc = X86EMUL_CONTINUE;
3888 
3889 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3890 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3891 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3892 		rc = jmp_rel(ctxt, ctxt->src.val);
3893 
3894 	return rc;
3895 }
3896 
3897 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3898 {
3899 	int rc = X86EMUL_CONTINUE;
3900 
3901 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3902 		rc = jmp_rel(ctxt, ctxt->src.val);
3903 
3904 	return rc;
3905 }
3906 
3907 static int em_in(struct x86_emulate_ctxt *ctxt)
3908 {
3909 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3910 			     &ctxt->dst.val))
3911 		return X86EMUL_IO_NEEDED;
3912 
3913 	return X86EMUL_CONTINUE;
3914 }
3915 
3916 static int em_out(struct x86_emulate_ctxt *ctxt)
3917 {
3918 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3919 				    &ctxt->src.val, 1);
3920 	/* Disable writeback. */
3921 	ctxt->dst.type = OP_NONE;
3922 	return X86EMUL_CONTINUE;
3923 }
3924 
3925 static int em_cli(struct x86_emulate_ctxt *ctxt)
3926 {
3927 	if (emulator_bad_iopl(ctxt))
3928 		return emulate_gp(ctxt, 0);
3929 
3930 	ctxt->eflags &= ~X86_EFLAGS_IF;
3931 	return X86EMUL_CONTINUE;
3932 }
3933 
3934 static int em_sti(struct x86_emulate_ctxt *ctxt)
3935 {
3936 	if (emulator_bad_iopl(ctxt))
3937 		return emulate_gp(ctxt, 0);
3938 
3939 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3940 	ctxt->eflags |= X86_EFLAGS_IF;
3941 	return X86EMUL_CONTINUE;
3942 }
3943 
3944 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3945 {
3946 	u32 eax, ebx, ecx, edx;
3947 	u64 msr = 0;
3948 
3949 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3950 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3951 	    ctxt->ops->cpl(ctxt)) {
3952 		return emulate_gp(ctxt, 0);
3953 	}
3954 
3955 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3956 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3957 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
3958 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3959 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3960 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3961 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3962 	return X86EMUL_CONTINUE;
3963 }
3964 
3965 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3966 {
3967 	u32 flags;
3968 
3969 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3970 		X86_EFLAGS_SF;
3971 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3972 
3973 	ctxt->eflags &= ~0xffUL;
3974 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3975 	return X86EMUL_CONTINUE;
3976 }
3977 
3978 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3979 {
3980 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3981 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3982 	return X86EMUL_CONTINUE;
3983 }
3984 
3985 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3986 {
3987 	switch (ctxt->op_bytes) {
3988 #ifdef CONFIG_X86_64
3989 	case 8:
3990 		asm("bswap %0" : "+r"(ctxt->dst.val));
3991 		break;
3992 #endif
3993 	default:
3994 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3995 		break;
3996 	}
3997 	return X86EMUL_CONTINUE;
3998 }
3999 
4000 static int em_clflush(struct x86_emulate_ctxt *ctxt)
4001 {
4002 	/* emulating clflush regardless of cpuid */
4003 	return X86EMUL_CONTINUE;
4004 }
4005 
4006 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
4007 {
4008 	ctxt->dst.val = (s32) ctxt->src.val;
4009 	return X86EMUL_CONTINUE;
4010 }
4011 
4012 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
4013 {
4014 	u32 eax = 1, ebx, ecx = 0, edx;
4015 
4016 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
4017 	if (!(edx & FFL(FXSR)))
4018 		return emulate_ud(ctxt);
4019 
4020 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
4021 		return emulate_nm(ctxt);
4022 
4023 	/*
4024 	 * Don't emulate a case that should never be hit, instead of working
4025 	 * around a lack of fxsave64/fxrstor64 on old compilers.
4026 	 */
4027 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
4028 		return X86EMUL_UNHANDLEABLE;
4029 
4030 	return X86EMUL_CONTINUE;
4031 }
4032 
4033 /*
4034  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
4035  * and restore MXCSR.
4036  */
4037 static size_t __fxstate_size(int nregs)
4038 {
4039 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
4040 }
4041 
4042 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
4043 {
4044 	bool cr4_osfxsr;
4045 	if (ctxt->mode == X86EMUL_MODE_PROT64)
4046 		return __fxstate_size(16);
4047 
4048 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
4049 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
4050 }
4051 
4052 /*
4053  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
4054  *  1) 16 bit mode
4055  *  2) 32 bit mode
4056  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
4057  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
4058  *       save and restore
4059  *  3) 64-bit mode with REX.W prefix
4060  *     - like (2), but XMM 8-15 are being saved and restored
4061  *  4) 64-bit mode without REX.W prefix
4062  *     - like (3), but FIP and FDP are 64 bit
4063  *
4064  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
4065  * desired result.  (4) is not emulated.
4066  *
4067  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
4068  * and FPU DS) should match.
4069  */
4070 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
4071 {
4072 	struct fxregs_state fx_state;
4073 	int rc;
4074 
4075 	rc = check_fxsr(ctxt);
4076 	if (rc != X86EMUL_CONTINUE)
4077 		return rc;
4078 
4079 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4080 
4081 	if (rc != X86EMUL_CONTINUE)
4082 		return rc;
4083 
4084 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4085 		                   fxstate_size(ctxt));
4086 }
4087 
4088 /*
4089  * FXRSTOR might restore XMM registers not provided by the guest. Fill
4090  * in the host registers (via FXSAVE) instead, so they won't be modified.
4091  * (preemption has to stay disabled until FXRSTOR).
4092  *
4093  * Use noinline to keep the stack for other functions called by callers small.
4094  */
4095 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
4096 				 const size_t used_size)
4097 {
4098 	struct fxregs_state fx_tmp;
4099 	int rc;
4100 
4101 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
4102 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
4103 	       __fxstate_size(16) - used_size);
4104 
4105 	return rc;
4106 }
4107 
4108 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4109 {
4110 	struct fxregs_state fx_state;
4111 	int rc;
4112 	size_t size;
4113 
4114 	rc = check_fxsr(ctxt);
4115 	if (rc != X86EMUL_CONTINUE)
4116 		return rc;
4117 
4118 	size = fxstate_size(ctxt);
4119 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4120 	if (rc != X86EMUL_CONTINUE)
4121 		return rc;
4122 
4123 	if (size < __fxstate_size(16)) {
4124 		rc = fxregs_fixup(&fx_state, size);
4125 		if (rc != X86EMUL_CONTINUE)
4126 			goto out;
4127 	}
4128 
4129 	if (fx_state.mxcsr >> 16) {
4130 		rc = emulate_gp(ctxt, 0);
4131 		goto out;
4132 	}
4133 
4134 	if (rc == X86EMUL_CONTINUE)
4135 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4136 
4137 out:
4138 	return rc;
4139 }
4140 
4141 static bool valid_cr(int nr)
4142 {
4143 	switch (nr) {
4144 	case 0:
4145 	case 2 ... 4:
4146 	case 8:
4147 		return true;
4148 	default:
4149 		return false;
4150 	}
4151 }
4152 
4153 static int check_cr_read(struct x86_emulate_ctxt *ctxt)
4154 {
4155 	if (!valid_cr(ctxt->modrm_reg))
4156 		return emulate_ud(ctxt);
4157 
4158 	return X86EMUL_CONTINUE;
4159 }
4160 
4161 static int check_cr_write(struct x86_emulate_ctxt *ctxt)
4162 {
4163 	u64 new_val = ctxt->src.val64;
4164 	int cr = ctxt->modrm_reg;
4165 	u64 efer = 0;
4166 
4167 	static u64 cr_reserved_bits[] = {
4168 		0xffffffff00000000ULL,
4169 		0, 0, 0, /* CR3 checked later */
4170 		CR4_RESERVED_BITS,
4171 		0, 0, 0,
4172 		CR8_RESERVED_BITS,
4173 	};
4174 
4175 	if (!valid_cr(cr))
4176 		return emulate_ud(ctxt);
4177 
4178 	if (new_val & cr_reserved_bits[cr])
4179 		return emulate_gp(ctxt, 0);
4180 
4181 	switch (cr) {
4182 	case 0: {
4183 		u64 cr4;
4184 		if (((new_val & X86_CR0_PG) && !(new_val & X86_CR0_PE)) ||
4185 		    ((new_val & X86_CR0_NW) && !(new_val & X86_CR0_CD)))
4186 			return emulate_gp(ctxt, 0);
4187 
4188 		cr4 = ctxt->ops->get_cr(ctxt, 4);
4189 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4190 
4191 		if ((new_val & X86_CR0_PG) && (efer & EFER_LME) &&
4192 		    !(cr4 & X86_CR4_PAE))
4193 			return emulate_gp(ctxt, 0);
4194 
4195 		break;
4196 		}
4197 	case 3: {
4198 		u64 rsvd = 0;
4199 
4200 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4201 		if (efer & EFER_LMA) {
4202 			u64 maxphyaddr;
4203 			u32 eax, ebx, ecx, edx;
4204 
4205 			eax = 0x80000008;
4206 			ecx = 0;
4207 			if (ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx,
4208 						 &edx, false))
4209 				maxphyaddr = eax & 0xff;
4210 			else
4211 				maxphyaddr = 36;
4212 			rsvd = rsvd_bits(maxphyaddr, 63);
4213 			if (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_PCIDE)
4214 				rsvd &= ~X86_CR3_PCID_NOFLUSH;
4215 		}
4216 
4217 		if (new_val & rsvd)
4218 			return emulate_gp(ctxt, 0);
4219 
4220 		break;
4221 		}
4222 	case 4: {
4223 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4224 
4225 		if ((efer & EFER_LMA) && !(new_val & X86_CR4_PAE))
4226 			return emulate_gp(ctxt, 0);
4227 
4228 		break;
4229 		}
4230 	}
4231 
4232 	return X86EMUL_CONTINUE;
4233 }
4234 
4235 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4236 {
4237 	unsigned long dr7;
4238 
4239 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4240 
4241 	/* Check if DR7.Global_Enable is set */
4242 	return dr7 & (1 << 13);
4243 }
4244 
4245 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4246 {
4247 	int dr = ctxt->modrm_reg;
4248 	u64 cr4;
4249 
4250 	if (dr > 7)
4251 		return emulate_ud(ctxt);
4252 
4253 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4254 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4255 		return emulate_ud(ctxt);
4256 
4257 	if (check_dr7_gd(ctxt)) {
4258 		ulong dr6;
4259 
4260 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4261 		dr6 &= ~15;
4262 		dr6 |= DR6_BD | DR6_RTM;
4263 		ctxt->ops->set_dr(ctxt, 6, dr6);
4264 		return emulate_db(ctxt);
4265 	}
4266 
4267 	return X86EMUL_CONTINUE;
4268 }
4269 
4270 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4271 {
4272 	u64 new_val = ctxt->src.val64;
4273 	int dr = ctxt->modrm_reg;
4274 
4275 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4276 		return emulate_gp(ctxt, 0);
4277 
4278 	return check_dr_read(ctxt);
4279 }
4280 
4281 static int check_svme(struct x86_emulate_ctxt *ctxt)
4282 {
4283 	u64 efer = 0;
4284 
4285 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4286 
4287 	if (!(efer & EFER_SVME))
4288 		return emulate_ud(ctxt);
4289 
4290 	return X86EMUL_CONTINUE;
4291 }
4292 
4293 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4294 {
4295 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4296 
4297 	/* Valid physical address? */
4298 	if (rax & 0xffff000000000000ULL)
4299 		return emulate_gp(ctxt, 0);
4300 
4301 	return check_svme(ctxt);
4302 }
4303 
4304 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4305 {
4306 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4307 
4308 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4309 		return emulate_ud(ctxt);
4310 
4311 	return X86EMUL_CONTINUE;
4312 }
4313 
4314 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4315 {
4316 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4317 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4318 
4319 	/*
4320 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
4321 	 * in Ring3 when CR4.PCE=0.
4322 	 */
4323 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
4324 		return X86EMUL_CONTINUE;
4325 
4326 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4327 	    ctxt->ops->check_pmc(ctxt, rcx))
4328 		return emulate_gp(ctxt, 0);
4329 
4330 	return X86EMUL_CONTINUE;
4331 }
4332 
4333 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4334 {
4335 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4336 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4337 		return emulate_gp(ctxt, 0);
4338 
4339 	return X86EMUL_CONTINUE;
4340 }
4341 
4342 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4343 {
4344 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4345 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4346 		return emulate_gp(ctxt, 0);
4347 
4348 	return X86EMUL_CONTINUE;
4349 }
4350 
4351 #define D(_y) { .flags = (_y) }
4352 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4353 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4354 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4355 #define N    D(NotImpl)
4356 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4357 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4358 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4359 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4360 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4361 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4362 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4363 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4364 #define II(_f, _e, _i) \
4365 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4366 #define IIP(_f, _e, _i, _p) \
4367 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4368 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4369 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4370 
4371 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4372 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4373 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4374 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4375 #define I2bvIP(_f, _e, _i, _p) \
4376 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4377 
4378 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4379 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4380 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4381 
4382 static const struct opcode group7_rm0[] = {
4383 	N,
4384 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4385 	N, N, N, N, N, N,
4386 };
4387 
4388 static const struct opcode group7_rm1[] = {
4389 	DI(SrcNone | Priv, monitor),
4390 	DI(SrcNone | Priv, mwait),
4391 	N, N, N, N, N, N,
4392 };
4393 
4394 static const struct opcode group7_rm3[] = {
4395 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4396 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4397 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4398 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4399 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4400 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4401 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4402 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4403 };
4404 
4405 static const struct opcode group7_rm7[] = {
4406 	N,
4407 	DIP(SrcNone, rdtscp, check_rdtsc),
4408 	N, N, N, N, N, N,
4409 };
4410 
4411 static const struct opcode group1[] = {
4412 	F(Lock, em_add),
4413 	F(Lock | PageTable, em_or),
4414 	F(Lock, em_adc),
4415 	F(Lock, em_sbb),
4416 	F(Lock | PageTable, em_and),
4417 	F(Lock, em_sub),
4418 	F(Lock, em_xor),
4419 	F(NoWrite, em_cmp),
4420 };
4421 
4422 static const struct opcode group1A[] = {
4423 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4424 };
4425 
4426 static const struct opcode group2[] = {
4427 	F(DstMem | ModRM, em_rol),
4428 	F(DstMem | ModRM, em_ror),
4429 	F(DstMem | ModRM, em_rcl),
4430 	F(DstMem | ModRM, em_rcr),
4431 	F(DstMem | ModRM, em_shl),
4432 	F(DstMem | ModRM, em_shr),
4433 	F(DstMem | ModRM, em_shl),
4434 	F(DstMem | ModRM, em_sar),
4435 };
4436 
4437 static const struct opcode group3[] = {
4438 	F(DstMem | SrcImm | NoWrite, em_test),
4439 	F(DstMem | SrcImm | NoWrite, em_test),
4440 	F(DstMem | SrcNone | Lock, em_not),
4441 	F(DstMem | SrcNone | Lock, em_neg),
4442 	F(DstXacc | Src2Mem, em_mul_ex),
4443 	F(DstXacc | Src2Mem, em_imul_ex),
4444 	F(DstXacc | Src2Mem, em_div_ex),
4445 	F(DstXacc | Src2Mem, em_idiv_ex),
4446 };
4447 
4448 static const struct opcode group4[] = {
4449 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4450 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4451 	N, N, N, N, N, N,
4452 };
4453 
4454 static const struct opcode group5[] = {
4455 	F(DstMem | SrcNone | Lock,		em_inc),
4456 	F(DstMem | SrcNone | Lock,		em_dec),
4457 	I(SrcMem | NearBranch,			em_call_near_abs),
4458 	I(SrcMemFAddr | ImplicitOps,		em_call_far),
4459 	I(SrcMem | NearBranch,			em_jmp_abs),
4460 	I(SrcMemFAddr | ImplicitOps,		em_jmp_far),
4461 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4462 };
4463 
4464 static const struct opcode group6[] = {
4465 	II(Prot | DstMem,	   em_sldt, sldt),
4466 	II(Prot | DstMem,	   em_str, str),
4467 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4468 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4469 	N, N, N, N,
4470 };
4471 
4472 static const struct group_dual group7 = { {
4473 	II(Mov | DstMem,			em_sgdt, sgdt),
4474 	II(Mov | DstMem,			em_sidt, sidt),
4475 	II(SrcMem | Priv,			em_lgdt, lgdt),
4476 	II(SrcMem | Priv,			em_lidt, lidt),
4477 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4478 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4479 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4480 }, {
4481 	EXT(0, group7_rm0),
4482 	EXT(0, group7_rm1),
4483 	N, EXT(0, group7_rm3),
4484 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4485 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4486 	EXT(0, group7_rm7),
4487 } };
4488 
4489 static const struct opcode group8[] = {
4490 	N, N, N, N,
4491 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4492 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4493 	F(DstMem | SrcImmByte | Lock,			em_btr),
4494 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4495 };
4496 
4497 /*
4498  * The "memory" destination is actually always a register, since we come
4499  * from the register case of group9.
4500  */
4501 static const struct gprefix pfx_0f_c7_7 = {
4502 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdtscp),
4503 };
4504 
4505 
4506 static const struct group_dual group9 = { {
4507 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4508 }, {
4509 	N, N, N, N, N, N, N,
4510 	GP(0, &pfx_0f_c7_7),
4511 } };
4512 
4513 static const struct opcode group11[] = {
4514 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4515 	X7(D(Undefined)),
4516 };
4517 
4518 static const struct gprefix pfx_0f_ae_7 = {
4519 	I(SrcMem | ByteOp, em_clflush), N, N, N,
4520 };
4521 
4522 static const struct group_dual group15 = { {
4523 	I(ModRM | Aligned16, em_fxsave),
4524 	I(ModRM | Aligned16, em_fxrstor),
4525 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4526 }, {
4527 	N, N, N, N, N, N, N, N,
4528 } };
4529 
4530 static const struct gprefix pfx_0f_6f_0f_7f = {
4531 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4532 };
4533 
4534 static const struct instr_dual instr_dual_0f_2b = {
4535 	I(0, em_mov), N
4536 };
4537 
4538 static const struct gprefix pfx_0f_2b = {
4539 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4540 };
4541 
4542 static const struct gprefix pfx_0f_10_0f_11 = {
4543 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4544 };
4545 
4546 static const struct gprefix pfx_0f_28_0f_29 = {
4547 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4548 };
4549 
4550 static const struct gprefix pfx_0f_e7 = {
4551 	N, I(Sse, em_mov), N, N,
4552 };
4553 
4554 static const struct escape escape_d9 = { {
4555 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4556 }, {
4557 	/* 0xC0 - 0xC7 */
4558 	N, N, N, N, N, N, N, N,
4559 	/* 0xC8 - 0xCF */
4560 	N, N, N, N, N, N, N, N,
4561 	/* 0xD0 - 0xC7 */
4562 	N, N, N, N, N, N, N, N,
4563 	/* 0xD8 - 0xDF */
4564 	N, N, N, N, N, N, N, N,
4565 	/* 0xE0 - 0xE7 */
4566 	N, N, N, N, N, N, N, N,
4567 	/* 0xE8 - 0xEF */
4568 	N, N, N, N, N, N, N, N,
4569 	/* 0xF0 - 0xF7 */
4570 	N, N, N, N, N, N, N, N,
4571 	/* 0xF8 - 0xFF */
4572 	N, N, N, N, N, N, N, N,
4573 } };
4574 
4575 static const struct escape escape_db = { {
4576 	N, N, N, N, N, N, N, N,
4577 }, {
4578 	/* 0xC0 - 0xC7 */
4579 	N, N, N, N, N, N, N, N,
4580 	/* 0xC8 - 0xCF */
4581 	N, N, N, N, N, N, N, N,
4582 	/* 0xD0 - 0xC7 */
4583 	N, N, N, N, N, N, N, N,
4584 	/* 0xD8 - 0xDF */
4585 	N, N, N, N, N, N, N, N,
4586 	/* 0xE0 - 0xE7 */
4587 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4588 	/* 0xE8 - 0xEF */
4589 	N, N, N, N, N, N, N, N,
4590 	/* 0xF0 - 0xF7 */
4591 	N, N, N, N, N, N, N, N,
4592 	/* 0xF8 - 0xFF */
4593 	N, N, N, N, N, N, N, N,
4594 } };
4595 
4596 static const struct escape escape_dd = { {
4597 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4598 }, {
4599 	/* 0xC0 - 0xC7 */
4600 	N, N, N, N, N, N, N, N,
4601 	/* 0xC8 - 0xCF */
4602 	N, N, N, N, N, N, N, N,
4603 	/* 0xD0 - 0xC7 */
4604 	N, N, N, N, N, N, N, N,
4605 	/* 0xD8 - 0xDF */
4606 	N, N, N, N, N, N, N, N,
4607 	/* 0xE0 - 0xE7 */
4608 	N, N, N, N, N, N, N, N,
4609 	/* 0xE8 - 0xEF */
4610 	N, N, N, N, N, N, N, N,
4611 	/* 0xF0 - 0xF7 */
4612 	N, N, N, N, N, N, N, N,
4613 	/* 0xF8 - 0xFF */
4614 	N, N, N, N, N, N, N, N,
4615 } };
4616 
4617 static const struct instr_dual instr_dual_0f_c3 = {
4618 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4619 };
4620 
4621 static const struct mode_dual mode_dual_63 = {
4622 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4623 };
4624 
4625 static const struct opcode opcode_table[256] = {
4626 	/* 0x00 - 0x07 */
4627 	F6ALU(Lock, em_add),
4628 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4629 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4630 	/* 0x08 - 0x0F */
4631 	F6ALU(Lock | PageTable, em_or),
4632 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4633 	N,
4634 	/* 0x10 - 0x17 */
4635 	F6ALU(Lock, em_adc),
4636 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4637 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4638 	/* 0x18 - 0x1F */
4639 	F6ALU(Lock, em_sbb),
4640 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4641 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4642 	/* 0x20 - 0x27 */
4643 	F6ALU(Lock | PageTable, em_and), N, N,
4644 	/* 0x28 - 0x2F */
4645 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4646 	/* 0x30 - 0x37 */
4647 	F6ALU(Lock, em_xor), N, N,
4648 	/* 0x38 - 0x3F */
4649 	F6ALU(NoWrite, em_cmp), N, N,
4650 	/* 0x40 - 0x4F */
4651 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4652 	/* 0x50 - 0x57 */
4653 	X8(I(SrcReg | Stack, em_push)),
4654 	/* 0x58 - 0x5F */
4655 	X8(I(DstReg | Stack, em_pop)),
4656 	/* 0x60 - 0x67 */
4657 	I(ImplicitOps | Stack | No64, em_pusha),
4658 	I(ImplicitOps | Stack | No64, em_popa),
4659 	N, MD(ModRM, &mode_dual_63),
4660 	N, N, N, N,
4661 	/* 0x68 - 0x6F */
4662 	I(SrcImm | Mov | Stack, em_push),
4663 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4664 	I(SrcImmByte | Mov | Stack, em_push),
4665 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4666 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4667 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4668 	/* 0x70 - 0x7F */
4669 	X16(D(SrcImmByte | NearBranch)),
4670 	/* 0x80 - 0x87 */
4671 	G(ByteOp | DstMem | SrcImm, group1),
4672 	G(DstMem | SrcImm, group1),
4673 	G(ByteOp | DstMem | SrcImm | No64, group1),
4674 	G(DstMem | SrcImmByte, group1),
4675 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4676 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4677 	/* 0x88 - 0x8F */
4678 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4679 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4680 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4681 	D(ModRM | SrcMem | NoAccess | DstReg),
4682 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4683 	G(0, group1A),
4684 	/* 0x90 - 0x97 */
4685 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4686 	/* 0x98 - 0x9F */
4687 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4688 	I(SrcImmFAddr | No64, em_call_far), N,
4689 	II(ImplicitOps | Stack, em_pushf, pushf),
4690 	II(ImplicitOps | Stack, em_popf, popf),
4691 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4692 	/* 0xA0 - 0xA7 */
4693 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4694 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4695 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4696 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4697 	/* 0xA8 - 0xAF */
4698 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4699 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4700 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4701 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4702 	/* 0xB0 - 0xB7 */
4703 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4704 	/* 0xB8 - 0xBF */
4705 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4706 	/* 0xC0 - 0xC7 */
4707 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4708 	I(ImplicitOps | NearBranch | SrcImmU16, em_ret_near_imm),
4709 	I(ImplicitOps | NearBranch, em_ret),
4710 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4711 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4712 	G(ByteOp, group11), G(0, group11),
4713 	/* 0xC8 - 0xCF */
4714 	I(Stack | SrcImmU16 | Src2ImmByte, em_enter), I(Stack, em_leave),
4715 	I(ImplicitOps | SrcImmU16, em_ret_far_imm),
4716 	I(ImplicitOps, em_ret_far),
4717 	D(ImplicitOps), DI(SrcImmByte, intn),
4718 	D(ImplicitOps | No64), II(ImplicitOps, em_iret, iret),
4719 	/* 0xD0 - 0xD7 */
4720 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4721 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4722 	I(DstAcc | SrcImmUByte | No64, em_aam),
4723 	I(DstAcc | SrcImmUByte | No64, em_aad),
4724 	F(DstAcc | ByteOp | No64, em_salc),
4725 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4726 	/* 0xD8 - 0xDF */
4727 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4728 	/* 0xE0 - 0xE7 */
4729 	X3(I(SrcImmByte | NearBranch, em_loop)),
4730 	I(SrcImmByte | NearBranch, em_jcxz),
4731 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4732 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4733 	/* 0xE8 - 0xEF */
4734 	I(SrcImm | NearBranch, em_call), D(SrcImm | ImplicitOps | NearBranch),
4735 	I(SrcImmFAddr | No64, em_jmp_far),
4736 	D(SrcImmByte | ImplicitOps | NearBranch),
4737 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4738 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4739 	/* 0xF0 - 0xF7 */
4740 	N, DI(ImplicitOps, icebp), N, N,
4741 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4742 	G(ByteOp, group3), G(0, group3),
4743 	/* 0xF8 - 0xFF */
4744 	D(ImplicitOps), D(ImplicitOps),
4745 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4746 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4747 };
4748 
4749 static const struct opcode twobyte_table[256] = {
4750 	/* 0x00 - 0x0F */
4751 	G(0, group6), GD(0, &group7), N, N,
4752 	N, I(ImplicitOps | EmulateOnUD, em_syscall),
4753 	II(ImplicitOps | Priv, em_clts, clts), N,
4754 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4755 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4756 	/* 0x10 - 0x1F */
4757 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4758 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4759 	N, N, N, N, N, N,
4760 	D(ImplicitOps | ModRM | SrcMem | NoAccess),
4761 	N, N, N, N, N, N, D(ImplicitOps | ModRM | SrcMem | NoAccess),
4762 	/* 0x20 - 0x2F */
4763 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_read),
4764 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4765 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4766 						check_cr_write),
4767 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4768 						check_dr_write),
4769 	N, N, N, N,
4770 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4771 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4772 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4773 	N, N, N, N,
4774 	/* 0x30 - 0x3F */
4775 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4776 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4777 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4778 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4779 	I(ImplicitOps | EmulateOnUD, em_sysenter),
4780 	I(ImplicitOps | Priv | EmulateOnUD, em_sysexit),
4781 	N, N,
4782 	N, N, N, N, N, N, N, N,
4783 	/* 0x40 - 0x4F */
4784 	X16(D(DstReg | SrcMem | ModRM)),
4785 	/* 0x50 - 0x5F */
4786 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4787 	/* 0x60 - 0x6F */
4788 	N, N, N, N,
4789 	N, N, N, N,
4790 	N, N, N, N,
4791 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4792 	/* 0x70 - 0x7F */
4793 	N, N, N, N,
4794 	N, N, N, N,
4795 	N, N, N, N,
4796 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4797 	/* 0x80 - 0x8F */
4798 	X16(D(SrcImm | NearBranch)),
4799 	/* 0x90 - 0x9F */
4800 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4801 	/* 0xA0 - 0xA7 */
4802 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4803 	II(ImplicitOps, em_cpuid, cpuid),
4804 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4805 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4806 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4807 	/* 0xA8 - 0xAF */
4808 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4809 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4810 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4811 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4812 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4813 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4814 	/* 0xB0 - 0xB7 */
4815 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4816 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4817 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4818 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4819 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4820 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4821 	/* 0xB8 - 0xBF */
4822 	N, N,
4823 	G(BitOp, group8),
4824 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4825 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4826 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4827 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4828 	/* 0xC0 - 0xC7 */
4829 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4830 	N, ID(0, &instr_dual_0f_c3),
4831 	N, N, N, GD(0, &group9),
4832 	/* 0xC8 - 0xCF */
4833 	X8(I(DstReg, em_bswap)),
4834 	/* 0xD0 - 0xDF */
4835 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4836 	/* 0xE0 - 0xEF */
4837 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4838 	N, N, N, N, N, N, N, N,
4839 	/* 0xF0 - 0xFF */
4840 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4841 };
4842 
4843 static const struct instr_dual instr_dual_0f_38_f0 = {
4844 	I(DstReg | SrcMem | Mov, em_movbe), N
4845 };
4846 
4847 static const struct instr_dual instr_dual_0f_38_f1 = {
4848 	I(DstMem | SrcReg | Mov, em_movbe), N
4849 };
4850 
4851 static const struct gprefix three_byte_0f_38_f0 = {
4852 	ID(0, &instr_dual_0f_38_f0), N, N, N
4853 };
4854 
4855 static const struct gprefix three_byte_0f_38_f1 = {
4856 	ID(0, &instr_dual_0f_38_f1), N, N, N
4857 };
4858 
4859 /*
4860  * Insns below are selected by the prefix which indexed by the third opcode
4861  * byte.
4862  */
4863 static const struct opcode opcode_map_0f_38[256] = {
4864 	/* 0x00 - 0x7f */
4865 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4866 	/* 0x80 - 0xef */
4867 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4868 	/* 0xf0 - 0xf1 */
4869 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4870 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4871 	/* 0xf2 - 0xff */
4872 	N, N, X4(N), X8(N)
4873 };
4874 
4875 #undef D
4876 #undef N
4877 #undef G
4878 #undef GD
4879 #undef I
4880 #undef GP
4881 #undef EXT
4882 #undef MD
4883 #undef ID
4884 
4885 #undef D2bv
4886 #undef D2bvIP
4887 #undef I2bv
4888 #undef I2bvIP
4889 #undef I6ALU
4890 
4891 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4892 {
4893 	unsigned size;
4894 
4895 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4896 	if (size == 8)
4897 		size = 4;
4898 	return size;
4899 }
4900 
4901 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4902 		      unsigned size, bool sign_extension)
4903 {
4904 	int rc = X86EMUL_CONTINUE;
4905 
4906 	op->type = OP_IMM;
4907 	op->bytes = size;
4908 	op->addr.mem.ea = ctxt->_eip;
4909 	/* NB. Immediates are sign-extended as necessary. */
4910 	switch (op->bytes) {
4911 	case 1:
4912 		op->val = insn_fetch(s8, ctxt);
4913 		break;
4914 	case 2:
4915 		op->val = insn_fetch(s16, ctxt);
4916 		break;
4917 	case 4:
4918 		op->val = insn_fetch(s32, ctxt);
4919 		break;
4920 	case 8:
4921 		op->val = insn_fetch(s64, ctxt);
4922 		break;
4923 	}
4924 	if (!sign_extension) {
4925 		switch (op->bytes) {
4926 		case 1:
4927 			op->val &= 0xff;
4928 			break;
4929 		case 2:
4930 			op->val &= 0xffff;
4931 			break;
4932 		case 4:
4933 			op->val &= 0xffffffff;
4934 			break;
4935 		}
4936 	}
4937 done:
4938 	return rc;
4939 }
4940 
4941 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4942 			  unsigned d)
4943 {
4944 	int rc = X86EMUL_CONTINUE;
4945 
4946 	switch (d) {
4947 	case OpReg:
4948 		decode_register_operand(ctxt, op);
4949 		break;
4950 	case OpImmUByte:
4951 		rc = decode_imm(ctxt, op, 1, false);
4952 		break;
4953 	case OpMem:
4954 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4955 	mem_common:
4956 		*op = ctxt->memop;
4957 		ctxt->memopp = op;
4958 		if (ctxt->d & BitOp)
4959 			fetch_bit_operand(ctxt);
4960 		op->orig_val = op->val;
4961 		break;
4962 	case OpMem64:
4963 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4964 		goto mem_common;
4965 	case OpAcc:
4966 		op->type = OP_REG;
4967 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4968 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4969 		fetch_register_operand(op);
4970 		op->orig_val = op->val;
4971 		break;
4972 	case OpAccLo:
4973 		op->type = OP_REG;
4974 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4975 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4976 		fetch_register_operand(op);
4977 		op->orig_val = op->val;
4978 		break;
4979 	case OpAccHi:
4980 		if (ctxt->d & ByteOp) {
4981 			op->type = OP_NONE;
4982 			break;
4983 		}
4984 		op->type = OP_REG;
4985 		op->bytes = ctxt->op_bytes;
4986 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4987 		fetch_register_operand(op);
4988 		op->orig_val = op->val;
4989 		break;
4990 	case OpDI:
4991 		op->type = OP_MEM;
4992 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4993 		op->addr.mem.ea =
4994 			register_address(ctxt, VCPU_REGS_RDI);
4995 		op->addr.mem.seg = VCPU_SREG_ES;
4996 		op->val = 0;
4997 		op->count = 1;
4998 		break;
4999 	case OpDX:
5000 		op->type = OP_REG;
5001 		op->bytes = 2;
5002 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
5003 		fetch_register_operand(op);
5004 		break;
5005 	case OpCL:
5006 		op->type = OP_IMM;
5007 		op->bytes = 1;
5008 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
5009 		break;
5010 	case OpImmByte:
5011 		rc = decode_imm(ctxt, op, 1, true);
5012 		break;
5013 	case OpOne:
5014 		op->type = OP_IMM;
5015 		op->bytes = 1;
5016 		op->val = 1;
5017 		break;
5018 	case OpImm:
5019 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
5020 		break;
5021 	case OpImm64:
5022 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
5023 		break;
5024 	case OpMem8:
5025 		ctxt->memop.bytes = 1;
5026 		if (ctxt->memop.type == OP_REG) {
5027 			ctxt->memop.addr.reg = decode_register(ctxt,
5028 					ctxt->modrm_rm, true);
5029 			fetch_register_operand(&ctxt->memop);
5030 		}
5031 		goto mem_common;
5032 	case OpMem16:
5033 		ctxt->memop.bytes = 2;
5034 		goto mem_common;
5035 	case OpMem32:
5036 		ctxt->memop.bytes = 4;
5037 		goto mem_common;
5038 	case OpImmU16:
5039 		rc = decode_imm(ctxt, op, 2, false);
5040 		break;
5041 	case OpImmU:
5042 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
5043 		break;
5044 	case OpSI:
5045 		op->type = OP_MEM;
5046 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5047 		op->addr.mem.ea =
5048 			register_address(ctxt, VCPU_REGS_RSI);
5049 		op->addr.mem.seg = ctxt->seg_override;
5050 		op->val = 0;
5051 		op->count = 1;
5052 		break;
5053 	case OpXLat:
5054 		op->type = OP_MEM;
5055 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5056 		op->addr.mem.ea =
5057 			address_mask(ctxt,
5058 				reg_read(ctxt, VCPU_REGS_RBX) +
5059 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
5060 		op->addr.mem.seg = ctxt->seg_override;
5061 		op->val = 0;
5062 		break;
5063 	case OpImmFAddr:
5064 		op->type = OP_IMM;
5065 		op->addr.mem.ea = ctxt->_eip;
5066 		op->bytes = ctxt->op_bytes + 2;
5067 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
5068 		break;
5069 	case OpMemFAddr:
5070 		ctxt->memop.bytes = ctxt->op_bytes + 2;
5071 		goto mem_common;
5072 	case OpES:
5073 		op->type = OP_IMM;
5074 		op->val = VCPU_SREG_ES;
5075 		break;
5076 	case OpCS:
5077 		op->type = OP_IMM;
5078 		op->val = VCPU_SREG_CS;
5079 		break;
5080 	case OpSS:
5081 		op->type = OP_IMM;
5082 		op->val = VCPU_SREG_SS;
5083 		break;
5084 	case OpDS:
5085 		op->type = OP_IMM;
5086 		op->val = VCPU_SREG_DS;
5087 		break;
5088 	case OpFS:
5089 		op->type = OP_IMM;
5090 		op->val = VCPU_SREG_FS;
5091 		break;
5092 	case OpGS:
5093 		op->type = OP_IMM;
5094 		op->val = VCPU_SREG_GS;
5095 		break;
5096 	case OpImplicit:
5097 		/* Special instructions do their own operand decoding. */
5098 	default:
5099 		op->type = OP_NONE; /* Disable writeback. */
5100 		break;
5101 	}
5102 
5103 done:
5104 	return rc;
5105 }
5106 
5107 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len)
5108 {
5109 	int rc = X86EMUL_CONTINUE;
5110 	int mode = ctxt->mode;
5111 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
5112 	bool op_prefix = false;
5113 	bool has_seg_override = false;
5114 	struct opcode opcode;
5115 	u16 dummy;
5116 	struct desc_struct desc;
5117 
5118 	ctxt->memop.type = OP_NONE;
5119 	ctxt->memopp = NULL;
5120 	ctxt->_eip = ctxt->eip;
5121 	ctxt->fetch.ptr = ctxt->fetch.data;
5122 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
5123 	ctxt->opcode_len = 1;
5124 	if (insn_len > 0)
5125 		memcpy(ctxt->fetch.data, insn, insn_len);
5126 	else {
5127 		rc = __do_insn_fetch_bytes(ctxt, 1);
5128 		if (rc != X86EMUL_CONTINUE)
5129 			return rc;
5130 	}
5131 
5132 	switch (mode) {
5133 	case X86EMUL_MODE_REAL:
5134 	case X86EMUL_MODE_VM86:
5135 		def_op_bytes = def_ad_bytes = 2;
5136 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
5137 		if (desc.d)
5138 			def_op_bytes = def_ad_bytes = 4;
5139 		break;
5140 	case X86EMUL_MODE_PROT16:
5141 		def_op_bytes = def_ad_bytes = 2;
5142 		break;
5143 	case X86EMUL_MODE_PROT32:
5144 		def_op_bytes = def_ad_bytes = 4;
5145 		break;
5146 #ifdef CONFIG_X86_64
5147 	case X86EMUL_MODE_PROT64:
5148 		def_op_bytes = 4;
5149 		def_ad_bytes = 8;
5150 		break;
5151 #endif
5152 	default:
5153 		return EMULATION_FAILED;
5154 	}
5155 
5156 	ctxt->op_bytes = def_op_bytes;
5157 	ctxt->ad_bytes = def_ad_bytes;
5158 
5159 	/* Legacy prefixes. */
5160 	for (;;) {
5161 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5162 		case 0x66:	/* operand-size override */
5163 			op_prefix = true;
5164 			/* switch between 2/4 bytes */
5165 			ctxt->op_bytes = def_op_bytes ^ 6;
5166 			break;
5167 		case 0x67:	/* address-size override */
5168 			if (mode == X86EMUL_MODE_PROT64)
5169 				/* switch between 4/8 bytes */
5170 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5171 			else
5172 				/* switch between 2/4 bytes */
5173 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5174 			break;
5175 		case 0x26:	/* ES override */
5176 		case 0x2e:	/* CS override */
5177 		case 0x36:	/* SS override */
5178 		case 0x3e:	/* DS override */
5179 			has_seg_override = true;
5180 			ctxt->seg_override = (ctxt->b >> 3) & 3;
5181 			break;
5182 		case 0x64:	/* FS override */
5183 		case 0x65:	/* GS override */
5184 			has_seg_override = true;
5185 			ctxt->seg_override = ctxt->b & 7;
5186 			break;
5187 		case 0x40 ... 0x4f: /* REX */
5188 			if (mode != X86EMUL_MODE_PROT64)
5189 				goto done_prefixes;
5190 			ctxt->rex_prefix = ctxt->b;
5191 			continue;
5192 		case 0xf0:	/* LOCK */
5193 			ctxt->lock_prefix = 1;
5194 			break;
5195 		case 0xf2:	/* REPNE/REPNZ */
5196 		case 0xf3:	/* REP/REPE/REPZ */
5197 			ctxt->rep_prefix = ctxt->b;
5198 			break;
5199 		default:
5200 			goto done_prefixes;
5201 		}
5202 
5203 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5204 
5205 		ctxt->rex_prefix = 0;
5206 	}
5207 
5208 done_prefixes:
5209 
5210 	/* REX prefix. */
5211 	if (ctxt->rex_prefix & 8)
5212 		ctxt->op_bytes = 8;	/* REX.W */
5213 
5214 	/* Opcode byte(s). */
5215 	opcode = opcode_table[ctxt->b];
5216 	/* Two-byte opcode? */
5217 	if (ctxt->b == 0x0f) {
5218 		ctxt->opcode_len = 2;
5219 		ctxt->b = insn_fetch(u8, ctxt);
5220 		opcode = twobyte_table[ctxt->b];
5221 
5222 		/* 0F_38 opcode map */
5223 		if (ctxt->b == 0x38) {
5224 			ctxt->opcode_len = 3;
5225 			ctxt->b = insn_fetch(u8, ctxt);
5226 			opcode = opcode_map_0f_38[ctxt->b];
5227 		}
5228 	}
5229 	ctxt->d = opcode.flags;
5230 
5231 	if (ctxt->d & ModRM)
5232 		ctxt->modrm = insn_fetch(u8, ctxt);
5233 
5234 	/* vex-prefix instructions are not implemented */
5235 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5236 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5237 		ctxt->d = NotImpl;
5238 	}
5239 
5240 	while (ctxt->d & GroupMask) {
5241 		switch (ctxt->d & GroupMask) {
5242 		case Group:
5243 			goffset = (ctxt->modrm >> 3) & 7;
5244 			opcode = opcode.u.group[goffset];
5245 			break;
5246 		case GroupDual:
5247 			goffset = (ctxt->modrm >> 3) & 7;
5248 			if ((ctxt->modrm >> 6) == 3)
5249 				opcode = opcode.u.gdual->mod3[goffset];
5250 			else
5251 				opcode = opcode.u.gdual->mod012[goffset];
5252 			break;
5253 		case RMExt:
5254 			goffset = ctxt->modrm & 7;
5255 			opcode = opcode.u.group[goffset];
5256 			break;
5257 		case Prefix:
5258 			if (ctxt->rep_prefix && op_prefix)
5259 				return EMULATION_FAILED;
5260 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5261 			switch (simd_prefix) {
5262 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5263 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5264 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5265 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5266 			}
5267 			break;
5268 		case Escape:
5269 			if (ctxt->modrm > 0xbf)
5270 				opcode = opcode.u.esc->high[ctxt->modrm - 0xc0];
5271 			else
5272 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5273 			break;
5274 		case InstrDual:
5275 			if ((ctxt->modrm >> 6) == 3)
5276 				opcode = opcode.u.idual->mod3;
5277 			else
5278 				opcode = opcode.u.idual->mod012;
5279 			break;
5280 		case ModeDual:
5281 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5282 				opcode = opcode.u.mdual->mode64;
5283 			else
5284 				opcode = opcode.u.mdual->mode32;
5285 			break;
5286 		default:
5287 			return EMULATION_FAILED;
5288 		}
5289 
5290 		ctxt->d &= ~(u64)GroupMask;
5291 		ctxt->d |= opcode.flags;
5292 	}
5293 
5294 	/* Unrecognised? */
5295 	if (ctxt->d == 0)
5296 		return EMULATION_FAILED;
5297 
5298 	ctxt->execute = opcode.u.execute;
5299 
5300 	if (unlikely(ctxt->ud) && likely(!(ctxt->d & EmulateOnUD)))
5301 		return EMULATION_FAILED;
5302 
5303 	if (unlikely(ctxt->d &
5304 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5305 	     No16))) {
5306 		/*
5307 		 * These are copied unconditionally here, and checked unconditionally
5308 		 * in x86_emulate_insn.
5309 		 */
5310 		ctxt->check_perm = opcode.check_perm;
5311 		ctxt->intercept = opcode.intercept;
5312 
5313 		if (ctxt->d & NotImpl)
5314 			return EMULATION_FAILED;
5315 
5316 		if (mode == X86EMUL_MODE_PROT64) {
5317 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5318 				ctxt->op_bytes = 8;
5319 			else if (ctxt->d & NearBranch)
5320 				ctxt->op_bytes = 8;
5321 		}
5322 
5323 		if (ctxt->d & Op3264) {
5324 			if (mode == X86EMUL_MODE_PROT64)
5325 				ctxt->op_bytes = 8;
5326 			else
5327 				ctxt->op_bytes = 4;
5328 		}
5329 
5330 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5331 			ctxt->op_bytes = 4;
5332 
5333 		if (ctxt->d & Sse)
5334 			ctxt->op_bytes = 16;
5335 		else if (ctxt->d & Mmx)
5336 			ctxt->op_bytes = 8;
5337 	}
5338 
5339 	/* ModRM and SIB bytes. */
5340 	if (ctxt->d & ModRM) {
5341 		rc = decode_modrm(ctxt, &ctxt->memop);
5342 		if (!has_seg_override) {
5343 			has_seg_override = true;
5344 			ctxt->seg_override = ctxt->modrm_seg;
5345 		}
5346 	} else if (ctxt->d & MemAbs)
5347 		rc = decode_abs(ctxt, &ctxt->memop);
5348 	if (rc != X86EMUL_CONTINUE)
5349 		goto done;
5350 
5351 	if (!has_seg_override)
5352 		ctxt->seg_override = VCPU_SREG_DS;
5353 
5354 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5355 
5356 	/*
5357 	 * Decode and fetch the source operand: register, memory
5358 	 * or immediate.
5359 	 */
5360 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5361 	if (rc != X86EMUL_CONTINUE)
5362 		goto done;
5363 
5364 	/*
5365 	 * Decode and fetch the second source operand: register, memory
5366 	 * or immediate.
5367 	 */
5368 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5369 	if (rc != X86EMUL_CONTINUE)
5370 		goto done;
5371 
5372 	/* Decode and fetch the destination operand: register or memory. */
5373 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5374 
5375 	if (ctxt->rip_relative && likely(ctxt->memopp))
5376 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5377 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5378 
5379 done:
5380 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5381 }
5382 
5383 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5384 {
5385 	return ctxt->d & PageTable;
5386 }
5387 
5388 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5389 {
5390 	/* The second termination condition only applies for REPE
5391 	 * and REPNE. Test if the repeat string operation prefix is
5392 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5393 	 * corresponding termination condition according to:
5394 	 * 	- if REPE/REPZ and ZF = 0 then done
5395 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5396 	 */
5397 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5398 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5399 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5400 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5401 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5402 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5403 		return true;
5404 
5405 	return false;
5406 }
5407 
5408 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5409 {
5410 	int rc;
5411 
5412 	rc = asm_safe("fwait");
5413 
5414 	if (unlikely(rc != X86EMUL_CONTINUE))
5415 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5416 
5417 	return X86EMUL_CONTINUE;
5418 }
5419 
5420 static void fetch_possible_mmx_operand(struct x86_emulate_ctxt *ctxt,
5421 				       struct operand *op)
5422 {
5423 	if (op->type == OP_MM)
5424 		read_mmx_reg(ctxt, &op->mm_val, op->addr.mm);
5425 }
5426 
5427 static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *))
5428 {
5429 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5430 
5431 	if (!(ctxt->d & ByteOp))
5432 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5433 
5434 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5435 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5436 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5437 	    : "c"(ctxt->src2.val));
5438 
5439 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5440 	if (!fop) /* exception is returned in fop variable */
5441 		return emulate_de(ctxt);
5442 	return X86EMUL_CONTINUE;
5443 }
5444 
5445 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5446 {
5447 	memset(&ctxt->rip_relative, 0,
5448 	       (void *)&ctxt->modrm - (void *)&ctxt->rip_relative);
5449 
5450 	ctxt->io_read.pos = 0;
5451 	ctxt->io_read.end = 0;
5452 	ctxt->mem_read.end = 0;
5453 }
5454 
5455 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5456 {
5457 	const struct x86_emulate_ops *ops = ctxt->ops;
5458 	int rc = X86EMUL_CONTINUE;
5459 	int saved_dst_type = ctxt->dst.type;
5460 	unsigned emul_flags;
5461 
5462 	ctxt->mem_read.pos = 0;
5463 
5464 	/* LOCK prefix is allowed only with some instructions */
5465 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5466 		rc = emulate_ud(ctxt);
5467 		goto done;
5468 	}
5469 
5470 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5471 		rc = emulate_ud(ctxt);
5472 		goto done;
5473 	}
5474 
5475 	emul_flags = ctxt->ops->get_hflags(ctxt);
5476 	if (unlikely(ctxt->d &
5477 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5478 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5479 				(ctxt->d & Undefined)) {
5480 			rc = emulate_ud(ctxt);
5481 			goto done;
5482 		}
5483 
5484 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5485 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5486 			rc = emulate_ud(ctxt);
5487 			goto done;
5488 		}
5489 
5490 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5491 			rc = emulate_nm(ctxt);
5492 			goto done;
5493 		}
5494 
5495 		if (ctxt->d & Mmx) {
5496 			rc = flush_pending_x87_faults(ctxt);
5497 			if (rc != X86EMUL_CONTINUE)
5498 				goto done;
5499 			/*
5500 			 * Now that we know the fpu is exception safe, we can fetch
5501 			 * operands from it.
5502 			 */
5503 			fetch_possible_mmx_operand(ctxt, &ctxt->src);
5504 			fetch_possible_mmx_operand(ctxt, &ctxt->src2);
5505 			if (!(ctxt->d & Mov))
5506 				fetch_possible_mmx_operand(ctxt, &ctxt->dst);
5507 		}
5508 
5509 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5510 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5511 						      X86_ICPT_PRE_EXCEPT);
5512 			if (rc != X86EMUL_CONTINUE)
5513 				goto done;
5514 		}
5515 
5516 		/* Instruction can only be executed in protected mode */
5517 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5518 			rc = emulate_ud(ctxt);
5519 			goto done;
5520 		}
5521 
5522 		/* Privileged instruction can be executed only in CPL=0 */
5523 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5524 			if (ctxt->d & PrivUD)
5525 				rc = emulate_ud(ctxt);
5526 			else
5527 				rc = emulate_gp(ctxt, 0);
5528 			goto done;
5529 		}
5530 
5531 		/* Do instruction specific permission checks */
5532 		if (ctxt->d & CheckPerm) {
5533 			rc = ctxt->check_perm(ctxt);
5534 			if (rc != X86EMUL_CONTINUE)
5535 				goto done;
5536 		}
5537 
5538 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5539 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5540 						      X86_ICPT_POST_EXCEPT);
5541 			if (rc != X86EMUL_CONTINUE)
5542 				goto done;
5543 		}
5544 
5545 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5546 			/* All REP prefixes have the same first termination condition */
5547 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5548 				string_registers_quirk(ctxt);
5549 				ctxt->eip = ctxt->_eip;
5550 				ctxt->eflags &= ~X86_EFLAGS_RF;
5551 				goto done;
5552 			}
5553 		}
5554 	}
5555 
5556 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5557 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5558 				    ctxt->src.valptr, ctxt->src.bytes);
5559 		if (rc != X86EMUL_CONTINUE)
5560 			goto done;
5561 		ctxt->src.orig_val64 = ctxt->src.val64;
5562 	}
5563 
5564 	if (ctxt->src2.type == OP_MEM) {
5565 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5566 				    &ctxt->src2.val, ctxt->src2.bytes);
5567 		if (rc != X86EMUL_CONTINUE)
5568 			goto done;
5569 	}
5570 
5571 	if ((ctxt->d & DstMask) == ImplicitOps)
5572 		goto special_insn;
5573 
5574 
5575 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5576 		/* optimisation - avoid slow emulated read if Mov */
5577 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5578 				   &ctxt->dst.val, ctxt->dst.bytes);
5579 		if (rc != X86EMUL_CONTINUE) {
5580 			if (!(ctxt->d & NoWrite) &&
5581 			    rc == X86EMUL_PROPAGATE_FAULT &&
5582 			    ctxt->exception.vector == PF_VECTOR)
5583 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5584 			goto done;
5585 		}
5586 	}
5587 	/* Copy full 64-bit value for CMPXCHG8B.  */
5588 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5589 
5590 special_insn:
5591 
5592 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5593 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5594 					      X86_ICPT_POST_MEMACCESS);
5595 		if (rc != X86EMUL_CONTINUE)
5596 			goto done;
5597 	}
5598 
5599 	if (ctxt->rep_prefix && (ctxt->d & String))
5600 		ctxt->eflags |= X86_EFLAGS_RF;
5601 	else
5602 		ctxt->eflags &= ~X86_EFLAGS_RF;
5603 
5604 	if (ctxt->execute) {
5605 		if (ctxt->d & Fastop) {
5606 			void (*fop)(struct fastop *) = (void *)ctxt->execute;
5607 			rc = fastop(ctxt, fop);
5608 			if (rc != X86EMUL_CONTINUE)
5609 				goto done;
5610 			goto writeback;
5611 		}
5612 		rc = ctxt->execute(ctxt);
5613 		if (rc != X86EMUL_CONTINUE)
5614 			goto done;
5615 		goto writeback;
5616 	}
5617 
5618 	if (ctxt->opcode_len == 2)
5619 		goto twobyte_insn;
5620 	else if (ctxt->opcode_len == 3)
5621 		goto threebyte_insn;
5622 
5623 	switch (ctxt->b) {
5624 	case 0x70 ... 0x7f: /* jcc (short) */
5625 		if (test_cc(ctxt->b, ctxt->eflags))
5626 			rc = jmp_rel(ctxt, ctxt->src.val);
5627 		break;
5628 	case 0x8d: /* lea r16/r32, m */
5629 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5630 		break;
5631 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5632 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5633 			ctxt->dst.type = OP_NONE;
5634 		else
5635 			rc = em_xchg(ctxt);
5636 		break;
5637 	case 0x98: /* cbw/cwde/cdqe */
5638 		switch (ctxt->op_bytes) {
5639 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5640 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5641 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5642 		}
5643 		break;
5644 	case 0xcc:		/* int3 */
5645 		rc = emulate_int(ctxt, 3);
5646 		break;
5647 	case 0xcd:		/* int n */
5648 		rc = emulate_int(ctxt, ctxt->src.val);
5649 		break;
5650 	case 0xce:		/* into */
5651 		if (ctxt->eflags & X86_EFLAGS_OF)
5652 			rc = emulate_int(ctxt, 4);
5653 		break;
5654 	case 0xe9: /* jmp rel */
5655 	case 0xeb: /* jmp rel short */
5656 		rc = jmp_rel(ctxt, ctxt->src.val);
5657 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5658 		break;
5659 	case 0xf4:              /* hlt */
5660 		ctxt->ops->halt(ctxt);
5661 		break;
5662 	case 0xf5:	/* cmc */
5663 		/* complement carry flag from eflags reg */
5664 		ctxt->eflags ^= X86_EFLAGS_CF;
5665 		break;
5666 	case 0xf8: /* clc */
5667 		ctxt->eflags &= ~X86_EFLAGS_CF;
5668 		break;
5669 	case 0xf9: /* stc */
5670 		ctxt->eflags |= X86_EFLAGS_CF;
5671 		break;
5672 	case 0xfc: /* cld */
5673 		ctxt->eflags &= ~X86_EFLAGS_DF;
5674 		break;
5675 	case 0xfd: /* std */
5676 		ctxt->eflags |= X86_EFLAGS_DF;
5677 		break;
5678 	default:
5679 		goto cannot_emulate;
5680 	}
5681 
5682 	if (rc != X86EMUL_CONTINUE)
5683 		goto done;
5684 
5685 writeback:
5686 	if (ctxt->d & SrcWrite) {
5687 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5688 		rc = writeback(ctxt, &ctxt->src);
5689 		if (rc != X86EMUL_CONTINUE)
5690 			goto done;
5691 	}
5692 	if (!(ctxt->d & NoWrite)) {
5693 		rc = writeback(ctxt, &ctxt->dst);
5694 		if (rc != X86EMUL_CONTINUE)
5695 			goto done;
5696 	}
5697 
5698 	/*
5699 	 * restore dst type in case the decoding will be reused
5700 	 * (happens for string instruction )
5701 	 */
5702 	ctxt->dst.type = saved_dst_type;
5703 
5704 	if ((ctxt->d & SrcMask) == SrcSI)
5705 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5706 
5707 	if ((ctxt->d & DstMask) == DstDI)
5708 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5709 
5710 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5711 		unsigned int count;
5712 		struct read_cache *r = &ctxt->io_read;
5713 		if ((ctxt->d & SrcMask) == SrcSI)
5714 			count = ctxt->src.count;
5715 		else
5716 			count = ctxt->dst.count;
5717 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5718 
5719 		if (!string_insn_completed(ctxt)) {
5720 			/*
5721 			 * Re-enter guest when pio read ahead buffer is empty
5722 			 * or, if it is not used, after each 1024 iteration.
5723 			 */
5724 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5725 			    (r->end == 0 || r->end != r->pos)) {
5726 				/*
5727 				 * Reset read cache. Usually happens before
5728 				 * decode, but since instruction is restarted
5729 				 * we have to do it here.
5730 				 */
5731 				ctxt->mem_read.end = 0;
5732 				writeback_registers(ctxt);
5733 				return EMULATION_RESTART;
5734 			}
5735 			goto done; /* skip rip writeback */
5736 		}
5737 		ctxt->eflags &= ~X86_EFLAGS_RF;
5738 	}
5739 
5740 	ctxt->eip = ctxt->_eip;
5741 
5742 done:
5743 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5744 		WARN_ON(ctxt->exception.vector > 0x1f);
5745 		ctxt->have_exception = true;
5746 	}
5747 	if (rc == X86EMUL_INTERCEPTED)
5748 		return EMULATION_INTERCEPTED;
5749 
5750 	if (rc == X86EMUL_CONTINUE)
5751 		writeback_registers(ctxt);
5752 
5753 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5754 
5755 twobyte_insn:
5756 	switch (ctxt->b) {
5757 	case 0x09:		/* wbinvd */
5758 		(ctxt->ops->wbinvd)(ctxt);
5759 		break;
5760 	case 0x08:		/* invd */
5761 	case 0x0d:		/* GrpP (prefetch) */
5762 	case 0x18:		/* Grp16 (prefetch/nop) */
5763 	case 0x1f:		/* nop */
5764 		break;
5765 	case 0x20: /* mov cr, reg */
5766 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5767 		break;
5768 	case 0x21: /* mov from dr to reg */
5769 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5770 		break;
5771 	case 0x40 ... 0x4f:	/* cmov */
5772 		if (test_cc(ctxt->b, ctxt->eflags))
5773 			ctxt->dst.val = ctxt->src.val;
5774 		else if (ctxt->op_bytes != 4)
5775 			ctxt->dst.type = OP_NONE; /* no writeback */
5776 		break;
5777 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5778 		if (test_cc(ctxt->b, ctxt->eflags))
5779 			rc = jmp_rel(ctxt, ctxt->src.val);
5780 		break;
5781 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5782 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5783 		break;
5784 	case 0xb6 ... 0xb7:	/* movzx */
5785 		ctxt->dst.bytes = ctxt->op_bytes;
5786 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5787 						       : (u16) ctxt->src.val;
5788 		break;
5789 	case 0xbe ... 0xbf:	/* movsx */
5790 		ctxt->dst.bytes = ctxt->op_bytes;
5791 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5792 							(s16) ctxt->src.val;
5793 		break;
5794 	default:
5795 		goto cannot_emulate;
5796 	}
5797 
5798 threebyte_insn:
5799 
5800 	if (rc != X86EMUL_CONTINUE)
5801 		goto done;
5802 
5803 	goto writeback;
5804 
5805 cannot_emulate:
5806 	return EMULATION_FAILED;
5807 }
5808 
5809 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5810 {
5811 	invalidate_registers(ctxt);
5812 }
5813 
5814 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5815 {
5816 	writeback_registers(ctxt);
5817 }
5818 
5819 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5820 {
5821 	if (ctxt->rep_prefix && (ctxt->d & String))
5822 		return false;
5823 
5824 	if (ctxt->d & TwoMemOp)
5825 		return false;
5826 
5827 	return true;
5828 }
5829