xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision d87c25e8)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3  * emulate.c
4  *
5  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
6  *
7  * Copyright (c) 2005 Keir Fraser
8  *
9  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
10  * privileged instructions:
11  *
12  * Copyright (C) 2006 Qumranet
13  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
14  *
15  *   Avi Kivity <avi@qumranet.com>
16  *   Yaniv Kamay <yaniv@qumranet.com>
17  *
18  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
19  */
20 
21 #include <linux/kvm_host.h>
22 #include "kvm_cache_regs.h"
23 #include "kvm_emulate.h"
24 #include <linux/stringify.h>
25 #include <asm/debugreg.h>
26 #include <asm/nospec-branch.h>
27 
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 #define IsBranch    ((u64)1 << 56)  /* Instruction is considered a branch. */
179 
180 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
181 
182 #define X2(x...) x, x
183 #define X3(x...) X2(x), x
184 #define X4(x...) X2(x), X2(x)
185 #define X5(x...) X4(x), x
186 #define X6(x...) X4(x), X2(x)
187 #define X7(x...) X4(x), X3(x)
188 #define X8(x...) X4(x), X4(x)
189 #define X16(x...) X8(x), X8(x)
190 
191 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1)
192 #define FASTOP_SIZE 8
193 
194 struct opcode {
195 	u64 flags;
196 	u8 intercept;
197 	u8 pad[7];
198 	union {
199 		int (*execute)(struct x86_emulate_ctxt *ctxt);
200 		const struct opcode *group;
201 		const struct group_dual *gdual;
202 		const struct gprefix *gprefix;
203 		const struct escape *esc;
204 		const struct instr_dual *idual;
205 		const struct mode_dual *mdual;
206 		void (*fastop)(struct fastop *fake);
207 	} u;
208 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
209 };
210 
211 struct group_dual {
212 	struct opcode mod012[8];
213 	struct opcode mod3[8];
214 };
215 
216 struct gprefix {
217 	struct opcode pfx_no;
218 	struct opcode pfx_66;
219 	struct opcode pfx_f2;
220 	struct opcode pfx_f3;
221 };
222 
223 struct escape {
224 	struct opcode op[8];
225 	struct opcode high[64];
226 };
227 
228 struct instr_dual {
229 	struct opcode mod012;
230 	struct opcode mod3;
231 };
232 
233 struct mode_dual {
234 	struct opcode mode32;
235 	struct opcode mode64;
236 };
237 
238 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
239 
240 enum x86_transfer_type {
241 	X86_TRANSFER_NONE,
242 	X86_TRANSFER_CALL_JMP,
243 	X86_TRANSFER_RET,
244 	X86_TRANSFER_TASK_SWITCH,
245 };
246 
247 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
248 {
249 	if (!(ctxt->regs_valid & (1 << nr))) {
250 		ctxt->regs_valid |= 1 << nr;
251 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
252 	}
253 	return ctxt->_regs[nr];
254 }
255 
256 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
257 {
258 	ctxt->regs_valid |= 1 << nr;
259 	ctxt->regs_dirty |= 1 << nr;
260 	return &ctxt->_regs[nr];
261 }
262 
263 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
264 {
265 	reg_read(ctxt, nr);
266 	return reg_write(ctxt, nr);
267 }
268 
269 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
270 {
271 	unsigned reg;
272 
273 	for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16)
274 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
275 }
276 
277 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
278 {
279 	ctxt->regs_dirty = 0;
280 	ctxt->regs_valid = 0;
281 }
282 
283 /*
284  * These EFLAGS bits are restored from saved value during emulation, and
285  * any changes are written back to the saved value after emulation.
286  */
287 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
288 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
289 
290 #ifdef CONFIG_X86_64
291 #define ON64(x) x
292 #else
293 #define ON64(x)
294 #endif
295 
296 /*
297  * fastop functions have a special calling convention:
298  *
299  * dst:    rax        (in/out)
300  * src:    rdx        (in/out)
301  * src2:   rcx        (in)
302  * flags:  rflags     (in/out)
303  * ex:     rsi        (in:fastop pointer, out:zero if exception)
304  *
305  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
306  * different operand sizes can be reached by calculation, rather than a jump
307  * table (which would be bigger than the code).
308  */
309 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
310 
311 #define __FOP_FUNC(name) \
312 	".align " __stringify(FASTOP_SIZE) " \n\t" \
313 	".type " name ", @function \n\t" \
314 	name ":\n\t"
315 
316 #define FOP_FUNC(name) \
317 	__FOP_FUNC(#name)
318 
319 #define __FOP_RET(name) \
320 	"11: " ASM_RET \
321 	".size " name ", .-" name "\n\t"
322 
323 #define FOP_RET(name) \
324 	__FOP_RET(#name)
325 
326 #define FOP_START(op) \
327 	extern void em_##op(struct fastop *fake); \
328 	asm(".pushsection .text, \"ax\" \n\t" \
329 	    ".global em_" #op " \n\t" \
330 	    ".align " __stringify(FASTOP_SIZE) " \n\t" \
331 	    "em_" #op ":\n\t"
332 
333 #define FOP_END \
334 	    ".popsection")
335 
336 #define __FOPNOP(name) \
337 	__FOP_FUNC(name) \
338 	__FOP_RET(name)
339 
340 #define FOPNOP() \
341 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
342 
343 #define FOP1E(op,  dst) \
344 	__FOP_FUNC(#op "_" #dst) \
345 	"10: " #op " %" #dst " \n\t" \
346 	__FOP_RET(#op "_" #dst)
347 
348 #define FOP1EEX(op,  dst) \
349 	FOP1E(op, dst) _ASM_EXTABLE_TYPE_REG(10b, 11b, EX_TYPE_ZERO_REG, %%esi)
350 
351 #define FASTOP1(op) \
352 	FOP_START(op) \
353 	FOP1E(op##b, al) \
354 	FOP1E(op##w, ax) \
355 	FOP1E(op##l, eax) \
356 	ON64(FOP1E(op##q, rax))	\
357 	FOP_END
358 
359 /* 1-operand, using src2 (for MUL/DIV r/m) */
360 #define FASTOP1SRC2(op, name) \
361 	FOP_START(name) \
362 	FOP1E(op, cl) \
363 	FOP1E(op, cx) \
364 	FOP1E(op, ecx) \
365 	ON64(FOP1E(op, rcx)) \
366 	FOP_END
367 
368 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
369 #define FASTOP1SRC2EX(op, name) \
370 	FOP_START(name) \
371 	FOP1EEX(op, cl) \
372 	FOP1EEX(op, cx) \
373 	FOP1EEX(op, ecx) \
374 	ON64(FOP1EEX(op, rcx)) \
375 	FOP_END
376 
377 #define FOP2E(op,  dst, src)	   \
378 	__FOP_FUNC(#op "_" #dst "_" #src) \
379 	#op " %" #src ", %" #dst " \n\t" \
380 	__FOP_RET(#op "_" #dst "_" #src)
381 
382 #define FASTOP2(op) \
383 	FOP_START(op) \
384 	FOP2E(op##b, al, dl) \
385 	FOP2E(op##w, ax, dx) \
386 	FOP2E(op##l, eax, edx) \
387 	ON64(FOP2E(op##q, rax, rdx)) \
388 	FOP_END
389 
390 /* 2 operand, word only */
391 #define FASTOP2W(op) \
392 	FOP_START(op) \
393 	FOPNOP() \
394 	FOP2E(op##w, ax, dx) \
395 	FOP2E(op##l, eax, edx) \
396 	ON64(FOP2E(op##q, rax, rdx)) \
397 	FOP_END
398 
399 /* 2 operand, src is CL */
400 #define FASTOP2CL(op) \
401 	FOP_START(op) \
402 	FOP2E(op##b, al, cl) \
403 	FOP2E(op##w, ax, cl) \
404 	FOP2E(op##l, eax, cl) \
405 	ON64(FOP2E(op##q, rax, cl)) \
406 	FOP_END
407 
408 /* 2 operand, src and dest are reversed */
409 #define FASTOP2R(op, name) \
410 	FOP_START(name) \
411 	FOP2E(op##b, dl, al) \
412 	FOP2E(op##w, dx, ax) \
413 	FOP2E(op##l, edx, eax) \
414 	ON64(FOP2E(op##q, rdx, rax)) \
415 	FOP_END
416 
417 #define FOP3E(op,  dst, src, src2) \
418 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
419 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
420 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
421 
422 /* 3-operand, word-only, src2=cl */
423 #define FASTOP3WCL(op) \
424 	FOP_START(op) \
425 	FOPNOP() \
426 	FOP3E(op##w, ax, dx, cl) \
427 	FOP3E(op##l, eax, edx, cl) \
428 	ON64(FOP3E(op##q, rax, rdx, cl)) \
429 	FOP_END
430 
431 /* Special case for SETcc - 1 instruction per cc */
432 #define FOP_SETCC(op) \
433 	".align 4 \n\t" \
434 	".type " #op ", @function \n\t" \
435 	#op ": \n\t" \
436 	#op " %al \n\t" \
437 	__FOP_RET(#op)
438 
439 FOP_START(setcc)
440 FOP_SETCC(seto)
441 FOP_SETCC(setno)
442 FOP_SETCC(setc)
443 FOP_SETCC(setnc)
444 FOP_SETCC(setz)
445 FOP_SETCC(setnz)
446 FOP_SETCC(setbe)
447 FOP_SETCC(setnbe)
448 FOP_SETCC(sets)
449 FOP_SETCC(setns)
450 FOP_SETCC(setp)
451 FOP_SETCC(setnp)
452 FOP_SETCC(setl)
453 FOP_SETCC(setnl)
454 FOP_SETCC(setle)
455 FOP_SETCC(setnle)
456 FOP_END;
457 
458 FOP_START(salc)
459 FOP_FUNC(salc)
460 "pushf; sbb %al, %al; popf \n\t"
461 FOP_RET(salc)
462 FOP_END;
463 
464 /*
465  * XXX: inoutclob user must know where the argument is being expanded.
466  *      Relying on CONFIG_CC_HAS_ASM_GOTO would allow us to remove _fault.
467  */
468 #define asm_safe(insn, inoutclob...) \
469 ({ \
470 	int _fault = 0; \
471  \
472 	asm volatile("1:" insn "\n" \
473 	             "2:\n" \
474 		     _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_ONE_REG, %[_fault]) \
475 	             : [_fault] "+r"(_fault) inoutclob ); \
476  \
477 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
478 })
479 
480 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
481 				    enum x86_intercept intercept,
482 				    enum x86_intercept_stage stage)
483 {
484 	struct x86_instruction_info info = {
485 		.intercept  = intercept,
486 		.rep_prefix = ctxt->rep_prefix,
487 		.modrm_mod  = ctxt->modrm_mod,
488 		.modrm_reg  = ctxt->modrm_reg,
489 		.modrm_rm   = ctxt->modrm_rm,
490 		.src_val    = ctxt->src.val64,
491 		.dst_val    = ctxt->dst.val64,
492 		.src_bytes  = ctxt->src.bytes,
493 		.dst_bytes  = ctxt->dst.bytes,
494 		.ad_bytes   = ctxt->ad_bytes,
495 		.next_rip   = ctxt->eip,
496 	};
497 
498 	return ctxt->ops->intercept(ctxt, &info, stage);
499 }
500 
501 static void assign_masked(ulong *dest, ulong src, ulong mask)
502 {
503 	*dest = (*dest & ~mask) | (src & mask);
504 }
505 
506 static void assign_register(unsigned long *reg, u64 val, int bytes)
507 {
508 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
509 	switch (bytes) {
510 	case 1:
511 		*(u8 *)reg = (u8)val;
512 		break;
513 	case 2:
514 		*(u16 *)reg = (u16)val;
515 		break;
516 	case 4:
517 		*reg = (u32)val;
518 		break;	/* 64b: zero-extend */
519 	case 8:
520 		*reg = val;
521 		break;
522 	}
523 }
524 
525 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
526 {
527 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
528 }
529 
530 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
531 {
532 	u16 sel;
533 	struct desc_struct ss;
534 
535 	if (ctxt->mode == X86EMUL_MODE_PROT64)
536 		return ~0UL;
537 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
538 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
539 }
540 
541 static int stack_size(struct x86_emulate_ctxt *ctxt)
542 {
543 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
544 }
545 
546 /* Access/update address held in a register, based on addressing mode. */
547 static inline unsigned long
548 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
549 {
550 	if (ctxt->ad_bytes == sizeof(unsigned long))
551 		return reg;
552 	else
553 		return reg & ad_mask(ctxt);
554 }
555 
556 static inline unsigned long
557 register_address(struct x86_emulate_ctxt *ctxt, int reg)
558 {
559 	return address_mask(ctxt, reg_read(ctxt, reg));
560 }
561 
562 static void masked_increment(ulong *reg, ulong mask, int inc)
563 {
564 	assign_masked(reg, *reg + inc, mask);
565 }
566 
567 static inline void
568 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
569 {
570 	ulong *preg = reg_rmw(ctxt, reg);
571 
572 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
573 }
574 
575 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
576 {
577 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
578 }
579 
580 static u32 desc_limit_scaled(struct desc_struct *desc)
581 {
582 	u32 limit = get_desc_limit(desc);
583 
584 	return desc->g ? (limit << 12) | 0xfff : limit;
585 }
586 
587 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
588 {
589 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
590 		return 0;
591 
592 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
593 }
594 
595 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
596 			     u32 error, bool valid)
597 {
598 	WARN_ON(vec > 0x1f);
599 	ctxt->exception.vector = vec;
600 	ctxt->exception.error_code = error;
601 	ctxt->exception.error_code_valid = valid;
602 	return X86EMUL_PROPAGATE_FAULT;
603 }
604 
605 static int emulate_db(struct x86_emulate_ctxt *ctxt)
606 {
607 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
608 }
609 
610 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
611 {
612 	return emulate_exception(ctxt, GP_VECTOR, err, true);
613 }
614 
615 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
616 {
617 	return emulate_exception(ctxt, SS_VECTOR, err, true);
618 }
619 
620 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
621 {
622 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
623 }
624 
625 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
626 {
627 	return emulate_exception(ctxt, TS_VECTOR, err, true);
628 }
629 
630 static int emulate_de(struct x86_emulate_ctxt *ctxt)
631 {
632 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
633 }
634 
635 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
636 {
637 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
638 }
639 
640 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
641 {
642 	u16 selector;
643 	struct desc_struct desc;
644 
645 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
646 	return selector;
647 }
648 
649 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
650 				 unsigned seg)
651 {
652 	u16 dummy;
653 	u32 base3;
654 	struct desc_struct desc;
655 
656 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
657 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
658 }
659 
660 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
661 {
662 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
663 }
664 
665 static inline bool emul_is_noncanonical_address(u64 la,
666 						struct x86_emulate_ctxt *ctxt)
667 {
668 	return get_canonical(la, ctxt_virt_addr_bits(ctxt)) != la;
669 }
670 
671 /*
672  * x86 defines three classes of vector instructions: explicitly
673  * aligned, explicitly unaligned, and the rest, which change behaviour
674  * depending on whether they're AVX encoded or not.
675  *
676  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
677  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
678  * 512 bytes of data must be aligned to a 16 byte boundary.
679  */
680 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
681 {
682 	u64 alignment = ctxt->d & AlignMask;
683 
684 	if (likely(size < 16))
685 		return 1;
686 
687 	switch (alignment) {
688 	case Unaligned:
689 	case Avx:
690 		return 1;
691 	case Aligned16:
692 		return 16;
693 	case Aligned:
694 	default:
695 		return size;
696 	}
697 }
698 
699 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
700 				       struct segmented_address addr,
701 				       unsigned *max_size, unsigned size,
702 				       bool write, bool fetch,
703 				       enum x86emul_mode mode, ulong *linear)
704 {
705 	struct desc_struct desc;
706 	bool usable;
707 	ulong la;
708 	u32 lim;
709 	u16 sel;
710 	u8  va_bits;
711 
712 	la = seg_base(ctxt, addr.seg) + addr.ea;
713 	*max_size = 0;
714 	switch (mode) {
715 	case X86EMUL_MODE_PROT64:
716 		*linear = la;
717 		va_bits = ctxt_virt_addr_bits(ctxt);
718 		if (get_canonical(la, va_bits) != la)
719 			goto bad;
720 
721 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
722 		if (size > *max_size)
723 			goto bad;
724 		break;
725 	default:
726 		*linear = la = (u32)la;
727 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
728 						addr.seg);
729 		if (!usable)
730 			goto bad;
731 		/* code segment in protected mode or read-only data segment */
732 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
733 					|| !(desc.type & 2)) && write)
734 			goto bad;
735 		/* unreadable code segment */
736 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
737 			goto bad;
738 		lim = desc_limit_scaled(&desc);
739 		if (!(desc.type & 8) && (desc.type & 4)) {
740 			/* expand-down segment */
741 			if (addr.ea <= lim)
742 				goto bad;
743 			lim = desc.d ? 0xffffffff : 0xffff;
744 		}
745 		if (addr.ea > lim)
746 			goto bad;
747 		if (lim == 0xffffffff)
748 			*max_size = ~0u;
749 		else {
750 			*max_size = (u64)lim + 1 - addr.ea;
751 			if (size > *max_size)
752 				goto bad;
753 		}
754 		break;
755 	}
756 	if (la & (insn_alignment(ctxt, size) - 1))
757 		return emulate_gp(ctxt, 0);
758 	return X86EMUL_CONTINUE;
759 bad:
760 	if (addr.seg == VCPU_SREG_SS)
761 		return emulate_ss(ctxt, 0);
762 	else
763 		return emulate_gp(ctxt, 0);
764 }
765 
766 static int linearize(struct x86_emulate_ctxt *ctxt,
767 		     struct segmented_address addr,
768 		     unsigned size, bool write,
769 		     ulong *linear)
770 {
771 	unsigned max_size;
772 	return __linearize(ctxt, addr, &max_size, size, write, false,
773 			   ctxt->mode, linear);
774 }
775 
776 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
777 			     enum x86emul_mode mode)
778 {
779 	ulong linear;
780 	int rc;
781 	unsigned max_size;
782 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
783 					   .ea = dst };
784 
785 	if (ctxt->op_bytes != sizeof(unsigned long))
786 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
787 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
788 	if (rc == X86EMUL_CONTINUE)
789 		ctxt->_eip = addr.ea;
790 	return rc;
791 }
792 
793 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
794 {
795 	return assign_eip(ctxt, dst, ctxt->mode);
796 }
797 
798 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
799 			  const struct desc_struct *cs_desc)
800 {
801 	enum x86emul_mode mode = ctxt->mode;
802 	int rc;
803 
804 #ifdef CONFIG_X86_64
805 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
806 		if (cs_desc->l) {
807 			u64 efer = 0;
808 
809 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
810 			if (efer & EFER_LMA)
811 				mode = X86EMUL_MODE_PROT64;
812 		} else
813 			mode = X86EMUL_MODE_PROT32; /* temporary value */
814 	}
815 #endif
816 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
817 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
818 	rc = assign_eip(ctxt, dst, mode);
819 	if (rc == X86EMUL_CONTINUE)
820 		ctxt->mode = mode;
821 	return rc;
822 }
823 
824 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
825 {
826 	return assign_eip_near(ctxt, ctxt->_eip + rel);
827 }
828 
829 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear,
830 			      void *data, unsigned size)
831 {
832 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true);
833 }
834 
835 static int linear_write_system(struct x86_emulate_ctxt *ctxt,
836 			       ulong linear, void *data,
837 			       unsigned int size)
838 {
839 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true);
840 }
841 
842 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
843 			      struct segmented_address addr,
844 			      void *data,
845 			      unsigned size)
846 {
847 	int rc;
848 	ulong linear;
849 
850 	rc = linearize(ctxt, addr, size, false, &linear);
851 	if (rc != X86EMUL_CONTINUE)
852 		return rc;
853 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false);
854 }
855 
856 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
857 			       struct segmented_address addr,
858 			       void *data,
859 			       unsigned int size)
860 {
861 	int rc;
862 	ulong linear;
863 
864 	rc = linearize(ctxt, addr, size, true, &linear);
865 	if (rc != X86EMUL_CONTINUE)
866 		return rc;
867 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false);
868 }
869 
870 /*
871  * Prefetch the remaining bytes of the instruction without crossing page
872  * boundary if they are not in fetch_cache yet.
873  */
874 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
875 {
876 	int rc;
877 	unsigned size, max_size;
878 	unsigned long linear;
879 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
880 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
881 					   .ea = ctxt->eip + cur_size };
882 
883 	/*
884 	 * We do not know exactly how many bytes will be needed, and
885 	 * __linearize is expensive, so fetch as much as possible.  We
886 	 * just have to avoid going beyond the 15 byte limit, the end
887 	 * of the segment, or the end of the page.
888 	 *
889 	 * __linearize is called with size 0 so that it does not do any
890 	 * boundary check itself.  Instead, we use max_size to check
891 	 * against op_size.
892 	 */
893 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
894 			 &linear);
895 	if (unlikely(rc != X86EMUL_CONTINUE))
896 		return rc;
897 
898 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
899 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
900 
901 	/*
902 	 * One instruction can only straddle two pages,
903 	 * and one has been loaded at the beginning of
904 	 * x86_decode_insn.  So, if not enough bytes
905 	 * still, we must have hit the 15-byte boundary.
906 	 */
907 	if (unlikely(size < op_size))
908 		return emulate_gp(ctxt, 0);
909 
910 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
911 			      size, &ctxt->exception);
912 	if (unlikely(rc != X86EMUL_CONTINUE))
913 		return rc;
914 	ctxt->fetch.end += size;
915 	return X86EMUL_CONTINUE;
916 }
917 
918 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
919 					       unsigned size)
920 {
921 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
922 
923 	if (unlikely(done_size < size))
924 		return __do_insn_fetch_bytes(ctxt, size - done_size);
925 	else
926 		return X86EMUL_CONTINUE;
927 }
928 
929 /* Fetch next part of the instruction being emulated. */
930 #define insn_fetch(_type, _ctxt)					\
931 ({	_type _x;							\
932 									\
933 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
934 	if (rc != X86EMUL_CONTINUE)					\
935 		goto done;						\
936 	ctxt->_eip += sizeof(_type);					\
937 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
938 	ctxt->fetch.ptr += sizeof(_type);				\
939 	_x;								\
940 })
941 
942 #define insn_fetch_arr(_arr, _size, _ctxt)				\
943 ({									\
944 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
945 	if (rc != X86EMUL_CONTINUE)					\
946 		goto done;						\
947 	ctxt->_eip += (_size);						\
948 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
949 	ctxt->fetch.ptr += (_size);					\
950 })
951 
952 /*
953  * Given the 'reg' portion of a ModRM byte, and a register block, return a
954  * pointer into the block that addresses the relevant register.
955  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
956  */
957 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
958 			     int byteop)
959 {
960 	void *p;
961 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
962 
963 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
964 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
965 	else
966 		p = reg_rmw(ctxt, modrm_reg);
967 	return p;
968 }
969 
970 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
971 			   struct segmented_address addr,
972 			   u16 *size, unsigned long *address, int op_bytes)
973 {
974 	int rc;
975 
976 	if (op_bytes == 2)
977 		op_bytes = 3;
978 	*address = 0;
979 	rc = segmented_read_std(ctxt, addr, size, 2);
980 	if (rc != X86EMUL_CONTINUE)
981 		return rc;
982 	addr.ea += 2;
983 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
984 	return rc;
985 }
986 
987 FASTOP2(add);
988 FASTOP2(or);
989 FASTOP2(adc);
990 FASTOP2(sbb);
991 FASTOP2(and);
992 FASTOP2(sub);
993 FASTOP2(xor);
994 FASTOP2(cmp);
995 FASTOP2(test);
996 
997 FASTOP1SRC2(mul, mul_ex);
998 FASTOP1SRC2(imul, imul_ex);
999 FASTOP1SRC2EX(div, div_ex);
1000 FASTOP1SRC2EX(idiv, idiv_ex);
1001 
1002 FASTOP3WCL(shld);
1003 FASTOP3WCL(shrd);
1004 
1005 FASTOP2W(imul);
1006 
1007 FASTOP1(not);
1008 FASTOP1(neg);
1009 FASTOP1(inc);
1010 FASTOP1(dec);
1011 
1012 FASTOP2CL(rol);
1013 FASTOP2CL(ror);
1014 FASTOP2CL(rcl);
1015 FASTOP2CL(rcr);
1016 FASTOP2CL(shl);
1017 FASTOP2CL(shr);
1018 FASTOP2CL(sar);
1019 
1020 FASTOP2W(bsf);
1021 FASTOP2W(bsr);
1022 FASTOP2W(bt);
1023 FASTOP2W(bts);
1024 FASTOP2W(btr);
1025 FASTOP2W(btc);
1026 
1027 FASTOP2(xadd);
1028 
1029 FASTOP2R(cmp, cmp_r);
1030 
1031 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1032 {
1033 	/* If src is zero, do not writeback, but update flags */
1034 	if (ctxt->src.val == 0)
1035 		ctxt->dst.type = OP_NONE;
1036 	return fastop(ctxt, em_bsf);
1037 }
1038 
1039 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1040 {
1041 	/* If src is zero, do not writeback, but update flags */
1042 	if (ctxt->src.val == 0)
1043 		ctxt->dst.type = OP_NONE;
1044 	return fastop(ctxt, em_bsr);
1045 }
1046 
1047 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1048 {
1049 	u8 rc;
1050 	void (*fop)(void) = (void *)em_setcc + 4 * (condition & 0xf);
1051 
1052 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1053 	asm("push %[flags]; popf; " CALL_NOSPEC
1054 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1055 	return rc;
1056 }
1057 
1058 static void fetch_register_operand(struct operand *op)
1059 {
1060 	switch (op->bytes) {
1061 	case 1:
1062 		op->val = *(u8 *)op->addr.reg;
1063 		break;
1064 	case 2:
1065 		op->val = *(u16 *)op->addr.reg;
1066 		break;
1067 	case 4:
1068 		op->val = *(u32 *)op->addr.reg;
1069 		break;
1070 	case 8:
1071 		op->val = *(u64 *)op->addr.reg;
1072 		break;
1073 	}
1074 }
1075 
1076 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1077 {
1078 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1079 		return emulate_nm(ctxt);
1080 
1081 	kvm_fpu_get();
1082 	asm volatile("fninit");
1083 	kvm_fpu_put();
1084 	return X86EMUL_CONTINUE;
1085 }
1086 
1087 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1088 {
1089 	u16 fcw;
1090 
1091 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1092 		return emulate_nm(ctxt);
1093 
1094 	kvm_fpu_get();
1095 	asm volatile("fnstcw %0": "+m"(fcw));
1096 	kvm_fpu_put();
1097 
1098 	ctxt->dst.val = fcw;
1099 
1100 	return X86EMUL_CONTINUE;
1101 }
1102 
1103 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1104 {
1105 	u16 fsw;
1106 
1107 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1108 		return emulate_nm(ctxt);
1109 
1110 	kvm_fpu_get();
1111 	asm volatile("fnstsw %0": "+m"(fsw));
1112 	kvm_fpu_put();
1113 
1114 	ctxt->dst.val = fsw;
1115 
1116 	return X86EMUL_CONTINUE;
1117 }
1118 
1119 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1120 				    struct operand *op)
1121 {
1122 	unsigned reg = ctxt->modrm_reg;
1123 
1124 	if (!(ctxt->d & ModRM))
1125 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1126 
1127 	if (ctxt->d & Sse) {
1128 		op->type = OP_XMM;
1129 		op->bytes = 16;
1130 		op->addr.xmm = reg;
1131 		kvm_read_sse_reg(reg, &op->vec_val);
1132 		return;
1133 	}
1134 	if (ctxt->d & Mmx) {
1135 		reg &= 7;
1136 		op->type = OP_MM;
1137 		op->bytes = 8;
1138 		op->addr.mm = reg;
1139 		return;
1140 	}
1141 
1142 	op->type = OP_REG;
1143 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1144 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1145 
1146 	fetch_register_operand(op);
1147 	op->orig_val = op->val;
1148 }
1149 
1150 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1151 {
1152 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1153 		ctxt->modrm_seg = VCPU_SREG_SS;
1154 }
1155 
1156 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1157 			struct operand *op)
1158 {
1159 	u8 sib;
1160 	int index_reg, base_reg, scale;
1161 	int rc = X86EMUL_CONTINUE;
1162 	ulong modrm_ea = 0;
1163 
1164 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1165 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1166 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1167 
1168 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1169 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1170 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1171 	ctxt->modrm_seg = VCPU_SREG_DS;
1172 
1173 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1174 		op->type = OP_REG;
1175 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1176 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1177 				ctxt->d & ByteOp);
1178 		if (ctxt->d & Sse) {
1179 			op->type = OP_XMM;
1180 			op->bytes = 16;
1181 			op->addr.xmm = ctxt->modrm_rm;
1182 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1183 			return rc;
1184 		}
1185 		if (ctxt->d & Mmx) {
1186 			op->type = OP_MM;
1187 			op->bytes = 8;
1188 			op->addr.mm = ctxt->modrm_rm & 7;
1189 			return rc;
1190 		}
1191 		fetch_register_operand(op);
1192 		return rc;
1193 	}
1194 
1195 	op->type = OP_MEM;
1196 
1197 	if (ctxt->ad_bytes == 2) {
1198 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1199 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1200 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1201 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1202 
1203 		/* 16-bit ModR/M decode. */
1204 		switch (ctxt->modrm_mod) {
1205 		case 0:
1206 			if (ctxt->modrm_rm == 6)
1207 				modrm_ea += insn_fetch(u16, ctxt);
1208 			break;
1209 		case 1:
1210 			modrm_ea += insn_fetch(s8, ctxt);
1211 			break;
1212 		case 2:
1213 			modrm_ea += insn_fetch(u16, ctxt);
1214 			break;
1215 		}
1216 		switch (ctxt->modrm_rm) {
1217 		case 0:
1218 			modrm_ea += bx + si;
1219 			break;
1220 		case 1:
1221 			modrm_ea += bx + di;
1222 			break;
1223 		case 2:
1224 			modrm_ea += bp + si;
1225 			break;
1226 		case 3:
1227 			modrm_ea += bp + di;
1228 			break;
1229 		case 4:
1230 			modrm_ea += si;
1231 			break;
1232 		case 5:
1233 			modrm_ea += di;
1234 			break;
1235 		case 6:
1236 			if (ctxt->modrm_mod != 0)
1237 				modrm_ea += bp;
1238 			break;
1239 		case 7:
1240 			modrm_ea += bx;
1241 			break;
1242 		}
1243 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1244 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1245 			ctxt->modrm_seg = VCPU_SREG_SS;
1246 		modrm_ea = (u16)modrm_ea;
1247 	} else {
1248 		/* 32/64-bit ModR/M decode. */
1249 		if ((ctxt->modrm_rm & 7) == 4) {
1250 			sib = insn_fetch(u8, ctxt);
1251 			index_reg |= (sib >> 3) & 7;
1252 			base_reg |= sib & 7;
1253 			scale = sib >> 6;
1254 
1255 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1256 				modrm_ea += insn_fetch(s32, ctxt);
1257 			else {
1258 				modrm_ea += reg_read(ctxt, base_reg);
1259 				adjust_modrm_seg(ctxt, base_reg);
1260 				/* Increment ESP on POP [ESP] */
1261 				if ((ctxt->d & IncSP) &&
1262 				    base_reg == VCPU_REGS_RSP)
1263 					modrm_ea += ctxt->op_bytes;
1264 			}
1265 			if (index_reg != 4)
1266 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1267 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1268 			modrm_ea += insn_fetch(s32, ctxt);
1269 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1270 				ctxt->rip_relative = 1;
1271 		} else {
1272 			base_reg = ctxt->modrm_rm;
1273 			modrm_ea += reg_read(ctxt, base_reg);
1274 			adjust_modrm_seg(ctxt, base_reg);
1275 		}
1276 		switch (ctxt->modrm_mod) {
1277 		case 1:
1278 			modrm_ea += insn_fetch(s8, ctxt);
1279 			break;
1280 		case 2:
1281 			modrm_ea += insn_fetch(s32, ctxt);
1282 			break;
1283 		}
1284 	}
1285 	op->addr.mem.ea = modrm_ea;
1286 	if (ctxt->ad_bytes != 8)
1287 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1288 
1289 done:
1290 	return rc;
1291 }
1292 
1293 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1294 		      struct operand *op)
1295 {
1296 	int rc = X86EMUL_CONTINUE;
1297 
1298 	op->type = OP_MEM;
1299 	switch (ctxt->ad_bytes) {
1300 	case 2:
1301 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1302 		break;
1303 	case 4:
1304 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1305 		break;
1306 	case 8:
1307 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1308 		break;
1309 	}
1310 done:
1311 	return rc;
1312 }
1313 
1314 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1315 {
1316 	long sv = 0, mask;
1317 
1318 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1319 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1320 
1321 		if (ctxt->src.bytes == 2)
1322 			sv = (s16)ctxt->src.val & (s16)mask;
1323 		else if (ctxt->src.bytes == 4)
1324 			sv = (s32)ctxt->src.val & (s32)mask;
1325 		else
1326 			sv = (s64)ctxt->src.val & (s64)mask;
1327 
1328 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1329 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1330 	}
1331 
1332 	/* only subword offset */
1333 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1334 }
1335 
1336 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1337 			 unsigned long addr, void *dest, unsigned size)
1338 {
1339 	int rc;
1340 	struct read_cache *mc = &ctxt->mem_read;
1341 
1342 	if (mc->pos < mc->end)
1343 		goto read_cached;
1344 
1345 	WARN_ON((mc->end + size) >= sizeof(mc->data));
1346 
1347 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1348 				      &ctxt->exception);
1349 	if (rc != X86EMUL_CONTINUE)
1350 		return rc;
1351 
1352 	mc->end += size;
1353 
1354 read_cached:
1355 	memcpy(dest, mc->data + mc->pos, size);
1356 	mc->pos += size;
1357 	return X86EMUL_CONTINUE;
1358 }
1359 
1360 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1361 			  struct segmented_address addr,
1362 			  void *data,
1363 			  unsigned size)
1364 {
1365 	int rc;
1366 	ulong linear;
1367 
1368 	rc = linearize(ctxt, addr, size, false, &linear);
1369 	if (rc != X86EMUL_CONTINUE)
1370 		return rc;
1371 	return read_emulated(ctxt, linear, data, size);
1372 }
1373 
1374 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1375 			   struct segmented_address addr,
1376 			   const void *data,
1377 			   unsigned size)
1378 {
1379 	int rc;
1380 	ulong linear;
1381 
1382 	rc = linearize(ctxt, addr, size, true, &linear);
1383 	if (rc != X86EMUL_CONTINUE)
1384 		return rc;
1385 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1386 					 &ctxt->exception);
1387 }
1388 
1389 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1390 			     struct segmented_address addr,
1391 			     const void *orig_data, const void *data,
1392 			     unsigned size)
1393 {
1394 	int rc;
1395 	ulong linear;
1396 
1397 	rc = linearize(ctxt, addr, size, true, &linear);
1398 	if (rc != X86EMUL_CONTINUE)
1399 		return rc;
1400 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1401 					   size, &ctxt->exception);
1402 }
1403 
1404 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1405 			   unsigned int size, unsigned short port,
1406 			   void *dest)
1407 {
1408 	struct read_cache *rc = &ctxt->io_read;
1409 
1410 	if (rc->pos == rc->end) { /* refill pio read ahead */
1411 		unsigned int in_page, n;
1412 		unsigned int count = ctxt->rep_prefix ?
1413 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1414 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1415 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1416 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1417 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1418 		if (n == 0)
1419 			n = 1;
1420 		rc->pos = rc->end = 0;
1421 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1422 			return 0;
1423 		rc->end = n * size;
1424 	}
1425 
1426 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1427 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1428 		ctxt->dst.data = rc->data + rc->pos;
1429 		ctxt->dst.type = OP_MEM_STR;
1430 		ctxt->dst.count = (rc->end - rc->pos) / size;
1431 		rc->pos = rc->end;
1432 	} else {
1433 		memcpy(dest, rc->data + rc->pos, size);
1434 		rc->pos += size;
1435 	}
1436 	return 1;
1437 }
1438 
1439 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1440 				     u16 index, struct desc_struct *desc)
1441 {
1442 	struct desc_ptr dt;
1443 	ulong addr;
1444 
1445 	ctxt->ops->get_idt(ctxt, &dt);
1446 
1447 	if (dt.size < index * 8 + 7)
1448 		return emulate_gp(ctxt, index << 3 | 0x2);
1449 
1450 	addr = dt.address + index * 8;
1451 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1452 }
1453 
1454 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1455 				     u16 selector, struct desc_ptr *dt)
1456 {
1457 	const struct x86_emulate_ops *ops = ctxt->ops;
1458 	u32 base3 = 0;
1459 
1460 	if (selector & 1 << 2) {
1461 		struct desc_struct desc;
1462 		u16 sel;
1463 
1464 		memset(dt, 0, sizeof(*dt));
1465 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1466 				      VCPU_SREG_LDTR))
1467 			return;
1468 
1469 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1470 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1471 	} else
1472 		ops->get_gdt(ctxt, dt);
1473 }
1474 
1475 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1476 			      u16 selector, ulong *desc_addr_p)
1477 {
1478 	struct desc_ptr dt;
1479 	u16 index = selector >> 3;
1480 	ulong addr;
1481 
1482 	get_descriptor_table_ptr(ctxt, selector, &dt);
1483 
1484 	if (dt.size < index * 8 + 7)
1485 		return emulate_gp(ctxt, selector & 0xfffc);
1486 
1487 	addr = dt.address + index * 8;
1488 
1489 #ifdef CONFIG_X86_64
1490 	if (addr >> 32 != 0) {
1491 		u64 efer = 0;
1492 
1493 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1494 		if (!(efer & EFER_LMA))
1495 			addr &= (u32)-1;
1496 	}
1497 #endif
1498 
1499 	*desc_addr_p = addr;
1500 	return X86EMUL_CONTINUE;
1501 }
1502 
1503 /* allowed just for 8 bytes segments */
1504 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1505 				   u16 selector, struct desc_struct *desc,
1506 				   ulong *desc_addr_p)
1507 {
1508 	int rc;
1509 
1510 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1511 	if (rc != X86EMUL_CONTINUE)
1512 		return rc;
1513 
1514 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1515 }
1516 
1517 /* allowed just for 8 bytes segments */
1518 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1519 				    u16 selector, struct desc_struct *desc)
1520 {
1521 	int rc;
1522 	ulong addr;
1523 
1524 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1525 	if (rc != X86EMUL_CONTINUE)
1526 		return rc;
1527 
1528 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1529 }
1530 
1531 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1532 				     u16 selector, int seg, u8 cpl,
1533 				     enum x86_transfer_type transfer,
1534 				     struct desc_struct *desc)
1535 {
1536 	struct desc_struct seg_desc, old_desc;
1537 	u8 dpl, rpl;
1538 	unsigned err_vec = GP_VECTOR;
1539 	u32 err_code = 0;
1540 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1541 	ulong desc_addr;
1542 	int ret;
1543 	u16 dummy;
1544 	u32 base3 = 0;
1545 
1546 	memset(&seg_desc, 0, sizeof(seg_desc));
1547 
1548 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1549 		/* set real mode segment descriptor (keep limit etc. for
1550 		 * unreal mode) */
1551 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1552 		set_desc_base(&seg_desc, selector << 4);
1553 		goto load;
1554 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1555 		/* VM86 needs a clean new segment descriptor */
1556 		set_desc_base(&seg_desc, selector << 4);
1557 		set_desc_limit(&seg_desc, 0xffff);
1558 		seg_desc.type = 3;
1559 		seg_desc.p = 1;
1560 		seg_desc.s = 1;
1561 		seg_desc.dpl = 3;
1562 		goto load;
1563 	}
1564 
1565 	rpl = selector & 3;
1566 
1567 	/* TR should be in GDT only */
1568 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1569 		goto exception;
1570 
1571 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1572 	if (null_selector) {
1573 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1574 			goto exception;
1575 
1576 		if (seg == VCPU_SREG_SS) {
1577 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1578 				goto exception;
1579 
1580 			/*
1581 			 * ctxt->ops->set_segment expects the CPL to be in
1582 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1583 			 */
1584 			seg_desc.type = 3;
1585 			seg_desc.p = 1;
1586 			seg_desc.s = 1;
1587 			seg_desc.dpl = cpl;
1588 			seg_desc.d = 1;
1589 			seg_desc.g = 1;
1590 		}
1591 
1592 		/* Skip all following checks */
1593 		goto load;
1594 	}
1595 
1596 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1597 	if (ret != X86EMUL_CONTINUE)
1598 		return ret;
1599 
1600 	err_code = selector & 0xfffc;
1601 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1602 							   GP_VECTOR;
1603 
1604 	/* can't load system descriptor into segment selector */
1605 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1606 		if (transfer == X86_TRANSFER_CALL_JMP)
1607 			return X86EMUL_UNHANDLEABLE;
1608 		goto exception;
1609 	}
1610 
1611 	if (!seg_desc.p) {
1612 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1613 		goto exception;
1614 	}
1615 
1616 	dpl = seg_desc.dpl;
1617 
1618 	switch (seg) {
1619 	case VCPU_SREG_SS:
1620 		/*
1621 		 * segment is not a writable data segment or segment
1622 		 * selector's RPL != CPL or segment selector's RPL != CPL
1623 		 */
1624 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1625 			goto exception;
1626 		break;
1627 	case VCPU_SREG_CS:
1628 		if (!(seg_desc.type & 8))
1629 			goto exception;
1630 
1631 		if (seg_desc.type & 4) {
1632 			/* conforming */
1633 			if (dpl > cpl)
1634 				goto exception;
1635 		} else {
1636 			/* nonconforming */
1637 			if (rpl > cpl || dpl != cpl)
1638 				goto exception;
1639 		}
1640 		/* in long-mode d/b must be clear if l is set */
1641 		if (seg_desc.d && seg_desc.l) {
1642 			u64 efer = 0;
1643 
1644 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1645 			if (efer & EFER_LMA)
1646 				goto exception;
1647 		}
1648 
1649 		/* CS(RPL) <- CPL */
1650 		selector = (selector & 0xfffc) | cpl;
1651 		break;
1652 	case VCPU_SREG_TR:
1653 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1654 			goto exception;
1655 		old_desc = seg_desc;
1656 		seg_desc.type |= 2; /* busy */
1657 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1658 						  sizeof(seg_desc), &ctxt->exception);
1659 		if (ret != X86EMUL_CONTINUE)
1660 			return ret;
1661 		break;
1662 	case VCPU_SREG_LDTR:
1663 		if (seg_desc.s || seg_desc.type != 2)
1664 			goto exception;
1665 		break;
1666 	default: /*  DS, ES, FS, or GS */
1667 		/*
1668 		 * segment is not a data or readable code segment or
1669 		 * ((segment is a data or nonconforming code segment)
1670 		 * and (both RPL and CPL > DPL))
1671 		 */
1672 		if ((seg_desc.type & 0xa) == 0x8 ||
1673 		    (((seg_desc.type & 0xc) != 0xc) &&
1674 		     (rpl > dpl && cpl > dpl)))
1675 			goto exception;
1676 		break;
1677 	}
1678 
1679 	if (seg_desc.s) {
1680 		/* mark segment as accessed */
1681 		if (!(seg_desc.type & 1)) {
1682 			seg_desc.type |= 1;
1683 			ret = write_segment_descriptor(ctxt, selector,
1684 						       &seg_desc);
1685 			if (ret != X86EMUL_CONTINUE)
1686 				return ret;
1687 		}
1688 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1689 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1690 		if (ret != X86EMUL_CONTINUE)
1691 			return ret;
1692 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1693 				((u64)base3 << 32), ctxt))
1694 			return emulate_gp(ctxt, 0);
1695 	}
1696 load:
1697 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1698 	if (desc)
1699 		*desc = seg_desc;
1700 	return X86EMUL_CONTINUE;
1701 exception:
1702 	return emulate_exception(ctxt, err_vec, err_code, true);
1703 }
1704 
1705 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1706 				   u16 selector, int seg)
1707 {
1708 	u8 cpl = ctxt->ops->cpl(ctxt);
1709 
1710 	/*
1711 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1712 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1713 	 * but it's wrong).
1714 	 *
1715 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1716 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1717 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1718 	 * and only forbid it here.
1719 	 */
1720 	if (seg == VCPU_SREG_SS && selector == 3 &&
1721 	    ctxt->mode == X86EMUL_MODE_PROT64)
1722 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1723 
1724 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1725 					 X86_TRANSFER_NONE, NULL);
1726 }
1727 
1728 static void write_register_operand(struct operand *op)
1729 {
1730 	return assign_register(op->addr.reg, op->val, op->bytes);
1731 }
1732 
1733 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1734 {
1735 	switch (op->type) {
1736 	case OP_REG:
1737 		write_register_operand(op);
1738 		break;
1739 	case OP_MEM:
1740 		if (ctxt->lock_prefix)
1741 			return segmented_cmpxchg(ctxt,
1742 						 op->addr.mem,
1743 						 &op->orig_val,
1744 						 &op->val,
1745 						 op->bytes);
1746 		else
1747 			return segmented_write(ctxt,
1748 					       op->addr.mem,
1749 					       &op->val,
1750 					       op->bytes);
1751 		break;
1752 	case OP_MEM_STR:
1753 		return segmented_write(ctxt,
1754 				       op->addr.mem,
1755 				       op->data,
1756 				       op->bytes * op->count);
1757 		break;
1758 	case OP_XMM:
1759 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1760 		break;
1761 	case OP_MM:
1762 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1763 		break;
1764 	case OP_NONE:
1765 		/* no writeback */
1766 		break;
1767 	default:
1768 		break;
1769 	}
1770 	return X86EMUL_CONTINUE;
1771 }
1772 
1773 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1774 {
1775 	struct segmented_address addr;
1776 
1777 	rsp_increment(ctxt, -bytes);
1778 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1779 	addr.seg = VCPU_SREG_SS;
1780 
1781 	return segmented_write(ctxt, addr, data, bytes);
1782 }
1783 
1784 static int em_push(struct x86_emulate_ctxt *ctxt)
1785 {
1786 	/* Disable writeback. */
1787 	ctxt->dst.type = OP_NONE;
1788 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1789 }
1790 
1791 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1792 		       void *dest, int len)
1793 {
1794 	int rc;
1795 	struct segmented_address addr;
1796 
1797 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1798 	addr.seg = VCPU_SREG_SS;
1799 	rc = segmented_read(ctxt, addr, dest, len);
1800 	if (rc != X86EMUL_CONTINUE)
1801 		return rc;
1802 
1803 	rsp_increment(ctxt, len);
1804 	return rc;
1805 }
1806 
1807 static int em_pop(struct x86_emulate_ctxt *ctxt)
1808 {
1809 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1810 }
1811 
1812 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1813 			void *dest, int len)
1814 {
1815 	int rc;
1816 	unsigned long val, change_mask;
1817 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1818 	int cpl = ctxt->ops->cpl(ctxt);
1819 
1820 	rc = emulate_pop(ctxt, &val, len);
1821 	if (rc != X86EMUL_CONTINUE)
1822 		return rc;
1823 
1824 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1825 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1826 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1827 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1828 
1829 	switch(ctxt->mode) {
1830 	case X86EMUL_MODE_PROT64:
1831 	case X86EMUL_MODE_PROT32:
1832 	case X86EMUL_MODE_PROT16:
1833 		if (cpl == 0)
1834 			change_mask |= X86_EFLAGS_IOPL;
1835 		if (cpl <= iopl)
1836 			change_mask |= X86_EFLAGS_IF;
1837 		break;
1838 	case X86EMUL_MODE_VM86:
1839 		if (iopl < 3)
1840 			return emulate_gp(ctxt, 0);
1841 		change_mask |= X86_EFLAGS_IF;
1842 		break;
1843 	default: /* real mode */
1844 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1845 		break;
1846 	}
1847 
1848 	*(unsigned long *)dest =
1849 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1850 
1851 	return rc;
1852 }
1853 
1854 static int em_popf(struct x86_emulate_ctxt *ctxt)
1855 {
1856 	ctxt->dst.type = OP_REG;
1857 	ctxt->dst.addr.reg = &ctxt->eflags;
1858 	ctxt->dst.bytes = ctxt->op_bytes;
1859 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1860 }
1861 
1862 static int em_enter(struct x86_emulate_ctxt *ctxt)
1863 {
1864 	int rc;
1865 	unsigned frame_size = ctxt->src.val;
1866 	unsigned nesting_level = ctxt->src2.val & 31;
1867 	ulong rbp;
1868 
1869 	if (nesting_level)
1870 		return X86EMUL_UNHANDLEABLE;
1871 
1872 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1873 	rc = push(ctxt, &rbp, stack_size(ctxt));
1874 	if (rc != X86EMUL_CONTINUE)
1875 		return rc;
1876 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1877 		      stack_mask(ctxt));
1878 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1879 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1880 		      stack_mask(ctxt));
1881 	return X86EMUL_CONTINUE;
1882 }
1883 
1884 static int em_leave(struct x86_emulate_ctxt *ctxt)
1885 {
1886 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1887 		      stack_mask(ctxt));
1888 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1889 }
1890 
1891 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1892 {
1893 	int seg = ctxt->src2.val;
1894 
1895 	ctxt->src.val = get_segment_selector(ctxt, seg);
1896 	if (ctxt->op_bytes == 4) {
1897 		rsp_increment(ctxt, -2);
1898 		ctxt->op_bytes = 2;
1899 	}
1900 
1901 	return em_push(ctxt);
1902 }
1903 
1904 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1905 {
1906 	int seg = ctxt->src2.val;
1907 	unsigned long selector;
1908 	int rc;
1909 
1910 	rc = emulate_pop(ctxt, &selector, 2);
1911 	if (rc != X86EMUL_CONTINUE)
1912 		return rc;
1913 
1914 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1915 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1916 	if (ctxt->op_bytes > 2)
1917 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1918 
1919 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1920 	return rc;
1921 }
1922 
1923 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1924 {
1925 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1926 	int rc = X86EMUL_CONTINUE;
1927 	int reg = VCPU_REGS_RAX;
1928 
1929 	while (reg <= VCPU_REGS_RDI) {
1930 		(reg == VCPU_REGS_RSP) ?
1931 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1932 
1933 		rc = em_push(ctxt);
1934 		if (rc != X86EMUL_CONTINUE)
1935 			return rc;
1936 
1937 		++reg;
1938 	}
1939 
1940 	return rc;
1941 }
1942 
1943 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1944 {
1945 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1946 	return em_push(ctxt);
1947 }
1948 
1949 static int em_popa(struct x86_emulate_ctxt *ctxt)
1950 {
1951 	int rc = X86EMUL_CONTINUE;
1952 	int reg = VCPU_REGS_RDI;
1953 	u32 val;
1954 
1955 	while (reg >= VCPU_REGS_RAX) {
1956 		if (reg == VCPU_REGS_RSP) {
1957 			rsp_increment(ctxt, ctxt->op_bytes);
1958 			--reg;
1959 		}
1960 
1961 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
1962 		if (rc != X86EMUL_CONTINUE)
1963 			break;
1964 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
1965 		--reg;
1966 	}
1967 	return rc;
1968 }
1969 
1970 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
1971 {
1972 	const struct x86_emulate_ops *ops = ctxt->ops;
1973 	int rc;
1974 	struct desc_ptr dt;
1975 	gva_t cs_addr;
1976 	gva_t eip_addr;
1977 	u16 cs, eip;
1978 
1979 	/* TODO: Add limit checks */
1980 	ctxt->src.val = ctxt->eflags;
1981 	rc = em_push(ctxt);
1982 	if (rc != X86EMUL_CONTINUE)
1983 		return rc;
1984 
1985 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
1986 
1987 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
1988 	rc = em_push(ctxt);
1989 	if (rc != X86EMUL_CONTINUE)
1990 		return rc;
1991 
1992 	ctxt->src.val = ctxt->_eip;
1993 	rc = em_push(ctxt);
1994 	if (rc != X86EMUL_CONTINUE)
1995 		return rc;
1996 
1997 	ops->get_idt(ctxt, &dt);
1998 
1999 	eip_addr = dt.address + (irq << 2);
2000 	cs_addr = dt.address + (irq << 2) + 2;
2001 
2002 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2003 	if (rc != X86EMUL_CONTINUE)
2004 		return rc;
2005 
2006 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2007 	if (rc != X86EMUL_CONTINUE)
2008 		return rc;
2009 
2010 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2011 	if (rc != X86EMUL_CONTINUE)
2012 		return rc;
2013 
2014 	ctxt->_eip = eip;
2015 
2016 	return rc;
2017 }
2018 
2019 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2020 {
2021 	int rc;
2022 
2023 	invalidate_registers(ctxt);
2024 	rc = __emulate_int_real(ctxt, irq);
2025 	if (rc == X86EMUL_CONTINUE)
2026 		writeback_registers(ctxt);
2027 	return rc;
2028 }
2029 
2030 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2031 {
2032 	switch(ctxt->mode) {
2033 	case X86EMUL_MODE_REAL:
2034 		return __emulate_int_real(ctxt, irq);
2035 	case X86EMUL_MODE_VM86:
2036 	case X86EMUL_MODE_PROT16:
2037 	case X86EMUL_MODE_PROT32:
2038 	case X86EMUL_MODE_PROT64:
2039 	default:
2040 		/* Protected mode interrupts unimplemented yet */
2041 		return X86EMUL_UNHANDLEABLE;
2042 	}
2043 }
2044 
2045 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2046 {
2047 	int rc = X86EMUL_CONTINUE;
2048 	unsigned long temp_eip = 0;
2049 	unsigned long temp_eflags = 0;
2050 	unsigned long cs = 0;
2051 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2052 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2053 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2054 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2055 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2056 			     X86_EFLAGS_FIXED;
2057 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2058 				  X86_EFLAGS_VIP;
2059 
2060 	/* TODO: Add stack limit check */
2061 
2062 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2063 
2064 	if (rc != X86EMUL_CONTINUE)
2065 		return rc;
2066 
2067 	if (temp_eip & ~0xffff)
2068 		return emulate_gp(ctxt, 0);
2069 
2070 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2071 
2072 	if (rc != X86EMUL_CONTINUE)
2073 		return rc;
2074 
2075 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2076 
2077 	if (rc != X86EMUL_CONTINUE)
2078 		return rc;
2079 
2080 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2081 
2082 	if (rc != X86EMUL_CONTINUE)
2083 		return rc;
2084 
2085 	ctxt->_eip = temp_eip;
2086 
2087 	if (ctxt->op_bytes == 4)
2088 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2089 	else if (ctxt->op_bytes == 2) {
2090 		ctxt->eflags &= ~0xffff;
2091 		ctxt->eflags |= temp_eflags;
2092 	}
2093 
2094 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2095 	ctxt->eflags |= X86_EFLAGS_FIXED;
2096 	ctxt->ops->set_nmi_mask(ctxt, false);
2097 
2098 	return rc;
2099 }
2100 
2101 static int em_iret(struct x86_emulate_ctxt *ctxt)
2102 {
2103 	switch(ctxt->mode) {
2104 	case X86EMUL_MODE_REAL:
2105 		return emulate_iret_real(ctxt);
2106 	case X86EMUL_MODE_VM86:
2107 	case X86EMUL_MODE_PROT16:
2108 	case X86EMUL_MODE_PROT32:
2109 	case X86EMUL_MODE_PROT64:
2110 	default:
2111 		/* iret from protected mode unimplemented yet */
2112 		return X86EMUL_UNHANDLEABLE;
2113 	}
2114 }
2115 
2116 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2117 {
2118 	int rc;
2119 	unsigned short sel;
2120 	struct desc_struct new_desc;
2121 	u8 cpl = ctxt->ops->cpl(ctxt);
2122 
2123 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2124 
2125 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2126 				       X86_TRANSFER_CALL_JMP,
2127 				       &new_desc);
2128 	if (rc != X86EMUL_CONTINUE)
2129 		return rc;
2130 
2131 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2132 	/* Error handling is not implemented. */
2133 	if (rc != X86EMUL_CONTINUE)
2134 		return X86EMUL_UNHANDLEABLE;
2135 
2136 	return rc;
2137 }
2138 
2139 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2140 {
2141 	return assign_eip_near(ctxt, ctxt->src.val);
2142 }
2143 
2144 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2145 {
2146 	int rc;
2147 	long int old_eip;
2148 
2149 	old_eip = ctxt->_eip;
2150 	rc = assign_eip_near(ctxt, ctxt->src.val);
2151 	if (rc != X86EMUL_CONTINUE)
2152 		return rc;
2153 	ctxt->src.val = old_eip;
2154 	rc = em_push(ctxt);
2155 	return rc;
2156 }
2157 
2158 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2159 {
2160 	u64 old = ctxt->dst.orig_val64;
2161 
2162 	if (ctxt->dst.bytes == 16)
2163 		return X86EMUL_UNHANDLEABLE;
2164 
2165 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2166 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2167 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2168 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2169 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2170 	} else {
2171 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2172 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2173 
2174 		ctxt->eflags |= X86_EFLAGS_ZF;
2175 	}
2176 	return X86EMUL_CONTINUE;
2177 }
2178 
2179 static int em_ret(struct x86_emulate_ctxt *ctxt)
2180 {
2181 	int rc;
2182 	unsigned long eip;
2183 
2184 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2185 	if (rc != X86EMUL_CONTINUE)
2186 		return rc;
2187 
2188 	return assign_eip_near(ctxt, eip);
2189 }
2190 
2191 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2192 {
2193 	int rc;
2194 	unsigned long eip, cs;
2195 	int cpl = ctxt->ops->cpl(ctxt);
2196 	struct desc_struct new_desc;
2197 
2198 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2199 	if (rc != X86EMUL_CONTINUE)
2200 		return rc;
2201 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2202 	if (rc != X86EMUL_CONTINUE)
2203 		return rc;
2204 	/* Outer-privilege level return is not implemented */
2205 	if (ctxt->mode >= X86EMUL_MODE_PROT16 && (cs & 3) > cpl)
2206 		return X86EMUL_UNHANDLEABLE;
2207 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2208 				       X86_TRANSFER_RET,
2209 				       &new_desc);
2210 	if (rc != X86EMUL_CONTINUE)
2211 		return rc;
2212 	rc = assign_eip_far(ctxt, eip, &new_desc);
2213 	/* Error handling is not implemented. */
2214 	if (rc != X86EMUL_CONTINUE)
2215 		return X86EMUL_UNHANDLEABLE;
2216 
2217 	return rc;
2218 }
2219 
2220 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2221 {
2222         int rc;
2223 
2224         rc = em_ret_far(ctxt);
2225         if (rc != X86EMUL_CONTINUE)
2226                 return rc;
2227         rsp_increment(ctxt, ctxt->src.val);
2228         return X86EMUL_CONTINUE;
2229 }
2230 
2231 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2232 {
2233 	/* Save real source value, then compare EAX against destination. */
2234 	ctxt->dst.orig_val = ctxt->dst.val;
2235 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2236 	ctxt->src.orig_val = ctxt->src.val;
2237 	ctxt->src.val = ctxt->dst.orig_val;
2238 	fastop(ctxt, em_cmp);
2239 
2240 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2241 		/* Success: write back to memory; no update of EAX */
2242 		ctxt->src.type = OP_NONE;
2243 		ctxt->dst.val = ctxt->src.orig_val;
2244 	} else {
2245 		/* Failure: write the value we saw to EAX. */
2246 		ctxt->src.type = OP_REG;
2247 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2248 		ctxt->src.val = ctxt->dst.orig_val;
2249 		/* Create write-cycle to dest by writing the same value */
2250 		ctxt->dst.val = ctxt->dst.orig_val;
2251 	}
2252 	return X86EMUL_CONTINUE;
2253 }
2254 
2255 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2256 {
2257 	int seg = ctxt->src2.val;
2258 	unsigned short sel;
2259 	int rc;
2260 
2261 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2262 
2263 	rc = load_segment_descriptor(ctxt, sel, seg);
2264 	if (rc != X86EMUL_CONTINUE)
2265 		return rc;
2266 
2267 	ctxt->dst.val = ctxt->src.val;
2268 	return rc;
2269 }
2270 
2271 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2272 {
2273 #ifdef CONFIG_X86_64
2274 	return ctxt->ops->guest_has_long_mode(ctxt);
2275 #else
2276 	return false;
2277 #endif
2278 }
2279 
2280 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2281 {
2282 	desc->g    = (flags >> 23) & 1;
2283 	desc->d    = (flags >> 22) & 1;
2284 	desc->l    = (flags >> 21) & 1;
2285 	desc->avl  = (flags >> 20) & 1;
2286 	desc->p    = (flags >> 15) & 1;
2287 	desc->dpl  = (flags >> 13) & 3;
2288 	desc->s    = (flags >> 12) & 1;
2289 	desc->type = (flags >>  8) & 15;
2290 }
2291 
2292 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, const char *smstate,
2293 			   int n)
2294 {
2295 	struct desc_struct desc;
2296 	int offset;
2297 	u16 selector;
2298 
2299 	selector = GET_SMSTATE(u32, smstate, 0x7fa8 + n * 4);
2300 
2301 	if (n < 3)
2302 		offset = 0x7f84 + n * 12;
2303 	else
2304 		offset = 0x7f2c + (n - 3) * 12;
2305 
2306 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2307 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2308 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, offset));
2309 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2310 	return X86EMUL_CONTINUE;
2311 }
2312 
2313 #ifdef CONFIG_X86_64
2314 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, const char *smstate,
2315 			   int n)
2316 {
2317 	struct desc_struct desc;
2318 	int offset;
2319 	u16 selector;
2320 	u32 base3;
2321 
2322 	offset = 0x7e00 + n * 16;
2323 
2324 	selector =                GET_SMSTATE(u16, smstate, offset);
2325 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smstate, offset + 2) << 8);
2326 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2327 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2328 	base3 =                   GET_SMSTATE(u32, smstate, offset + 12);
2329 
2330 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2331 	return X86EMUL_CONTINUE;
2332 }
2333 #endif
2334 
2335 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2336 				    u64 cr0, u64 cr3, u64 cr4)
2337 {
2338 	int bad;
2339 	u64 pcid;
2340 
2341 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
2342 	pcid = 0;
2343 	if (cr4 & X86_CR4_PCIDE) {
2344 		pcid = cr3 & 0xfff;
2345 		cr3 &= ~0xfff;
2346 	}
2347 
2348 	bad = ctxt->ops->set_cr(ctxt, 3, cr3);
2349 	if (bad)
2350 		return X86EMUL_UNHANDLEABLE;
2351 
2352 	/*
2353 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2354 	 * Then enable protected mode.	However, PCID cannot be enabled
2355 	 * if EFER.LMA=0, so set it separately.
2356 	 */
2357 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2358 	if (bad)
2359 		return X86EMUL_UNHANDLEABLE;
2360 
2361 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2362 	if (bad)
2363 		return X86EMUL_UNHANDLEABLE;
2364 
2365 	if (cr4 & X86_CR4_PCIDE) {
2366 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2367 		if (bad)
2368 			return X86EMUL_UNHANDLEABLE;
2369 		if (pcid) {
2370 			bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid);
2371 			if (bad)
2372 				return X86EMUL_UNHANDLEABLE;
2373 		}
2374 
2375 	}
2376 
2377 	return X86EMUL_CONTINUE;
2378 }
2379 
2380 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
2381 			     const char *smstate)
2382 {
2383 	struct desc_struct desc;
2384 	struct desc_ptr dt;
2385 	u16 selector;
2386 	u32 val, cr0, cr3, cr4;
2387 	int i;
2388 
2389 	cr0 =                      GET_SMSTATE(u32, smstate, 0x7ffc);
2390 	cr3 =                      GET_SMSTATE(u32, smstate, 0x7ff8);
2391 	ctxt->eflags =             GET_SMSTATE(u32, smstate, 0x7ff4) | X86_EFLAGS_FIXED;
2392 	ctxt->_eip =               GET_SMSTATE(u32, smstate, 0x7ff0);
2393 
2394 	for (i = 0; i < 8; i++)
2395 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smstate, 0x7fd0 + i * 4);
2396 
2397 	val = GET_SMSTATE(u32, smstate, 0x7fcc);
2398 
2399 	if (ctxt->ops->set_dr(ctxt, 6, val))
2400 		return X86EMUL_UNHANDLEABLE;
2401 
2402 	val = GET_SMSTATE(u32, smstate, 0x7fc8);
2403 
2404 	if (ctxt->ops->set_dr(ctxt, 7, val))
2405 		return X86EMUL_UNHANDLEABLE;
2406 
2407 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc4);
2408 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f64));
2409 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f60));
2410 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f5c));
2411 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2412 
2413 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc0);
2414 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f80));
2415 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f7c));
2416 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f78));
2417 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2418 
2419 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f74);
2420 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f70);
2421 	ctxt->ops->set_gdt(ctxt, &dt);
2422 
2423 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f58);
2424 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f54);
2425 	ctxt->ops->set_idt(ctxt, &dt);
2426 
2427 	for (i = 0; i < 6; i++) {
2428 		int r = rsm_load_seg_32(ctxt, smstate, i);
2429 		if (r != X86EMUL_CONTINUE)
2430 			return r;
2431 	}
2432 
2433 	cr4 = GET_SMSTATE(u32, smstate, 0x7f14);
2434 
2435 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7ef8));
2436 
2437 	return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2438 }
2439 
2440 #ifdef CONFIG_X86_64
2441 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
2442 			     const char *smstate)
2443 {
2444 	struct desc_struct desc;
2445 	struct desc_ptr dt;
2446 	u64 val, cr0, cr3, cr4;
2447 	u32 base3;
2448 	u16 selector;
2449 	int i, r;
2450 
2451 	for (i = 0; i < 16; i++)
2452 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smstate, 0x7ff8 - i * 8);
2453 
2454 	ctxt->_eip   = GET_SMSTATE(u64, smstate, 0x7f78);
2455 	ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7f70) | X86_EFLAGS_FIXED;
2456 
2457 	val = GET_SMSTATE(u64, smstate, 0x7f68);
2458 
2459 	if (ctxt->ops->set_dr(ctxt, 6, val))
2460 		return X86EMUL_UNHANDLEABLE;
2461 
2462 	val = GET_SMSTATE(u64, smstate, 0x7f60);
2463 
2464 	if (ctxt->ops->set_dr(ctxt, 7, val))
2465 		return X86EMUL_UNHANDLEABLE;
2466 
2467 	cr0 =                       GET_SMSTATE(u64, smstate, 0x7f58);
2468 	cr3 =                       GET_SMSTATE(u64, smstate, 0x7f50);
2469 	cr4 =                       GET_SMSTATE(u64, smstate, 0x7f48);
2470 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7f00));
2471 	val =                       GET_SMSTATE(u64, smstate, 0x7ed0);
2472 
2473 	if (ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA))
2474 		return X86EMUL_UNHANDLEABLE;
2475 
2476 	selector =                  GET_SMSTATE(u32, smstate, 0x7e90);
2477 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e92) << 8);
2478 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e94));
2479 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e98));
2480 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e9c);
2481 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2482 
2483 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e84);
2484 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e88);
2485 	ctxt->ops->set_idt(ctxt, &dt);
2486 
2487 	selector =                  GET_SMSTATE(u32, smstate, 0x7e70);
2488 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e72) << 8);
2489 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e74));
2490 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e78));
2491 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e7c);
2492 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2493 
2494 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e64);
2495 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e68);
2496 	ctxt->ops->set_gdt(ctxt, &dt);
2497 
2498 	r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2499 	if (r != X86EMUL_CONTINUE)
2500 		return r;
2501 
2502 	for (i = 0; i < 6; i++) {
2503 		r = rsm_load_seg_64(ctxt, smstate, i);
2504 		if (r != X86EMUL_CONTINUE)
2505 			return r;
2506 	}
2507 
2508 	return X86EMUL_CONTINUE;
2509 }
2510 #endif
2511 
2512 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2513 {
2514 	unsigned long cr0, cr4, efer;
2515 	char buf[512];
2516 	u64 smbase;
2517 	int ret;
2518 
2519 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2520 		return emulate_ud(ctxt);
2521 
2522 	smbase = ctxt->ops->get_smbase(ctxt);
2523 
2524 	ret = ctxt->ops->read_phys(ctxt, smbase + 0xfe00, buf, sizeof(buf));
2525 	if (ret != X86EMUL_CONTINUE)
2526 		return X86EMUL_UNHANDLEABLE;
2527 
2528 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2529 		ctxt->ops->set_nmi_mask(ctxt, false);
2530 
2531 	ctxt->ops->exiting_smm(ctxt);
2532 
2533 	/*
2534 	 * Get back to real mode, to prepare a safe state in which to load
2535 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2536 	 * supports long mode.
2537 	 */
2538 	if (emulator_has_longmode(ctxt)) {
2539 		struct desc_struct cs_desc;
2540 
2541 		/* Zero CR4.PCIDE before CR0.PG.  */
2542 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2543 		if (cr4 & X86_CR4_PCIDE)
2544 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2545 
2546 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2547 		memset(&cs_desc, 0, sizeof(cs_desc));
2548 		cs_desc.type = 0xb;
2549 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2550 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2551 	}
2552 
2553 	/* For the 64-bit case, this will clear EFER.LMA.  */
2554 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2555 	if (cr0 & X86_CR0_PE)
2556 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2557 
2558 	if (emulator_has_longmode(ctxt)) {
2559 		/* Clear CR4.PAE before clearing EFER.LME. */
2560 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2561 		if (cr4 & X86_CR4_PAE)
2562 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2563 
2564 		/* And finally go back to 32-bit mode.  */
2565 		efer = 0;
2566 		ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2567 	}
2568 
2569 	/*
2570 	 * Give leave_smm() a chance to make ISA-specific changes to the vCPU
2571 	 * state (e.g. enter guest mode) before loading state from the SMM
2572 	 * state-save area.
2573 	 */
2574 	if (ctxt->ops->leave_smm(ctxt, buf))
2575 		goto emulate_shutdown;
2576 
2577 #ifdef CONFIG_X86_64
2578 	if (emulator_has_longmode(ctxt))
2579 		ret = rsm_load_state_64(ctxt, buf);
2580 	else
2581 #endif
2582 		ret = rsm_load_state_32(ctxt, buf);
2583 
2584 	if (ret != X86EMUL_CONTINUE)
2585 		goto emulate_shutdown;
2586 
2587 	/*
2588 	 * Note, the ctxt->ops callbacks are responsible for handling side
2589 	 * effects when writing MSRs and CRs, e.g. MMU context resets, CPUID
2590 	 * runtime updates, etc...  If that changes, e.g. this flow is moved
2591 	 * out of the emulator to make it look more like enter_smm(), then
2592 	 * those side effects need to be explicitly handled for both success
2593 	 * and shutdown.
2594 	 */
2595 	return X86EMUL_CONTINUE;
2596 
2597 emulate_shutdown:
2598 	ctxt->ops->triple_fault(ctxt);
2599 	return X86EMUL_CONTINUE;
2600 }
2601 
2602 static void
2603 setup_syscalls_segments(struct x86_emulate_ctxt *ctxt,
2604 			struct desc_struct *cs, struct desc_struct *ss)
2605 {
2606 	cs->l = 0;		/* will be adjusted later */
2607 	set_desc_base(cs, 0);	/* flat segment */
2608 	cs->g = 1;		/* 4kb granularity */
2609 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2610 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2611 	cs->s = 1;
2612 	cs->dpl = 0;		/* will be adjusted later */
2613 	cs->p = 1;
2614 	cs->d = 1;
2615 	cs->avl = 0;
2616 
2617 	set_desc_base(ss, 0);	/* flat segment */
2618 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2619 	ss->g = 1;		/* 4kb granularity */
2620 	ss->s = 1;
2621 	ss->type = 0x03;	/* Read/Write, Accessed */
2622 	ss->d = 1;		/* 32bit stack segment */
2623 	ss->dpl = 0;
2624 	ss->p = 1;
2625 	ss->l = 0;
2626 	ss->avl = 0;
2627 }
2628 
2629 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2630 {
2631 	u32 eax, ebx, ecx, edx;
2632 
2633 	eax = ecx = 0;
2634 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2635 	return is_guest_vendor_intel(ebx, ecx, edx);
2636 }
2637 
2638 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2639 {
2640 	const struct x86_emulate_ops *ops = ctxt->ops;
2641 	u32 eax, ebx, ecx, edx;
2642 
2643 	/*
2644 	 * syscall should always be enabled in longmode - so only become
2645 	 * vendor specific (cpuid) if other modes are active...
2646 	 */
2647 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2648 		return true;
2649 
2650 	eax = 0x00000000;
2651 	ecx = 0x00000000;
2652 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2653 	/*
2654 	 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a
2655 	 * 64bit guest with a 32bit compat-app running will #UD !! While this
2656 	 * behaviour can be fixed (by emulating) into AMD response - CPUs of
2657 	 * AMD can't behave like Intel.
2658 	 */
2659 	if (is_guest_vendor_intel(ebx, ecx, edx))
2660 		return false;
2661 
2662 	if (is_guest_vendor_amd(ebx, ecx, edx) ||
2663 	    is_guest_vendor_hygon(ebx, ecx, edx))
2664 		return true;
2665 
2666 	/*
2667 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2668 	 * stricter rules...
2669 	 */
2670 	return false;
2671 }
2672 
2673 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2674 {
2675 	const struct x86_emulate_ops *ops = ctxt->ops;
2676 	struct desc_struct cs, ss;
2677 	u64 msr_data;
2678 	u16 cs_sel, ss_sel;
2679 	u64 efer = 0;
2680 
2681 	/* syscall is not available in real mode */
2682 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2683 	    ctxt->mode == X86EMUL_MODE_VM86)
2684 		return emulate_ud(ctxt);
2685 
2686 	if (!(em_syscall_is_enabled(ctxt)))
2687 		return emulate_ud(ctxt);
2688 
2689 	ops->get_msr(ctxt, MSR_EFER, &efer);
2690 	if (!(efer & EFER_SCE))
2691 		return emulate_ud(ctxt);
2692 
2693 	setup_syscalls_segments(ctxt, &cs, &ss);
2694 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2695 	msr_data >>= 32;
2696 	cs_sel = (u16)(msr_data & 0xfffc);
2697 	ss_sel = (u16)(msr_data + 8);
2698 
2699 	if (efer & EFER_LMA) {
2700 		cs.d = 0;
2701 		cs.l = 1;
2702 	}
2703 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2704 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2705 
2706 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2707 	if (efer & EFER_LMA) {
2708 #ifdef CONFIG_X86_64
2709 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2710 
2711 		ops->get_msr(ctxt,
2712 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2713 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2714 		ctxt->_eip = msr_data;
2715 
2716 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2717 		ctxt->eflags &= ~msr_data;
2718 		ctxt->eflags |= X86_EFLAGS_FIXED;
2719 #endif
2720 	} else {
2721 		/* legacy mode */
2722 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2723 		ctxt->_eip = (u32)msr_data;
2724 
2725 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2726 	}
2727 
2728 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2729 	return X86EMUL_CONTINUE;
2730 }
2731 
2732 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2733 {
2734 	const struct x86_emulate_ops *ops = ctxt->ops;
2735 	struct desc_struct cs, ss;
2736 	u64 msr_data;
2737 	u16 cs_sel, ss_sel;
2738 	u64 efer = 0;
2739 
2740 	ops->get_msr(ctxt, MSR_EFER, &efer);
2741 	/* inject #GP if in real mode */
2742 	if (ctxt->mode == X86EMUL_MODE_REAL)
2743 		return emulate_gp(ctxt, 0);
2744 
2745 	/*
2746 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2747 	 * mode).
2748 	 */
2749 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2750 	    && !vendor_intel(ctxt))
2751 		return emulate_ud(ctxt);
2752 
2753 	/* sysenter/sysexit have not been tested in 64bit mode. */
2754 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2755 		return X86EMUL_UNHANDLEABLE;
2756 
2757 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2758 	if ((msr_data & 0xfffc) == 0x0)
2759 		return emulate_gp(ctxt, 0);
2760 
2761 	setup_syscalls_segments(ctxt, &cs, &ss);
2762 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2763 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2764 	ss_sel = cs_sel + 8;
2765 	if (efer & EFER_LMA) {
2766 		cs.d = 0;
2767 		cs.l = 1;
2768 	}
2769 
2770 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2771 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2772 
2773 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2774 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2775 
2776 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2777 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2778 							      (u32)msr_data;
2779 	if (efer & EFER_LMA)
2780 		ctxt->mode = X86EMUL_MODE_PROT64;
2781 
2782 	return X86EMUL_CONTINUE;
2783 }
2784 
2785 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2786 {
2787 	const struct x86_emulate_ops *ops = ctxt->ops;
2788 	struct desc_struct cs, ss;
2789 	u64 msr_data, rcx, rdx;
2790 	int usermode;
2791 	u16 cs_sel = 0, ss_sel = 0;
2792 
2793 	/* inject #GP if in real mode or Virtual 8086 mode */
2794 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2795 	    ctxt->mode == X86EMUL_MODE_VM86)
2796 		return emulate_gp(ctxt, 0);
2797 
2798 	setup_syscalls_segments(ctxt, &cs, &ss);
2799 
2800 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2801 		usermode = X86EMUL_MODE_PROT64;
2802 	else
2803 		usermode = X86EMUL_MODE_PROT32;
2804 
2805 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2806 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2807 
2808 	cs.dpl = 3;
2809 	ss.dpl = 3;
2810 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2811 	switch (usermode) {
2812 	case X86EMUL_MODE_PROT32:
2813 		cs_sel = (u16)(msr_data + 16);
2814 		if ((msr_data & 0xfffc) == 0x0)
2815 			return emulate_gp(ctxt, 0);
2816 		ss_sel = (u16)(msr_data + 24);
2817 		rcx = (u32)rcx;
2818 		rdx = (u32)rdx;
2819 		break;
2820 	case X86EMUL_MODE_PROT64:
2821 		cs_sel = (u16)(msr_data + 32);
2822 		if (msr_data == 0x0)
2823 			return emulate_gp(ctxt, 0);
2824 		ss_sel = cs_sel + 8;
2825 		cs.d = 0;
2826 		cs.l = 1;
2827 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2828 		    emul_is_noncanonical_address(rdx, ctxt))
2829 			return emulate_gp(ctxt, 0);
2830 		break;
2831 	}
2832 	cs_sel |= SEGMENT_RPL_MASK;
2833 	ss_sel |= SEGMENT_RPL_MASK;
2834 
2835 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2836 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2837 
2838 	ctxt->_eip = rdx;
2839 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2840 
2841 	return X86EMUL_CONTINUE;
2842 }
2843 
2844 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2845 {
2846 	int iopl;
2847 	if (ctxt->mode == X86EMUL_MODE_REAL)
2848 		return false;
2849 	if (ctxt->mode == X86EMUL_MODE_VM86)
2850 		return true;
2851 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2852 	return ctxt->ops->cpl(ctxt) > iopl;
2853 }
2854 
2855 #define VMWARE_PORT_VMPORT	(0x5658)
2856 #define VMWARE_PORT_VMRPC	(0x5659)
2857 
2858 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2859 					    u16 port, u16 len)
2860 {
2861 	const struct x86_emulate_ops *ops = ctxt->ops;
2862 	struct desc_struct tr_seg;
2863 	u32 base3;
2864 	int r;
2865 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2866 	unsigned mask = (1 << len) - 1;
2867 	unsigned long base;
2868 
2869 	/*
2870 	 * VMware allows access to these ports even if denied
2871 	 * by TSS I/O permission bitmap. Mimic behavior.
2872 	 */
2873 	if (enable_vmware_backdoor &&
2874 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2875 		return true;
2876 
2877 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2878 	if (!tr_seg.p)
2879 		return false;
2880 	if (desc_limit_scaled(&tr_seg) < 103)
2881 		return false;
2882 	base = get_desc_base(&tr_seg);
2883 #ifdef CONFIG_X86_64
2884 	base |= ((u64)base3) << 32;
2885 #endif
2886 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2887 	if (r != X86EMUL_CONTINUE)
2888 		return false;
2889 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2890 		return false;
2891 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2892 	if (r != X86EMUL_CONTINUE)
2893 		return false;
2894 	if ((perm >> bit_idx) & mask)
2895 		return false;
2896 	return true;
2897 }
2898 
2899 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2900 				 u16 port, u16 len)
2901 {
2902 	if (ctxt->perm_ok)
2903 		return true;
2904 
2905 	if (emulator_bad_iopl(ctxt))
2906 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2907 			return false;
2908 
2909 	ctxt->perm_ok = true;
2910 
2911 	return true;
2912 }
2913 
2914 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2915 {
2916 	/*
2917 	 * Intel CPUs mask the counter and pointers in quite strange
2918 	 * manner when ECX is zero due to REP-string optimizations.
2919 	 */
2920 #ifdef CONFIG_X86_64
2921 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2922 		return;
2923 
2924 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2925 
2926 	switch (ctxt->b) {
2927 	case 0xa4:	/* movsb */
2928 	case 0xa5:	/* movsd/w */
2929 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2930 		fallthrough;
2931 	case 0xaa:	/* stosb */
2932 	case 0xab:	/* stosd/w */
2933 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2934 	}
2935 #endif
2936 }
2937 
2938 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2939 				struct tss_segment_16 *tss)
2940 {
2941 	tss->ip = ctxt->_eip;
2942 	tss->flag = ctxt->eflags;
2943 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2944 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2945 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2946 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2947 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2948 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2949 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2950 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2951 
2952 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2953 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2954 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2955 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2956 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2957 }
2958 
2959 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2960 				 struct tss_segment_16 *tss)
2961 {
2962 	int ret;
2963 	u8 cpl;
2964 
2965 	ctxt->_eip = tss->ip;
2966 	ctxt->eflags = tss->flag | 2;
2967 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2968 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2969 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2970 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2971 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2972 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2973 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2974 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2975 
2976 	/*
2977 	 * SDM says that segment selectors are loaded before segment
2978 	 * descriptors
2979 	 */
2980 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
2981 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2982 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2983 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2984 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2985 
2986 	cpl = tss->cs & 3;
2987 
2988 	/*
2989 	 * Now load segment descriptors. If fault happens at this stage
2990 	 * it is handled in a context of new task
2991 	 */
2992 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
2993 					X86_TRANSFER_TASK_SWITCH, NULL);
2994 	if (ret != X86EMUL_CONTINUE)
2995 		return ret;
2996 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
2997 					X86_TRANSFER_TASK_SWITCH, NULL);
2998 	if (ret != X86EMUL_CONTINUE)
2999 		return ret;
3000 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3001 					X86_TRANSFER_TASK_SWITCH, NULL);
3002 	if (ret != X86EMUL_CONTINUE)
3003 		return ret;
3004 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3005 					X86_TRANSFER_TASK_SWITCH, NULL);
3006 	if (ret != X86EMUL_CONTINUE)
3007 		return ret;
3008 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3009 					X86_TRANSFER_TASK_SWITCH, NULL);
3010 	if (ret != X86EMUL_CONTINUE)
3011 		return ret;
3012 
3013 	return X86EMUL_CONTINUE;
3014 }
3015 
3016 static int task_switch_16(struct x86_emulate_ctxt *ctxt,
3017 			  u16 tss_selector, u16 old_tss_sel,
3018 			  ulong old_tss_base, struct desc_struct *new_desc)
3019 {
3020 	struct tss_segment_16 tss_seg;
3021 	int ret;
3022 	u32 new_tss_base = get_desc_base(new_desc);
3023 
3024 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3025 	if (ret != X86EMUL_CONTINUE)
3026 		return ret;
3027 
3028 	save_state_to_tss16(ctxt, &tss_seg);
3029 
3030 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3031 	if (ret != X86EMUL_CONTINUE)
3032 		return ret;
3033 
3034 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3035 	if (ret != X86EMUL_CONTINUE)
3036 		return ret;
3037 
3038 	if (old_tss_sel != 0xffff) {
3039 		tss_seg.prev_task_link = old_tss_sel;
3040 
3041 		ret = linear_write_system(ctxt, new_tss_base,
3042 					  &tss_seg.prev_task_link,
3043 					  sizeof(tss_seg.prev_task_link));
3044 		if (ret != X86EMUL_CONTINUE)
3045 			return ret;
3046 	}
3047 
3048 	return load_state_from_tss16(ctxt, &tss_seg);
3049 }
3050 
3051 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3052 				struct tss_segment_32 *tss)
3053 {
3054 	/* CR3 and ldt selector are not saved intentionally */
3055 	tss->eip = ctxt->_eip;
3056 	tss->eflags = ctxt->eflags;
3057 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3058 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3059 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3060 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3061 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3062 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3063 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3064 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3065 
3066 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3067 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3068 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3069 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3070 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3071 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3072 }
3073 
3074 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3075 				 struct tss_segment_32 *tss)
3076 {
3077 	int ret;
3078 	u8 cpl;
3079 
3080 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3081 		return emulate_gp(ctxt, 0);
3082 	ctxt->_eip = tss->eip;
3083 	ctxt->eflags = tss->eflags | 2;
3084 
3085 	/* General purpose registers */
3086 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3087 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3088 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3089 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3090 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3091 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3092 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3093 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3094 
3095 	/*
3096 	 * SDM says that segment selectors are loaded before segment
3097 	 * descriptors.  This is important because CPL checks will
3098 	 * use CS.RPL.
3099 	 */
3100 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3101 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3102 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3103 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3104 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3105 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3106 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3107 
3108 	/*
3109 	 * If we're switching between Protected Mode and VM86, we need to make
3110 	 * sure to update the mode before loading the segment descriptors so
3111 	 * that the selectors are interpreted correctly.
3112 	 */
3113 	if (ctxt->eflags & X86_EFLAGS_VM) {
3114 		ctxt->mode = X86EMUL_MODE_VM86;
3115 		cpl = 3;
3116 	} else {
3117 		ctxt->mode = X86EMUL_MODE_PROT32;
3118 		cpl = tss->cs & 3;
3119 	}
3120 
3121 	/*
3122 	 * Now load segment descriptors. If fault happens at this stage
3123 	 * it is handled in a context of new task
3124 	 */
3125 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3126 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3127 	if (ret != X86EMUL_CONTINUE)
3128 		return ret;
3129 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3130 					X86_TRANSFER_TASK_SWITCH, NULL);
3131 	if (ret != X86EMUL_CONTINUE)
3132 		return ret;
3133 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3134 					X86_TRANSFER_TASK_SWITCH, NULL);
3135 	if (ret != X86EMUL_CONTINUE)
3136 		return ret;
3137 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3138 					X86_TRANSFER_TASK_SWITCH, NULL);
3139 	if (ret != X86EMUL_CONTINUE)
3140 		return ret;
3141 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3142 					X86_TRANSFER_TASK_SWITCH, NULL);
3143 	if (ret != X86EMUL_CONTINUE)
3144 		return ret;
3145 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3146 					X86_TRANSFER_TASK_SWITCH, NULL);
3147 	if (ret != X86EMUL_CONTINUE)
3148 		return ret;
3149 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3150 					X86_TRANSFER_TASK_SWITCH, NULL);
3151 
3152 	return ret;
3153 }
3154 
3155 static int task_switch_32(struct x86_emulate_ctxt *ctxt,
3156 			  u16 tss_selector, u16 old_tss_sel,
3157 			  ulong old_tss_base, struct desc_struct *new_desc)
3158 {
3159 	struct tss_segment_32 tss_seg;
3160 	int ret;
3161 	u32 new_tss_base = get_desc_base(new_desc);
3162 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3163 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3164 
3165 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3166 	if (ret != X86EMUL_CONTINUE)
3167 		return ret;
3168 
3169 	save_state_to_tss32(ctxt, &tss_seg);
3170 
3171 	/* Only GP registers and segment selectors are saved */
3172 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3173 				  ldt_sel_offset - eip_offset);
3174 	if (ret != X86EMUL_CONTINUE)
3175 		return ret;
3176 
3177 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3178 	if (ret != X86EMUL_CONTINUE)
3179 		return ret;
3180 
3181 	if (old_tss_sel != 0xffff) {
3182 		tss_seg.prev_task_link = old_tss_sel;
3183 
3184 		ret = linear_write_system(ctxt, new_tss_base,
3185 					  &tss_seg.prev_task_link,
3186 					  sizeof(tss_seg.prev_task_link));
3187 		if (ret != X86EMUL_CONTINUE)
3188 			return ret;
3189 	}
3190 
3191 	return load_state_from_tss32(ctxt, &tss_seg);
3192 }
3193 
3194 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3195 				   u16 tss_selector, int idt_index, int reason,
3196 				   bool has_error_code, u32 error_code)
3197 {
3198 	const struct x86_emulate_ops *ops = ctxt->ops;
3199 	struct desc_struct curr_tss_desc, next_tss_desc;
3200 	int ret;
3201 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3202 	ulong old_tss_base =
3203 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3204 	u32 desc_limit;
3205 	ulong desc_addr, dr7;
3206 
3207 	/* FIXME: old_tss_base == ~0 ? */
3208 
3209 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3210 	if (ret != X86EMUL_CONTINUE)
3211 		return ret;
3212 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3213 	if (ret != X86EMUL_CONTINUE)
3214 		return ret;
3215 
3216 	/* FIXME: check that next_tss_desc is tss */
3217 
3218 	/*
3219 	 * Check privileges. The three cases are task switch caused by...
3220 	 *
3221 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3222 	 * 2. Exception/IRQ/iret: No check is performed
3223 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3224 	 *    hardware checks it before exiting.
3225 	 */
3226 	if (reason == TASK_SWITCH_GATE) {
3227 		if (idt_index != -1) {
3228 			/* Software interrupts */
3229 			struct desc_struct task_gate_desc;
3230 			int dpl;
3231 
3232 			ret = read_interrupt_descriptor(ctxt, idt_index,
3233 							&task_gate_desc);
3234 			if (ret != X86EMUL_CONTINUE)
3235 				return ret;
3236 
3237 			dpl = task_gate_desc.dpl;
3238 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3239 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3240 		}
3241 	}
3242 
3243 	desc_limit = desc_limit_scaled(&next_tss_desc);
3244 	if (!next_tss_desc.p ||
3245 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3246 	     desc_limit < 0x2b)) {
3247 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3248 	}
3249 
3250 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3251 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3252 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3253 	}
3254 
3255 	if (reason == TASK_SWITCH_IRET)
3256 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3257 
3258 	/* set back link to prev task only if NT bit is set in eflags
3259 	   note that old_tss_sel is not used after this point */
3260 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3261 		old_tss_sel = 0xffff;
3262 
3263 	if (next_tss_desc.type & 8)
3264 		ret = task_switch_32(ctxt, tss_selector, old_tss_sel,
3265 				     old_tss_base, &next_tss_desc);
3266 	else
3267 		ret = task_switch_16(ctxt, tss_selector, old_tss_sel,
3268 				     old_tss_base, &next_tss_desc);
3269 	if (ret != X86EMUL_CONTINUE)
3270 		return ret;
3271 
3272 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3273 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3274 
3275 	if (reason != TASK_SWITCH_IRET) {
3276 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3277 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3278 	}
3279 
3280 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3281 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3282 
3283 	if (has_error_code) {
3284 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3285 		ctxt->lock_prefix = 0;
3286 		ctxt->src.val = (unsigned long) error_code;
3287 		ret = em_push(ctxt);
3288 	}
3289 
3290 	ops->get_dr(ctxt, 7, &dr7);
3291 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3292 
3293 	return ret;
3294 }
3295 
3296 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3297 			 u16 tss_selector, int idt_index, int reason,
3298 			 bool has_error_code, u32 error_code)
3299 {
3300 	int rc;
3301 
3302 	invalidate_registers(ctxt);
3303 	ctxt->_eip = ctxt->eip;
3304 	ctxt->dst.type = OP_NONE;
3305 
3306 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3307 				     has_error_code, error_code);
3308 
3309 	if (rc == X86EMUL_CONTINUE) {
3310 		ctxt->eip = ctxt->_eip;
3311 		writeback_registers(ctxt);
3312 	}
3313 
3314 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3315 }
3316 
3317 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3318 		struct operand *op)
3319 {
3320 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3321 
3322 	register_address_increment(ctxt, reg, df * op->bytes);
3323 	op->addr.mem.ea = register_address(ctxt, reg);
3324 }
3325 
3326 static int em_das(struct x86_emulate_ctxt *ctxt)
3327 {
3328 	u8 al, old_al;
3329 	bool af, cf, old_cf;
3330 
3331 	cf = ctxt->eflags & X86_EFLAGS_CF;
3332 	al = ctxt->dst.val;
3333 
3334 	old_al = al;
3335 	old_cf = cf;
3336 	cf = false;
3337 	af = ctxt->eflags & X86_EFLAGS_AF;
3338 	if ((al & 0x0f) > 9 || af) {
3339 		al -= 6;
3340 		cf = old_cf | (al >= 250);
3341 		af = true;
3342 	} else {
3343 		af = false;
3344 	}
3345 	if (old_al > 0x99 || old_cf) {
3346 		al -= 0x60;
3347 		cf = true;
3348 	}
3349 
3350 	ctxt->dst.val = al;
3351 	/* Set PF, ZF, SF */
3352 	ctxt->src.type = OP_IMM;
3353 	ctxt->src.val = 0;
3354 	ctxt->src.bytes = 1;
3355 	fastop(ctxt, em_or);
3356 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3357 	if (cf)
3358 		ctxt->eflags |= X86_EFLAGS_CF;
3359 	if (af)
3360 		ctxt->eflags |= X86_EFLAGS_AF;
3361 	return X86EMUL_CONTINUE;
3362 }
3363 
3364 static int em_aam(struct x86_emulate_ctxt *ctxt)
3365 {
3366 	u8 al, ah;
3367 
3368 	if (ctxt->src.val == 0)
3369 		return emulate_de(ctxt);
3370 
3371 	al = ctxt->dst.val & 0xff;
3372 	ah = al / ctxt->src.val;
3373 	al %= ctxt->src.val;
3374 
3375 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3376 
3377 	/* Set PF, ZF, SF */
3378 	ctxt->src.type = OP_IMM;
3379 	ctxt->src.val = 0;
3380 	ctxt->src.bytes = 1;
3381 	fastop(ctxt, em_or);
3382 
3383 	return X86EMUL_CONTINUE;
3384 }
3385 
3386 static int em_aad(struct x86_emulate_ctxt *ctxt)
3387 {
3388 	u8 al = ctxt->dst.val & 0xff;
3389 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3390 
3391 	al = (al + (ah * ctxt->src.val)) & 0xff;
3392 
3393 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3394 
3395 	/* Set PF, ZF, SF */
3396 	ctxt->src.type = OP_IMM;
3397 	ctxt->src.val = 0;
3398 	ctxt->src.bytes = 1;
3399 	fastop(ctxt, em_or);
3400 
3401 	return X86EMUL_CONTINUE;
3402 }
3403 
3404 static int em_call(struct x86_emulate_ctxt *ctxt)
3405 {
3406 	int rc;
3407 	long rel = ctxt->src.val;
3408 
3409 	ctxt->src.val = (unsigned long)ctxt->_eip;
3410 	rc = jmp_rel(ctxt, rel);
3411 	if (rc != X86EMUL_CONTINUE)
3412 		return rc;
3413 	return em_push(ctxt);
3414 }
3415 
3416 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3417 {
3418 	u16 sel, old_cs;
3419 	ulong old_eip;
3420 	int rc;
3421 	struct desc_struct old_desc, new_desc;
3422 	const struct x86_emulate_ops *ops = ctxt->ops;
3423 	int cpl = ctxt->ops->cpl(ctxt);
3424 	enum x86emul_mode prev_mode = ctxt->mode;
3425 
3426 	old_eip = ctxt->_eip;
3427 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3428 
3429 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3430 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3431 				       X86_TRANSFER_CALL_JMP, &new_desc);
3432 	if (rc != X86EMUL_CONTINUE)
3433 		return rc;
3434 
3435 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3436 	if (rc != X86EMUL_CONTINUE)
3437 		goto fail;
3438 
3439 	ctxt->src.val = old_cs;
3440 	rc = em_push(ctxt);
3441 	if (rc != X86EMUL_CONTINUE)
3442 		goto fail;
3443 
3444 	ctxt->src.val = old_eip;
3445 	rc = em_push(ctxt);
3446 	/* If we failed, we tainted the memory, but the very least we should
3447 	   restore cs */
3448 	if (rc != X86EMUL_CONTINUE) {
3449 		pr_warn_once("faulting far call emulation tainted memory\n");
3450 		goto fail;
3451 	}
3452 	return rc;
3453 fail:
3454 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3455 	ctxt->mode = prev_mode;
3456 	return rc;
3457 
3458 }
3459 
3460 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3461 {
3462 	int rc;
3463 	unsigned long eip;
3464 
3465 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3466 	if (rc != X86EMUL_CONTINUE)
3467 		return rc;
3468 	rc = assign_eip_near(ctxt, eip);
3469 	if (rc != X86EMUL_CONTINUE)
3470 		return rc;
3471 	rsp_increment(ctxt, ctxt->src.val);
3472 	return X86EMUL_CONTINUE;
3473 }
3474 
3475 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3476 {
3477 	/* Write back the register source. */
3478 	ctxt->src.val = ctxt->dst.val;
3479 	write_register_operand(&ctxt->src);
3480 
3481 	/* Write back the memory destination with implicit LOCK prefix. */
3482 	ctxt->dst.val = ctxt->src.orig_val;
3483 	ctxt->lock_prefix = 1;
3484 	return X86EMUL_CONTINUE;
3485 }
3486 
3487 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3488 {
3489 	ctxt->dst.val = ctxt->src2.val;
3490 	return fastop(ctxt, em_imul);
3491 }
3492 
3493 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3494 {
3495 	ctxt->dst.type = OP_REG;
3496 	ctxt->dst.bytes = ctxt->src.bytes;
3497 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3498 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3499 
3500 	return X86EMUL_CONTINUE;
3501 }
3502 
3503 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3504 {
3505 	u64 tsc_aux = 0;
3506 
3507 	if (ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux))
3508 		return emulate_ud(ctxt);
3509 	ctxt->dst.val = tsc_aux;
3510 	return X86EMUL_CONTINUE;
3511 }
3512 
3513 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3514 {
3515 	u64 tsc = 0;
3516 
3517 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3518 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3519 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3520 	return X86EMUL_CONTINUE;
3521 }
3522 
3523 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3524 {
3525 	u64 pmc;
3526 
3527 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3528 		return emulate_gp(ctxt, 0);
3529 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3530 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3531 	return X86EMUL_CONTINUE;
3532 }
3533 
3534 static int em_mov(struct x86_emulate_ctxt *ctxt)
3535 {
3536 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3537 	return X86EMUL_CONTINUE;
3538 }
3539 
3540 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3541 {
3542 	u16 tmp;
3543 
3544 	if (!ctxt->ops->guest_has_movbe(ctxt))
3545 		return emulate_ud(ctxt);
3546 
3547 	switch (ctxt->op_bytes) {
3548 	case 2:
3549 		/*
3550 		 * From MOVBE definition: "...When the operand size is 16 bits,
3551 		 * the upper word of the destination register remains unchanged
3552 		 * ..."
3553 		 *
3554 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3555 		 * rules so we have to do the operation almost per hand.
3556 		 */
3557 		tmp = (u16)ctxt->src.val;
3558 		ctxt->dst.val &= ~0xffffUL;
3559 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3560 		break;
3561 	case 4:
3562 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3563 		break;
3564 	case 8:
3565 		ctxt->dst.val = swab64(ctxt->src.val);
3566 		break;
3567 	default:
3568 		BUG();
3569 	}
3570 	return X86EMUL_CONTINUE;
3571 }
3572 
3573 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3574 {
3575 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3576 		return emulate_gp(ctxt, 0);
3577 
3578 	/* Disable writeback. */
3579 	ctxt->dst.type = OP_NONE;
3580 	return X86EMUL_CONTINUE;
3581 }
3582 
3583 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3584 {
3585 	unsigned long val;
3586 
3587 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3588 		val = ctxt->src.val & ~0ULL;
3589 	else
3590 		val = ctxt->src.val & ~0U;
3591 
3592 	/* #UD condition is already handled. */
3593 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3594 		return emulate_gp(ctxt, 0);
3595 
3596 	/* Disable writeback. */
3597 	ctxt->dst.type = OP_NONE;
3598 	return X86EMUL_CONTINUE;
3599 }
3600 
3601 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3602 {
3603 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3604 	u64 msr_data;
3605 	int r;
3606 
3607 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3608 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3609 	r = ctxt->ops->set_msr(ctxt, msr_index, msr_data);
3610 
3611 	if (r == X86EMUL_IO_NEEDED)
3612 		return r;
3613 
3614 	if (r > 0)
3615 		return emulate_gp(ctxt, 0);
3616 
3617 	return r < 0 ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
3618 }
3619 
3620 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3621 {
3622 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3623 	u64 msr_data;
3624 	int r;
3625 
3626 	r = ctxt->ops->get_msr(ctxt, msr_index, &msr_data);
3627 
3628 	if (r == X86EMUL_IO_NEEDED)
3629 		return r;
3630 
3631 	if (r)
3632 		return emulate_gp(ctxt, 0);
3633 
3634 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3635 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3636 	return X86EMUL_CONTINUE;
3637 }
3638 
3639 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3640 {
3641 	if (segment > VCPU_SREG_GS &&
3642 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3643 	    ctxt->ops->cpl(ctxt) > 0)
3644 		return emulate_gp(ctxt, 0);
3645 
3646 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3647 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3648 		ctxt->dst.bytes = 2;
3649 	return X86EMUL_CONTINUE;
3650 }
3651 
3652 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3653 {
3654 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3655 		return emulate_ud(ctxt);
3656 
3657 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3658 }
3659 
3660 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3661 {
3662 	u16 sel = ctxt->src.val;
3663 
3664 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3665 		return emulate_ud(ctxt);
3666 
3667 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3668 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3669 
3670 	/* Disable writeback. */
3671 	ctxt->dst.type = OP_NONE;
3672 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3673 }
3674 
3675 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3676 {
3677 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3678 }
3679 
3680 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3681 {
3682 	u16 sel = ctxt->src.val;
3683 
3684 	/* Disable writeback. */
3685 	ctxt->dst.type = OP_NONE;
3686 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3687 }
3688 
3689 static int em_str(struct x86_emulate_ctxt *ctxt)
3690 {
3691 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3692 }
3693 
3694 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3695 {
3696 	u16 sel = ctxt->src.val;
3697 
3698 	/* Disable writeback. */
3699 	ctxt->dst.type = OP_NONE;
3700 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3701 }
3702 
3703 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3704 {
3705 	int rc;
3706 	ulong linear;
3707 
3708 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3709 	if (rc == X86EMUL_CONTINUE)
3710 		ctxt->ops->invlpg(ctxt, linear);
3711 	/* Disable writeback. */
3712 	ctxt->dst.type = OP_NONE;
3713 	return X86EMUL_CONTINUE;
3714 }
3715 
3716 static int em_clts(struct x86_emulate_ctxt *ctxt)
3717 {
3718 	ulong cr0;
3719 
3720 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3721 	cr0 &= ~X86_CR0_TS;
3722 	ctxt->ops->set_cr(ctxt, 0, cr0);
3723 	return X86EMUL_CONTINUE;
3724 }
3725 
3726 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3727 {
3728 	int rc = ctxt->ops->fix_hypercall(ctxt);
3729 
3730 	if (rc != X86EMUL_CONTINUE)
3731 		return rc;
3732 
3733 	/* Let the processor re-execute the fixed hypercall */
3734 	ctxt->_eip = ctxt->eip;
3735 	/* Disable writeback. */
3736 	ctxt->dst.type = OP_NONE;
3737 	return X86EMUL_CONTINUE;
3738 }
3739 
3740 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3741 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3742 					      struct desc_ptr *ptr))
3743 {
3744 	struct desc_ptr desc_ptr;
3745 
3746 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3747 	    ctxt->ops->cpl(ctxt) > 0)
3748 		return emulate_gp(ctxt, 0);
3749 
3750 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3751 		ctxt->op_bytes = 8;
3752 	get(ctxt, &desc_ptr);
3753 	if (ctxt->op_bytes == 2) {
3754 		ctxt->op_bytes = 4;
3755 		desc_ptr.address &= 0x00ffffff;
3756 	}
3757 	/* Disable writeback. */
3758 	ctxt->dst.type = OP_NONE;
3759 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3760 				   &desc_ptr, 2 + ctxt->op_bytes);
3761 }
3762 
3763 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3764 {
3765 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3766 }
3767 
3768 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3769 {
3770 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3771 }
3772 
3773 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3774 {
3775 	struct desc_ptr desc_ptr;
3776 	int rc;
3777 
3778 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3779 		ctxt->op_bytes = 8;
3780 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3781 			     &desc_ptr.size, &desc_ptr.address,
3782 			     ctxt->op_bytes);
3783 	if (rc != X86EMUL_CONTINUE)
3784 		return rc;
3785 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3786 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3787 		return emulate_gp(ctxt, 0);
3788 	if (lgdt)
3789 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3790 	else
3791 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3792 	/* Disable writeback. */
3793 	ctxt->dst.type = OP_NONE;
3794 	return X86EMUL_CONTINUE;
3795 }
3796 
3797 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3798 {
3799 	return em_lgdt_lidt(ctxt, true);
3800 }
3801 
3802 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3803 {
3804 	return em_lgdt_lidt(ctxt, false);
3805 }
3806 
3807 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3808 {
3809 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3810 	    ctxt->ops->cpl(ctxt) > 0)
3811 		return emulate_gp(ctxt, 0);
3812 
3813 	if (ctxt->dst.type == OP_MEM)
3814 		ctxt->dst.bytes = 2;
3815 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3816 	return X86EMUL_CONTINUE;
3817 }
3818 
3819 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3820 {
3821 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3822 			  | (ctxt->src.val & 0x0f));
3823 	ctxt->dst.type = OP_NONE;
3824 	return X86EMUL_CONTINUE;
3825 }
3826 
3827 static int em_loop(struct x86_emulate_ctxt *ctxt)
3828 {
3829 	int rc = X86EMUL_CONTINUE;
3830 
3831 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3832 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3833 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3834 		rc = jmp_rel(ctxt, ctxt->src.val);
3835 
3836 	return rc;
3837 }
3838 
3839 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3840 {
3841 	int rc = X86EMUL_CONTINUE;
3842 
3843 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3844 		rc = jmp_rel(ctxt, ctxt->src.val);
3845 
3846 	return rc;
3847 }
3848 
3849 static int em_in(struct x86_emulate_ctxt *ctxt)
3850 {
3851 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3852 			     &ctxt->dst.val))
3853 		return X86EMUL_IO_NEEDED;
3854 
3855 	return X86EMUL_CONTINUE;
3856 }
3857 
3858 static int em_out(struct x86_emulate_ctxt *ctxt)
3859 {
3860 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3861 				    &ctxt->src.val, 1);
3862 	/* Disable writeback. */
3863 	ctxt->dst.type = OP_NONE;
3864 	return X86EMUL_CONTINUE;
3865 }
3866 
3867 static int em_cli(struct x86_emulate_ctxt *ctxt)
3868 {
3869 	if (emulator_bad_iopl(ctxt))
3870 		return emulate_gp(ctxt, 0);
3871 
3872 	ctxt->eflags &= ~X86_EFLAGS_IF;
3873 	return X86EMUL_CONTINUE;
3874 }
3875 
3876 static int em_sti(struct x86_emulate_ctxt *ctxt)
3877 {
3878 	if (emulator_bad_iopl(ctxt))
3879 		return emulate_gp(ctxt, 0);
3880 
3881 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3882 	ctxt->eflags |= X86_EFLAGS_IF;
3883 	return X86EMUL_CONTINUE;
3884 }
3885 
3886 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3887 {
3888 	u32 eax, ebx, ecx, edx;
3889 	u64 msr = 0;
3890 
3891 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3892 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3893 	    ctxt->ops->cpl(ctxt)) {
3894 		return emulate_gp(ctxt, 0);
3895 	}
3896 
3897 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3898 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3899 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3900 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3901 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3902 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3903 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3904 	return X86EMUL_CONTINUE;
3905 }
3906 
3907 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3908 {
3909 	u32 flags;
3910 
3911 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3912 		X86_EFLAGS_SF;
3913 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3914 
3915 	ctxt->eflags &= ~0xffUL;
3916 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3917 	return X86EMUL_CONTINUE;
3918 }
3919 
3920 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3921 {
3922 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3923 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3924 	return X86EMUL_CONTINUE;
3925 }
3926 
3927 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3928 {
3929 	switch (ctxt->op_bytes) {
3930 #ifdef CONFIG_X86_64
3931 	case 8:
3932 		asm("bswap %0" : "+r"(ctxt->dst.val));
3933 		break;
3934 #endif
3935 	default:
3936 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3937 		break;
3938 	}
3939 	return X86EMUL_CONTINUE;
3940 }
3941 
3942 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3943 {
3944 	/* emulating clflush regardless of cpuid */
3945 	return X86EMUL_CONTINUE;
3946 }
3947 
3948 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
3949 {
3950 	/* emulating clflushopt regardless of cpuid */
3951 	return X86EMUL_CONTINUE;
3952 }
3953 
3954 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3955 {
3956 	ctxt->dst.val = (s32) ctxt->src.val;
3957 	return X86EMUL_CONTINUE;
3958 }
3959 
3960 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3961 {
3962 	if (!ctxt->ops->guest_has_fxsr(ctxt))
3963 		return emulate_ud(ctxt);
3964 
3965 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3966 		return emulate_nm(ctxt);
3967 
3968 	/*
3969 	 * Don't emulate a case that should never be hit, instead of working
3970 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3971 	 */
3972 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3973 		return X86EMUL_UNHANDLEABLE;
3974 
3975 	return X86EMUL_CONTINUE;
3976 }
3977 
3978 /*
3979  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
3980  * and restore MXCSR.
3981  */
3982 static size_t __fxstate_size(int nregs)
3983 {
3984 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
3985 }
3986 
3987 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
3988 {
3989 	bool cr4_osfxsr;
3990 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3991 		return __fxstate_size(16);
3992 
3993 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
3994 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
3995 }
3996 
3997 /*
3998  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
3999  *  1) 16 bit mode
4000  *  2) 32 bit mode
4001  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
4002  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
4003  *       save and restore
4004  *  3) 64-bit mode with REX.W prefix
4005  *     - like (2), but XMM 8-15 are being saved and restored
4006  *  4) 64-bit mode without REX.W prefix
4007  *     - like (3), but FIP and FDP are 64 bit
4008  *
4009  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
4010  * desired result.  (4) is not emulated.
4011  *
4012  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
4013  * and FPU DS) should match.
4014  */
4015 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
4016 {
4017 	struct fxregs_state fx_state;
4018 	int rc;
4019 
4020 	rc = check_fxsr(ctxt);
4021 	if (rc != X86EMUL_CONTINUE)
4022 		return rc;
4023 
4024 	kvm_fpu_get();
4025 
4026 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4027 
4028 	kvm_fpu_put();
4029 
4030 	if (rc != X86EMUL_CONTINUE)
4031 		return rc;
4032 
4033 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4034 		                   fxstate_size(ctxt));
4035 }
4036 
4037 /*
4038  * FXRSTOR might restore XMM registers not provided by the guest. Fill
4039  * in the host registers (via FXSAVE) instead, so they won't be modified.
4040  * (preemption has to stay disabled until FXRSTOR).
4041  *
4042  * Use noinline to keep the stack for other functions called by callers small.
4043  */
4044 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
4045 				 const size_t used_size)
4046 {
4047 	struct fxregs_state fx_tmp;
4048 	int rc;
4049 
4050 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
4051 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
4052 	       __fxstate_size(16) - used_size);
4053 
4054 	return rc;
4055 }
4056 
4057 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4058 {
4059 	struct fxregs_state fx_state;
4060 	int rc;
4061 	size_t size;
4062 
4063 	rc = check_fxsr(ctxt);
4064 	if (rc != X86EMUL_CONTINUE)
4065 		return rc;
4066 
4067 	size = fxstate_size(ctxt);
4068 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4069 	if (rc != X86EMUL_CONTINUE)
4070 		return rc;
4071 
4072 	kvm_fpu_get();
4073 
4074 	if (size < __fxstate_size(16)) {
4075 		rc = fxregs_fixup(&fx_state, size);
4076 		if (rc != X86EMUL_CONTINUE)
4077 			goto out;
4078 	}
4079 
4080 	if (fx_state.mxcsr >> 16) {
4081 		rc = emulate_gp(ctxt, 0);
4082 		goto out;
4083 	}
4084 
4085 	if (rc == X86EMUL_CONTINUE)
4086 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4087 
4088 out:
4089 	kvm_fpu_put();
4090 
4091 	return rc;
4092 }
4093 
4094 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
4095 {
4096 	u32 eax, ecx, edx;
4097 
4098 	eax = reg_read(ctxt, VCPU_REGS_RAX);
4099 	edx = reg_read(ctxt, VCPU_REGS_RDX);
4100 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
4101 
4102 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
4103 		return emulate_gp(ctxt, 0);
4104 
4105 	return X86EMUL_CONTINUE;
4106 }
4107 
4108 static bool valid_cr(int nr)
4109 {
4110 	switch (nr) {
4111 	case 0:
4112 	case 2 ... 4:
4113 	case 8:
4114 		return true;
4115 	default:
4116 		return false;
4117 	}
4118 }
4119 
4120 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
4121 {
4122 	if (!valid_cr(ctxt->modrm_reg))
4123 		return emulate_ud(ctxt);
4124 
4125 	return X86EMUL_CONTINUE;
4126 }
4127 
4128 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4129 {
4130 	unsigned long dr7;
4131 
4132 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4133 
4134 	/* Check if DR7.Global_Enable is set */
4135 	return dr7 & (1 << 13);
4136 }
4137 
4138 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4139 {
4140 	int dr = ctxt->modrm_reg;
4141 	u64 cr4;
4142 
4143 	if (dr > 7)
4144 		return emulate_ud(ctxt);
4145 
4146 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4147 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4148 		return emulate_ud(ctxt);
4149 
4150 	if (check_dr7_gd(ctxt)) {
4151 		ulong dr6;
4152 
4153 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4154 		dr6 &= ~DR_TRAP_BITS;
4155 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
4156 		ctxt->ops->set_dr(ctxt, 6, dr6);
4157 		return emulate_db(ctxt);
4158 	}
4159 
4160 	return X86EMUL_CONTINUE;
4161 }
4162 
4163 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4164 {
4165 	u64 new_val = ctxt->src.val64;
4166 	int dr = ctxt->modrm_reg;
4167 
4168 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4169 		return emulate_gp(ctxt, 0);
4170 
4171 	return check_dr_read(ctxt);
4172 }
4173 
4174 static int check_svme(struct x86_emulate_ctxt *ctxt)
4175 {
4176 	u64 efer = 0;
4177 
4178 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4179 
4180 	if (!(efer & EFER_SVME))
4181 		return emulate_ud(ctxt);
4182 
4183 	return X86EMUL_CONTINUE;
4184 }
4185 
4186 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4187 {
4188 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4189 
4190 	/* Valid physical address? */
4191 	if (rax & 0xffff000000000000ULL)
4192 		return emulate_gp(ctxt, 0);
4193 
4194 	return check_svme(ctxt);
4195 }
4196 
4197 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4198 {
4199 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4200 
4201 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4202 		return emulate_gp(ctxt, 0);
4203 
4204 	return X86EMUL_CONTINUE;
4205 }
4206 
4207 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4208 {
4209 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4210 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4211 
4212 	/*
4213 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
4214 	 * in Ring3 when CR4.PCE=0.
4215 	 */
4216 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
4217 		return X86EMUL_CONTINUE;
4218 
4219 	/*
4220 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
4221 	 * check however is unnecessary because CPL is always 0 outside
4222 	 * protected mode.
4223 	 */
4224 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4225 	    ctxt->ops->check_pmc(ctxt, rcx))
4226 		return emulate_gp(ctxt, 0);
4227 
4228 	return X86EMUL_CONTINUE;
4229 }
4230 
4231 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4232 {
4233 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4234 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4235 		return emulate_gp(ctxt, 0);
4236 
4237 	return X86EMUL_CONTINUE;
4238 }
4239 
4240 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4241 {
4242 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4243 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4244 		return emulate_gp(ctxt, 0);
4245 
4246 	return X86EMUL_CONTINUE;
4247 }
4248 
4249 #define D(_y) { .flags = (_y) }
4250 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4251 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4252 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4253 #define N    D(NotImpl)
4254 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4255 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4256 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4257 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4258 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4259 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4260 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4261 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4262 #define II(_f, _e, _i) \
4263 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4264 #define IIP(_f, _e, _i, _p) \
4265 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4266 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4267 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4268 
4269 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4270 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4271 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4272 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4273 #define I2bvIP(_f, _e, _i, _p) \
4274 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4275 
4276 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4277 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4278 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4279 
4280 static const struct opcode group7_rm0[] = {
4281 	N,
4282 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4283 	N, N, N, N, N, N,
4284 };
4285 
4286 static const struct opcode group7_rm1[] = {
4287 	DI(SrcNone | Priv, monitor),
4288 	DI(SrcNone | Priv, mwait),
4289 	N, N, N, N, N, N,
4290 };
4291 
4292 static const struct opcode group7_rm2[] = {
4293 	N,
4294 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4295 	N, N, N, N, N, N,
4296 };
4297 
4298 static const struct opcode group7_rm3[] = {
4299 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4300 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4301 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4302 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4303 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4304 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4305 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4306 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4307 };
4308 
4309 static const struct opcode group7_rm7[] = {
4310 	N,
4311 	DIP(SrcNone, rdtscp, check_rdtsc),
4312 	N, N, N, N, N, N,
4313 };
4314 
4315 static const struct opcode group1[] = {
4316 	F(Lock, em_add),
4317 	F(Lock | PageTable, em_or),
4318 	F(Lock, em_adc),
4319 	F(Lock, em_sbb),
4320 	F(Lock | PageTable, em_and),
4321 	F(Lock, em_sub),
4322 	F(Lock, em_xor),
4323 	F(NoWrite, em_cmp),
4324 };
4325 
4326 static const struct opcode group1A[] = {
4327 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4328 };
4329 
4330 static const struct opcode group2[] = {
4331 	F(DstMem | ModRM, em_rol),
4332 	F(DstMem | ModRM, em_ror),
4333 	F(DstMem | ModRM, em_rcl),
4334 	F(DstMem | ModRM, em_rcr),
4335 	F(DstMem | ModRM, em_shl),
4336 	F(DstMem | ModRM, em_shr),
4337 	F(DstMem | ModRM, em_shl),
4338 	F(DstMem | ModRM, em_sar),
4339 };
4340 
4341 static const struct opcode group3[] = {
4342 	F(DstMem | SrcImm | NoWrite, em_test),
4343 	F(DstMem | SrcImm | NoWrite, em_test),
4344 	F(DstMem | SrcNone | Lock, em_not),
4345 	F(DstMem | SrcNone | Lock, em_neg),
4346 	F(DstXacc | Src2Mem, em_mul_ex),
4347 	F(DstXacc | Src2Mem, em_imul_ex),
4348 	F(DstXacc | Src2Mem, em_div_ex),
4349 	F(DstXacc | Src2Mem, em_idiv_ex),
4350 };
4351 
4352 static const struct opcode group4[] = {
4353 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4354 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4355 	N, N, N, N, N, N,
4356 };
4357 
4358 static const struct opcode group5[] = {
4359 	F(DstMem | SrcNone | Lock,		em_inc),
4360 	F(DstMem | SrcNone | Lock,		em_dec),
4361 	I(SrcMem | NearBranch | IsBranch,       em_call_near_abs),
4362 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_call_far),
4363 	I(SrcMem | NearBranch | IsBranch,       em_jmp_abs),
4364 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_jmp_far),
4365 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4366 };
4367 
4368 static const struct opcode group6[] = {
4369 	II(Prot | DstMem,	   em_sldt, sldt),
4370 	II(Prot | DstMem,	   em_str, str),
4371 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4372 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4373 	N, N, N, N,
4374 };
4375 
4376 static const struct group_dual group7 = { {
4377 	II(Mov | DstMem,			em_sgdt, sgdt),
4378 	II(Mov | DstMem,			em_sidt, sidt),
4379 	II(SrcMem | Priv,			em_lgdt, lgdt),
4380 	II(SrcMem | Priv,			em_lidt, lidt),
4381 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4382 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4383 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4384 }, {
4385 	EXT(0, group7_rm0),
4386 	EXT(0, group7_rm1),
4387 	EXT(0, group7_rm2),
4388 	EXT(0, group7_rm3),
4389 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4390 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4391 	EXT(0, group7_rm7),
4392 } };
4393 
4394 static const struct opcode group8[] = {
4395 	N, N, N, N,
4396 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4397 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4398 	F(DstMem | SrcImmByte | Lock,			em_btr),
4399 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4400 };
4401 
4402 /*
4403  * The "memory" destination is actually always a register, since we come
4404  * from the register case of group9.
4405  */
4406 static const struct gprefix pfx_0f_c7_7 = {
4407 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4408 };
4409 
4410 
4411 static const struct group_dual group9 = { {
4412 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4413 }, {
4414 	N, N, N, N, N, N, N,
4415 	GP(0, &pfx_0f_c7_7),
4416 } };
4417 
4418 static const struct opcode group11[] = {
4419 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4420 	X7(D(Undefined)),
4421 };
4422 
4423 static const struct gprefix pfx_0f_ae_7 = {
4424 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4425 };
4426 
4427 static const struct group_dual group15 = { {
4428 	I(ModRM | Aligned16, em_fxsave),
4429 	I(ModRM | Aligned16, em_fxrstor),
4430 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4431 }, {
4432 	N, N, N, N, N, N, N, N,
4433 } };
4434 
4435 static const struct gprefix pfx_0f_6f_0f_7f = {
4436 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4437 };
4438 
4439 static const struct instr_dual instr_dual_0f_2b = {
4440 	I(0, em_mov), N
4441 };
4442 
4443 static const struct gprefix pfx_0f_2b = {
4444 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4445 };
4446 
4447 static const struct gprefix pfx_0f_10_0f_11 = {
4448 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4449 };
4450 
4451 static const struct gprefix pfx_0f_28_0f_29 = {
4452 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4453 };
4454 
4455 static const struct gprefix pfx_0f_e7 = {
4456 	N, I(Sse, em_mov), N, N,
4457 };
4458 
4459 static const struct escape escape_d9 = { {
4460 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4461 }, {
4462 	/* 0xC0 - 0xC7 */
4463 	N, N, N, N, N, N, N, N,
4464 	/* 0xC8 - 0xCF */
4465 	N, N, N, N, N, N, N, N,
4466 	/* 0xD0 - 0xC7 */
4467 	N, N, N, N, N, N, N, N,
4468 	/* 0xD8 - 0xDF */
4469 	N, N, N, N, N, N, N, N,
4470 	/* 0xE0 - 0xE7 */
4471 	N, N, N, N, N, N, N, N,
4472 	/* 0xE8 - 0xEF */
4473 	N, N, N, N, N, N, N, N,
4474 	/* 0xF0 - 0xF7 */
4475 	N, N, N, N, N, N, N, N,
4476 	/* 0xF8 - 0xFF */
4477 	N, N, N, N, N, N, N, N,
4478 } };
4479 
4480 static const struct escape escape_db = { {
4481 	N, N, N, N, N, N, N, N,
4482 }, {
4483 	/* 0xC0 - 0xC7 */
4484 	N, N, N, N, N, N, N, N,
4485 	/* 0xC8 - 0xCF */
4486 	N, N, N, N, N, N, N, N,
4487 	/* 0xD0 - 0xC7 */
4488 	N, N, N, N, N, N, N, N,
4489 	/* 0xD8 - 0xDF */
4490 	N, N, N, N, N, N, N, N,
4491 	/* 0xE0 - 0xE7 */
4492 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4493 	/* 0xE8 - 0xEF */
4494 	N, N, N, N, N, N, N, N,
4495 	/* 0xF0 - 0xF7 */
4496 	N, N, N, N, N, N, N, N,
4497 	/* 0xF8 - 0xFF */
4498 	N, N, N, N, N, N, N, N,
4499 } };
4500 
4501 static const struct escape escape_dd = { {
4502 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4503 }, {
4504 	/* 0xC0 - 0xC7 */
4505 	N, N, N, N, N, N, N, N,
4506 	/* 0xC8 - 0xCF */
4507 	N, N, N, N, N, N, N, N,
4508 	/* 0xD0 - 0xC7 */
4509 	N, N, N, N, N, N, N, N,
4510 	/* 0xD8 - 0xDF */
4511 	N, N, N, N, N, N, N, N,
4512 	/* 0xE0 - 0xE7 */
4513 	N, N, N, N, N, N, N, N,
4514 	/* 0xE8 - 0xEF */
4515 	N, N, N, N, N, N, N, N,
4516 	/* 0xF0 - 0xF7 */
4517 	N, N, N, N, N, N, N, N,
4518 	/* 0xF8 - 0xFF */
4519 	N, N, N, N, N, N, N, N,
4520 } };
4521 
4522 static const struct instr_dual instr_dual_0f_c3 = {
4523 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4524 };
4525 
4526 static const struct mode_dual mode_dual_63 = {
4527 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4528 };
4529 
4530 static const struct opcode opcode_table[256] = {
4531 	/* 0x00 - 0x07 */
4532 	F6ALU(Lock, em_add),
4533 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4534 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4535 	/* 0x08 - 0x0F */
4536 	F6ALU(Lock | PageTable, em_or),
4537 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4538 	N,
4539 	/* 0x10 - 0x17 */
4540 	F6ALU(Lock, em_adc),
4541 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4542 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4543 	/* 0x18 - 0x1F */
4544 	F6ALU(Lock, em_sbb),
4545 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4546 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4547 	/* 0x20 - 0x27 */
4548 	F6ALU(Lock | PageTable, em_and), N, N,
4549 	/* 0x28 - 0x2F */
4550 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4551 	/* 0x30 - 0x37 */
4552 	F6ALU(Lock, em_xor), N, N,
4553 	/* 0x38 - 0x3F */
4554 	F6ALU(NoWrite, em_cmp), N, N,
4555 	/* 0x40 - 0x4F */
4556 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4557 	/* 0x50 - 0x57 */
4558 	X8(I(SrcReg | Stack, em_push)),
4559 	/* 0x58 - 0x5F */
4560 	X8(I(DstReg | Stack, em_pop)),
4561 	/* 0x60 - 0x67 */
4562 	I(ImplicitOps | Stack | No64, em_pusha),
4563 	I(ImplicitOps | Stack | No64, em_popa),
4564 	N, MD(ModRM, &mode_dual_63),
4565 	N, N, N, N,
4566 	/* 0x68 - 0x6F */
4567 	I(SrcImm | Mov | Stack, em_push),
4568 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4569 	I(SrcImmByte | Mov | Stack, em_push),
4570 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4571 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4572 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4573 	/* 0x70 - 0x7F */
4574 	X16(D(SrcImmByte | NearBranch | IsBranch)),
4575 	/* 0x80 - 0x87 */
4576 	G(ByteOp | DstMem | SrcImm, group1),
4577 	G(DstMem | SrcImm, group1),
4578 	G(ByteOp | DstMem | SrcImm | No64, group1),
4579 	G(DstMem | SrcImmByte, group1),
4580 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4581 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4582 	/* 0x88 - 0x8F */
4583 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4584 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4585 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4586 	D(ModRM | SrcMem | NoAccess | DstReg),
4587 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4588 	G(0, group1A),
4589 	/* 0x90 - 0x97 */
4590 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4591 	/* 0x98 - 0x9F */
4592 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4593 	I(SrcImmFAddr | No64 | IsBranch, em_call_far), N,
4594 	II(ImplicitOps | Stack, em_pushf, pushf),
4595 	II(ImplicitOps | Stack, em_popf, popf),
4596 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4597 	/* 0xA0 - 0xA7 */
4598 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4599 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4600 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4601 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4602 	/* 0xA8 - 0xAF */
4603 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4604 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4605 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4606 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4607 	/* 0xB0 - 0xB7 */
4608 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4609 	/* 0xB8 - 0xBF */
4610 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4611 	/* 0xC0 - 0xC7 */
4612 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4613 	I(ImplicitOps | NearBranch | SrcImmU16 | IsBranch, em_ret_near_imm),
4614 	I(ImplicitOps | NearBranch | IsBranch, em_ret),
4615 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4616 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4617 	G(ByteOp, group11), G(0, group11),
4618 	/* 0xC8 - 0xCF */
4619 	I(Stack | SrcImmU16 | Src2ImmByte | IsBranch, em_enter),
4620 	I(Stack | IsBranch, em_leave),
4621 	I(ImplicitOps | SrcImmU16 | IsBranch, em_ret_far_imm),
4622 	I(ImplicitOps | IsBranch, em_ret_far),
4623 	D(ImplicitOps | IsBranch), DI(SrcImmByte | IsBranch, intn),
4624 	D(ImplicitOps | No64 | IsBranch),
4625 	II(ImplicitOps | IsBranch, em_iret, iret),
4626 	/* 0xD0 - 0xD7 */
4627 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4628 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4629 	I(DstAcc | SrcImmUByte | No64, em_aam),
4630 	I(DstAcc | SrcImmUByte | No64, em_aad),
4631 	F(DstAcc | ByteOp | No64, em_salc),
4632 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4633 	/* 0xD8 - 0xDF */
4634 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4635 	/* 0xE0 - 0xE7 */
4636 	X3(I(SrcImmByte | NearBranch | IsBranch, em_loop)),
4637 	I(SrcImmByte | NearBranch | IsBranch, em_jcxz),
4638 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4639 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4640 	/* 0xE8 - 0xEF */
4641 	I(SrcImm | NearBranch | IsBranch, em_call),
4642 	D(SrcImm | ImplicitOps | NearBranch | IsBranch),
4643 	I(SrcImmFAddr | No64 | IsBranch, em_jmp_far),
4644 	D(SrcImmByte | ImplicitOps | NearBranch | IsBranch),
4645 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4646 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4647 	/* 0xF0 - 0xF7 */
4648 	N, DI(ImplicitOps, icebp), N, N,
4649 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4650 	G(ByteOp, group3), G(0, group3),
4651 	/* 0xF8 - 0xFF */
4652 	D(ImplicitOps), D(ImplicitOps),
4653 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4654 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4655 };
4656 
4657 static const struct opcode twobyte_table[256] = {
4658 	/* 0x00 - 0x0F */
4659 	G(0, group6), GD(0, &group7), N, N,
4660 	N, I(ImplicitOps | EmulateOnUD | IsBranch, em_syscall),
4661 	II(ImplicitOps | Priv, em_clts, clts), N,
4662 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4663 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4664 	/* 0x10 - 0x1F */
4665 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4666 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4667 	N, N, N, N, N, N,
4668 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4669 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4670 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4671 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4672 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4673 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4674 	/* 0x20 - 0x2F */
4675 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4676 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4677 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4678 						check_cr_access),
4679 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4680 						check_dr_write),
4681 	N, N, N, N,
4682 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4683 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4684 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4685 	N, N, N, N,
4686 	/* 0x30 - 0x3F */
4687 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4688 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4689 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4690 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4691 	I(ImplicitOps | EmulateOnUD | IsBranch, em_sysenter),
4692 	I(ImplicitOps | Priv | EmulateOnUD | IsBranch, em_sysexit),
4693 	N, N,
4694 	N, N, N, N, N, N, N, N,
4695 	/* 0x40 - 0x4F */
4696 	X16(D(DstReg | SrcMem | ModRM)),
4697 	/* 0x50 - 0x5F */
4698 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4699 	/* 0x60 - 0x6F */
4700 	N, N, N, N,
4701 	N, N, N, N,
4702 	N, N, N, N,
4703 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4704 	/* 0x70 - 0x7F */
4705 	N, N, N, N,
4706 	N, N, N, N,
4707 	N, N, N, N,
4708 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4709 	/* 0x80 - 0x8F */
4710 	X16(D(SrcImm | NearBranch | IsBranch)),
4711 	/* 0x90 - 0x9F */
4712 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4713 	/* 0xA0 - 0xA7 */
4714 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4715 	II(ImplicitOps, em_cpuid, cpuid),
4716 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4717 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4718 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4719 	/* 0xA8 - 0xAF */
4720 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4721 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4722 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4723 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4724 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4725 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4726 	/* 0xB0 - 0xB7 */
4727 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4728 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4729 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4730 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4731 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4732 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4733 	/* 0xB8 - 0xBF */
4734 	N, N,
4735 	G(BitOp, group8),
4736 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4737 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4738 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4739 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4740 	/* 0xC0 - 0xC7 */
4741 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4742 	N, ID(0, &instr_dual_0f_c3),
4743 	N, N, N, GD(0, &group9),
4744 	/* 0xC8 - 0xCF */
4745 	X8(I(DstReg, em_bswap)),
4746 	/* 0xD0 - 0xDF */
4747 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4748 	/* 0xE0 - 0xEF */
4749 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4750 	N, N, N, N, N, N, N, N,
4751 	/* 0xF0 - 0xFF */
4752 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4753 };
4754 
4755 static const struct instr_dual instr_dual_0f_38_f0 = {
4756 	I(DstReg | SrcMem | Mov, em_movbe), N
4757 };
4758 
4759 static const struct instr_dual instr_dual_0f_38_f1 = {
4760 	I(DstMem | SrcReg | Mov, em_movbe), N
4761 };
4762 
4763 static const struct gprefix three_byte_0f_38_f0 = {
4764 	ID(0, &instr_dual_0f_38_f0), N, N, N
4765 };
4766 
4767 static const struct gprefix three_byte_0f_38_f1 = {
4768 	ID(0, &instr_dual_0f_38_f1), N, N, N
4769 };
4770 
4771 /*
4772  * Insns below are selected by the prefix which indexed by the third opcode
4773  * byte.
4774  */
4775 static const struct opcode opcode_map_0f_38[256] = {
4776 	/* 0x00 - 0x7f */
4777 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4778 	/* 0x80 - 0xef */
4779 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4780 	/* 0xf0 - 0xf1 */
4781 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4782 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4783 	/* 0xf2 - 0xff */
4784 	N, N, X4(N), X8(N)
4785 };
4786 
4787 #undef D
4788 #undef N
4789 #undef G
4790 #undef GD
4791 #undef I
4792 #undef GP
4793 #undef EXT
4794 #undef MD
4795 #undef ID
4796 
4797 #undef D2bv
4798 #undef D2bvIP
4799 #undef I2bv
4800 #undef I2bvIP
4801 #undef I6ALU
4802 
4803 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4804 {
4805 	unsigned size;
4806 
4807 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4808 	if (size == 8)
4809 		size = 4;
4810 	return size;
4811 }
4812 
4813 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4814 		      unsigned size, bool sign_extension)
4815 {
4816 	int rc = X86EMUL_CONTINUE;
4817 
4818 	op->type = OP_IMM;
4819 	op->bytes = size;
4820 	op->addr.mem.ea = ctxt->_eip;
4821 	/* NB. Immediates are sign-extended as necessary. */
4822 	switch (op->bytes) {
4823 	case 1:
4824 		op->val = insn_fetch(s8, ctxt);
4825 		break;
4826 	case 2:
4827 		op->val = insn_fetch(s16, ctxt);
4828 		break;
4829 	case 4:
4830 		op->val = insn_fetch(s32, ctxt);
4831 		break;
4832 	case 8:
4833 		op->val = insn_fetch(s64, ctxt);
4834 		break;
4835 	}
4836 	if (!sign_extension) {
4837 		switch (op->bytes) {
4838 		case 1:
4839 			op->val &= 0xff;
4840 			break;
4841 		case 2:
4842 			op->val &= 0xffff;
4843 			break;
4844 		case 4:
4845 			op->val &= 0xffffffff;
4846 			break;
4847 		}
4848 	}
4849 done:
4850 	return rc;
4851 }
4852 
4853 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4854 			  unsigned d)
4855 {
4856 	int rc = X86EMUL_CONTINUE;
4857 
4858 	switch (d) {
4859 	case OpReg:
4860 		decode_register_operand(ctxt, op);
4861 		break;
4862 	case OpImmUByte:
4863 		rc = decode_imm(ctxt, op, 1, false);
4864 		break;
4865 	case OpMem:
4866 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4867 	mem_common:
4868 		*op = ctxt->memop;
4869 		ctxt->memopp = op;
4870 		if (ctxt->d & BitOp)
4871 			fetch_bit_operand(ctxt);
4872 		op->orig_val = op->val;
4873 		break;
4874 	case OpMem64:
4875 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4876 		goto mem_common;
4877 	case OpAcc:
4878 		op->type = OP_REG;
4879 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4880 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4881 		fetch_register_operand(op);
4882 		op->orig_val = op->val;
4883 		break;
4884 	case OpAccLo:
4885 		op->type = OP_REG;
4886 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4887 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4888 		fetch_register_operand(op);
4889 		op->orig_val = op->val;
4890 		break;
4891 	case OpAccHi:
4892 		if (ctxt->d & ByteOp) {
4893 			op->type = OP_NONE;
4894 			break;
4895 		}
4896 		op->type = OP_REG;
4897 		op->bytes = ctxt->op_bytes;
4898 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4899 		fetch_register_operand(op);
4900 		op->orig_val = op->val;
4901 		break;
4902 	case OpDI:
4903 		op->type = OP_MEM;
4904 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4905 		op->addr.mem.ea =
4906 			register_address(ctxt, VCPU_REGS_RDI);
4907 		op->addr.mem.seg = VCPU_SREG_ES;
4908 		op->val = 0;
4909 		op->count = 1;
4910 		break;
4911 	case OpDX:
4912 		op->type = OP_REG;
4913 		op->bytes = 2;
4914 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4915 		fetch_register_operand(op);
4916 		break;
4917 	case OpCL:
4918 		op->type = OP_IMM;
4919 		op->bytes = 1;
4920 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4921 		break;
4922 	case OpImmByte:
4923 		rc = decode_imm(ctxt, op, 1, true);
4924 		break;
4925 	case OpOne:
4926 		op->type = OP_IMM;
4927 		op->bytes = 1;
4928 		op->val = 1;
4929 		break;
4930 	case OpImm:
4931 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4932 		break;
4933 	case OpImm64:
4934 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4935 		break;
4936 	case OpMem8:
4937 		ctxt->memop.bytes = 1;
4938 		if (ctxt->memop.type == OP_REG) {
4939 			ctxt->memop.addr.reg = decode_register(ctxt,
4940 					ctxt->modrm_rm, true);
4941 			fetch_register_operand(&ctxt->memop);
4942 		}
4943 		goto mem_common;
4944 	case OpMem16:
4945 		ctxt->memop.bytes = 2;
4946 		goto mem_common;
4947 	case OpMem32:
4948 		ctxt->memop.bytes = 4;
4949 		goto mem_common;
4950 	case OpImmU16:
4951 		rc = decode_imm(ctxt, op, 2, false);
4952 		break;
4953 	case OpImmU:
4954 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4955 		break;
4956 	case OpSI:
4957 		op->type = OP_MEM;
4958 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4959 		op->addr.mem.ea =
4960 			register_address(ctxt, VCPU_REGS_RSI);
4961 		op->addr.mem.seg = ctxt->seg_override;
4962 		op->val = 0;
4963 		op->count = 1;
4964 		break;
4965 	case OpXLat:
4966 		op->type = OP_MEM;
4967 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4968 		op->addr.mem.ea =
4969 			address_mask(ctxt,
4970 				reg_read(ctxt, VCPU_REGS_RBX) +
4971 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
4972 		op->addr.mem.seg = ctxt->seg_override;
4973 		op->val = 0;
4974 		break;
4975 	case OpImmFAddr:
4976 		op->type = OP_IMM;
4977 		op->addr.mem.ea = ctxt->_eip;
4978 		op->bytes = ctxt->op_bytes + 2;
4979 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
4980 		break;
4981 	case OpMemFAddr:
4982 		ctxt->memop.bytes = ctxt->op_bytes + 2;
4983 		goto mem_common;
4984 	case OpES:
4985 		op->type = OP_IMM;
4986 		op->val = VCPU_SREG_ES;
4987 		break;
4988 	case OpCS:
4989 		op->type = OP_IMM;
4990 		op->val = VCPU_SREG_CS;
4991 		break;
4992 	case OpSS:
4993 		op->type = OP_IMM;
4994 		op->val = VCPU_SREG_SS;
4995 		break;
4996 	case OpDS:
4997 		op->type = OP_IMM;
4998 		op->val = VCPU_SREG_DS;
4999 		break;
5000 	case OpFS:
5001 		op->type = OP_IMM;
5002 		op->val = VCPU_SREG_FS;
5003 		break;
5004 	case OpGS:
5005 		op->type = OP_IMM;
5006 		op->val = VCPU_SREG_GS;
5007 		break;
5008 	case OpImplicit:
5009 		/* Special instructions do their own operand decoding. */
5010 	default:
5011 		op->type = OP_NONE; /* Disable writeback. */
5012 		break;
5013 	}
5014 
5015 done:
5016 	return rc;
5017 }
5018 
5019 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
5020 {
5021 	int rc = X86EMUL_CONTINUE;
5022 	int mode = ctxt->mode;
5023 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
5024 	bool op_prefix = false;
5025 	bool has_seg_override = false;
5026 	struct opcode opcode;
5027 	u16 dummy;
5028 	struct desc_struct desc;
5029 
5030 	ctxt->memop.type = OP_NONE;
5031 	ctxt->memopp = NULL;
5032 	ctxt->_eip = ctxt->eip;
5033 	ctxt->fetch.ptr = ctxt->fetch.data;
5034 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
5035 	ctxt->opcode_len = 1;
5036 	ctxt->intercept = x86_intercept_none;
5037 	if (insn_len > 0)
5038 		memcpy(ctxt->fetch.data, insn, insn_len);
5039 	else {
5040 		rc = __do_insn_fetch_bytes(ctxt, 1);
5041 		if (rc != X86EMUL_CONTINUE)
5042 			goto done;
5043 	}
5044 
5045 	switch (mode) {
5046 	case X86EMUL_MODE_REAL:
5047 	case X86EMUL_MODE_VM86:
5048 		def_op_bytes = def_ad_bytes = 2;
5049 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
5050 		if (desc.d)
5051 			def_op_bytes = def_ad_bytes = 4;
5052 		break;
5053 	case X86EMUL_MODE_PROT16:
5054 		def_op_bytes = def_ad_bytes = 2;
5055 		break;
5056 	case X86EMUL_MODE_PROT32:
5057 		def_op_bytes = def_ad_bytes = 4;
5058 		break;
5059 #ifdef CONFIG_X86_64
5060 	case X86EMUL_MODE_PROT64:
5061 		def_op_bytes = 4;
5062 		def_ad_bytes = 8;
5063 		break;
5064 #endif
5065 	default:
5066 		return EMULATION_FAILED;
5067 	}
5068 
5069 	ctxt->op_bytes = def_op_bytes;
5070 	ctxt->ad_bytes = def_ad_bytes;
5071 
5072 	/* Legacy prefixes. */
5073 	for (;;) {
5074 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5075 		case 0x66:	/* operand-size override */
5076 			op_prefix = true;
5077 			/* switch between 2/4 bytes */
5078 			ctxt->op_bytes = def_op_bytes ^ 6;
5079 			break;
5080 		case 0x67:	/* address-size override */
5081 			if (mode == X86EMUL_MODE_PROT64)
5082 				/* switch between 4/8 bytes */
5083 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5084 			else
5085 				/* switch between 2/4 bytes */
5086 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5087 			break;
5088 		case 0x26:	/* ES override */
5089 			has_seg_override = true;
5090 			ctxt->seg_override = VCPU_SREG_ES;
5091 			break;
5092 		case 0x2e:	/* CS override */
5093 			has_seg_override = true;
5094 			ctxt->seg_override = VCPU_SREG_CS;
5095 			break;
5096 		case 0x36:	/* SS override */
5097 			has_seg_override = true;
5098 			ctxt->seg_override = VCPU_SREG_SS;
5099 			break;
5100 		case 0x3e:	/* DS override */
5101 			has_seg_override = true;
5102 			ctxt->seg_override = VCPU_SREG_DS;
5103 			break;
5104 		case 0x64:	/* FS override */
5105 			has_seg_override = true;
5106 			ctxt->seg_override = VCPU_SREG_FS;
5107 			break;
5108 		case 0x65:	/* GS override */
5109 			has_seg_override = true;
5110 			ctxt->seg_override = VCPU_SREG_GS;
5111 			break;
5112 		case 0x40 ... 0x4f: /* REX */
5113 			if (mode != X86EMUL_MODE_PROT64)
5114 				goto done_prefixes;
5115 			ctxt->rex_prefix = ctxt->b;
5116 			continue;
5117 		case 0xf0:	/* LOCK */
5118 			ctxt->lock_prefix = 1;
5119 			break;
5120 		case 0xf2:	/* REPNE/REPNZ */
5121 		case 0xf3:	/* REP/REPE/REPZ */
5122 			ctxt->rep_prefix = ctxt->b;
5123 			break;
5124 		default:
5125 			goto done_prefixes;
5126 		}
5127 
5128 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5129 
5130 		ctxt->rex_prefix = 0;
5131 	}
5132 
5133 done_prefixes:
5134 
5135 	/* REX prefix. */
5136 	if (ctxt->rex_prefix & 8)
5137 		ctxt->op_bytes = 8;	/* REX.W */
5138 
5139 	/* Opcode byte(s). */
5140 	opcode = opcode_table[ctxt->b];
5141 	/* Two-byte opcode? */
5142 	if (ctxt->b == 0x0f) {
5143 		ctxt->opcode_len = 2;
5144 		ctxt->b = insn_fetch(u8, ctxt);
5145 		opcode = twobyte_table[ctxt->b];
5146 
5147 		/* 0F_38 opcode map */
5148 		if (ctxt->b == 0x38) {
5149 			ctxt->opcode_len = 3;
5150 			ctxt->b = insn_fetch(u8, ctxt);
5151 			opcode = opcode_map_0f_38[ctxt->b];
5152 		}
5153 	}
5154 	ctxt->d = opcode.flags;
5155 
5156 	if (ctxt->d & ModRM)
5157 		ctxt->modrm = insn_fetch(u8, ctxt);
5158 
5159 	/* vex-prefix instructions are not implemented */
5160 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5161 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5162 		ctxt->d = NotImpl;
5163 	}
5164 
5165 	while (ctxt->d & GroupMask) {
5166 		switch (ctxt->d & GroupMask) {
5167 		case Group:
5168 			goffset = (ctxt->modrm >> 3) & 7;
5169 			opcode = opcode.u.group[goffset];
5170 			break;
5171 		case GroupDual:
5172 			goffset = (ctxt->modrm >> 3) & 7;
5173 			if ((ctxt->modrm >> 6) == 3)
5174 				opcode = opcode.u.gdual->mod3[goffset];
5175 			else
5176 				opcode = opcode.u.gdual->mod012[goffset];
5177 			break;
5178 		case RMExt:
5179 			goffset = ctxt->modrm & 7;
5180 			opcode = opcode.u.group[goffset];
5181 			break;
5182 		case Prefix:
5183 			if (ctxt->rep_prefix && op_prefix)
5184 				return EMULATION_FAILED;
5185 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5186 			switch (simd_prefix) {
5187 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5188 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5189 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5190 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5191 			}
5192 			break;
5193 		case Escape:
5194 			if (ctxt->modrm > 0xbf) {
5195 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
5196 				u32 index = array_index_nospec(
5197 					ctxt->modrm - 0xc0, size);
5198 
5199 				opcode = opcode.u.esc->high[index];
5200 			} else {
5201 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5202 			}
5203 			break;
5204 		case InstrDual:
5205 			if ((ctxt->modrm >> 6) == 3)
5206 				opcode = opcode.u.idual->mod3;
5207 			else
5208 				opcode = opcode.u.idual->mod012;
5209 			break;
5210 		case ModeDual:
5211 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5212 				opcode = opcode.u.mdual->mode64;
5213 			else
5214 				opcode = opcode.u.mdual->mode32;
5215 			break;
5216 		default:
5217 			return EMULATION_FAILED;
5218 		}
5219 
5220 		ctxt->d &= ~(u64)GroupMask;
5221 		ctxt->d |= opcode.flags;
5222 	}
5223 
5224 	ctxt->is_branch = opcode.flags & IsBranch;
5225 
5226 	/* Unrecognised? */
5227 	if (ctxt->d == 0)
5228 		return EMULATION_FAILED;
5229 
5230 	ctxt->execute = opcode.u.execute;
5231 
5232 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
5233 	    likely(!(ctxt->d & EmulateOnUD)))
5234 		return EMULATION_FAILED;
5235 
5236 	if (unlikely(ctxt->d &
5237 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5238 	     No16))) {
5239 		/*
5240 		 * These are copied unconditionally here, and checked unconditionally
5241 		 * in x86_emulate_insn.
5242 		 */
5243 		ctxt->check_perm = opcode.check_perm;
5244 		ctxt->intercept = opcode.intercept;
5245 
5246 		if (ctxt->d & NotImpl)
5247 			return EMULATION_FAILED;
5248 
5249 		if (mode == X86EMUL_MODE_PROT64) {
5250 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5251 				ctxt->op_bytes = 8;
5252 			else if (ctxt->d & NearBranch)
5253 				ctxt->op_bytes = 8;
5254 		}
5255 
5256 		if (ctxt->d & Op3264) {
5257 			if (mode == X86EMUL_MODE_PROT64)
5258 				ctxt->op_bytes = 8;
5259 			else
5260 				ctxt->op_bytes = 4;
5261 		}
5262 
5263 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5264 			ctxt->op_bytes = 4;
5265 
5266 		if (ctxt->d & Sse)
5267 			ctxt->op_bytes = 16;
5268 		else if (ctxt->d & Mmx)
5269 			ctxt->op_bytes = 8;
5270 	}
5271 
5272 	/* ModRM and SIB bytes. */
5273 	if (ctxt->d & ModRM) {
5274 		rc = decode_modrm(ctxt, &ctxt->memop);
5275 		if (!has_seg_override) {
5276 			has_seg_override = true;
5277 			ctxt->seg_override = ctxt->modrm_seg;
5278 		}
5279 	} else if (ctxt->d & MemAbs)
5280 		rc = decode_abs(ctxt, &ctxt->memop);
5281 	if (rc != X86EMUL_CONTINUE)
5282 		goto done;
5283 
5284 	if (!has_seg_override)
5285 		ctxt->seg_override = VCPU_SREG_DS;
5286 
5287 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5288 
5289 	/*
5290 	 * Decode and fetch the source operand: register, memory
5291 	 * or immediate.
5292 	 */
5293 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5294 	if (rc != X86EMUL_CONTINUE)
5295 		goto done;
5296 
5297 	/*
5298 	 * Decode and fetch the second source operand: register, memory
5299 	 * or immediate.
5300 	 */
5301 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5302 	if (rc != X86EMUL_CONTINUE)
5303 		goto done;
5304 
5305 	/* Decode and fetch the destination operand: register or memory. */
5306 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5307 
5308 	if (ctxt->rip_relative && likely(ctxt->memopp))
5309 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5310 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5311 
5312 done:
5313 	if (rc == X86EMUL_PROPAGATE_FAULT)
5314 		ctxt->have_exception = true;
5315 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5316 }
5317 
5318 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5319 {
5320 	return ctxt->d & PageTable;
5321 }
5322 
5323 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5324 {
5325 	/* The second termination condition only applies for REPE
5326 	 * and REPNE. Test if the repeat string operation prefix is
5327 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5328 	 * corresponding termination condition according to:
5329 	 * 	- if REPE/REPZ and ZF = 0 then done
5330 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5331 	 */
5332 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5333 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5334 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5335 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5336 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5337 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5338 		return true;
5339 
5340 	return false;
5341 }
5342 
5343 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5344 {
5345 	int rc;
5346 
5347 	kvm_fpu_get();
5348 	rc = asm_safe("fwait");
5349 	kvm_fpu_put();
5350 
5351 	if (unlikely(rc != X86EMUL_CONTINUE))
5352 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5353 
5354 	return X86EMUL_CONTINUE;
5355 }
5356 
5357 static void fetch_possible_mmx_operand(struct operand *op)
5358 {
5359 	if (op->type == OP_MM)
5360 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5361 }
5362 
5363 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5364 {
5365 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5366 
5367 	if (!(ctxt->d & ByteOp))
5368 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5369 
5370 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5371 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5372 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5373 	    : "c"(ctxt->src2.val));
5374 
5375 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5376 	if (!fop) /* exception is returned in fop variable */
5377 		return emulate_de(ctxt);
5378 	return X86EMUL_CONTINUE;
5379 }
5380 
5381 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5382 {
5383 	memset(&ctxt->rip_relative, 0,
5384 	       (void *)&ctxt->modrm - (void *)&ctxt->rip_relative);
5385 
5386 	ctxt->io_read.pos = 0;
5387 	ctxt->io_read.end = 0;
5388 	ctxt->mem_read.end = 0;
5389 }
5390 
5391 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5392 {
5393 	const struct x86_emulate_ops *ops = ctxt->ops;
5394 	int rc = X86EMUL_CONTINUE;
5395 	int saved_dst_type = ctxt->dst.type;
5396 	unsigned emul_flags;
5397 
5398 	ctxt->mem_read.pos = 0;
5399 
5400 	/* LOCK prefix is allowed only with some instructions */
5401 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5402 		rc = emulate_ud(ctxt);
5403 		goto done;
5404 	}
5405 
5406 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5407 		rc = emulate_ud(ctxt);
5408 		goto done;
5409 	}
5410 
5411 	emul_flags = ctxt->ops->get_hflags(ctxt);
5412 	if (unlikely(ctxt->d &
5413 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5414 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5415 				(ctxt->d & Undefined)) {
5416 			rc = emulate_ud(ctxt);
5417 			goto done;
5418 		}
5419 
5420 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5421 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5422 			rc = emulate_ud(ctxt);
5423 			goto done;
5424 		}
5425 
5426 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5427 			rc = emulate_nm(ctxt);
5428 			goto done;
5429 		}
5430 
5431 		if (ctxt->d & Mmx) {
5432 			rc = flush_pending_x87_faults(ctxt);
5433 			if (rc != X86EMUL_CONTINUE)
5434 				goto done;
5435 			/*
5436 			 * Now that we know the fpu is exception safe, we can fetch
5437 			 * operands from it.
5438 			 */
5439 			fetch_possible_mmx_operand(&ctxt->src);
5440 			fetch_possible_mmx_operand(&ctxt->src2);
5441 			if (!(ctxt->d & Mov))
5442 				fetch_possible_mmx_operand(&ctxt->dst);
5443 		}
5444 
5445 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5446 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5447 						      X86_ICPT_PRE_EXCEPT);
5448 			if (rc != X86EMUL_CONTINUE)
5449 				goto done;
5450 		}
5451 
5452 		/* Instruction can only be executed in protected mode */
5453 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5454 			rc = emulate_ud(ctxt);
5455 			goto done;
5456 		}
5457 
5458 		/* Privileged instruction can be executed only in CPL=0 */
5459 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5460 			if (ctxt->d & PrivUD)
5461 				rc = emulate_ud(ctxt);
5462 			else
5463 				rc = emulate_gp(ctxt, 0);
5464 			goto done;
5465 		}
5466 
5467 		/* Do instruction specific permission checks */
5468 		if (ctxt->d & CheckPerm) {
5469 			rc = ctxt->check_perm(ctxt);
5470 			if (rc != X86EMUL_CONTINUE)
5471 				goto done;
5472 		}
5473 
5474 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5475 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5476 						      X86_ICPT_POST_EXCEPT);
5477 			if (rc != X86EMUL_CONTINUE)
5478 				goto done;
5479 		}
5480 
5481 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5482 			/* All REP prefixes have the same first termination condition */
5483 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5484 				string_registers_quirk(ctxt);
5485 				ctxt->eip = ctxt->_eip;
5486 				ctxt->eflags &= ~X86_EFLAGS_RF;
5487 				goto done;
5488 			}
5489 		}
5490 	}
5491 
5492 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5493 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5494 				    ctxt->src.valptr, ctxt->src.bytes);
5495 		if (rc != X86EMUL_CONTINUE)
5496 			goto done;
5497 		ctxt->src.orig_val64 = ctxt->src.val64;
5498 	}
5499 
5500 	if (ctxt->src2.type == OP_MEM) {
5501 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5502 				    &ctxt->src2.val, ctxt->src2.bytes);
5503 		if (rc != X86EMUL_CONTINUE)
5504 			goto done;
5505 	}
5506 
5507 	if ((ctxt->d & DstMask) == ImplicitOps)
5508 		goto special_insn;
5509 
5510 
5511 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5512 		/* optimisation - avoid slow emulated read if Mov */
5513 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5514 				   &ctxt->dst.val, ctxt->dst.bytes);
5515 		if (rc != X86EMUL_CONTINUE) {
5516 			if (!(ctxt->d & NoWrite) &&
5517 			    rc == X86EMUL_PROPAGATE_FAULT &&
5518 			    ctxt->exception.vector == PF_VECTOR)
5519 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5520 			goto done;
5521 		}
5522 	}
5523 	/* Copy full 64-bit value for CMPXCHG8B.  */
5524 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5525 
5526 special_insn:
5527 
5528 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5529 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5530 					      X86_ICPT_POST_MEMACCESS);
5531 		if (rc != X86EMUL_CONTINUE)
5532 			goto done;
5533 	}
5534 
5535 	if (ctxt->rep_prefix && (ctxt->d & String))
5536 		ctxt->eflags |= X86_EFLAGS_RF;
5537 	else
5538 		ctxt->eflags &= ~X86_EFLAGS_RF;
5539 
5540 	if (ctxt->execute) {
5541 		if (ctxt->d & Fastop)
5542 			rc = fastop(ctxt, ctxt->fop);
5543 		else
5544 			rc = ctxt->execute(ctxt);
5545 		if (rc != X86EMUL_CONTINUE)
5546 			goto done;
5547 		goto writeback;
5548 	}
5549 
5550 	if (ctxt->opcode_len == 2)
5551 		goto twobyte_insn;
5552 	else if (ctxt->opcode_len == 3)
5553 		goto threebyte_insn;
5554 
5555 	switch (ctxt->b) {
5556 	case 0x70 ... 0x7f: /* jcc (short) */
5557 		if (test_cc(ctxt->b, ctxt->eflags))
5558 			rc = jmp_rel(ctxt, ctxt->src.val);
5559 		break;
5560 	case 0x8d: /* lea r16/r32, m */
5561 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5562 		break;
5563 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5564 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5565 			ctxt->dst.type = OP_NONE;
5566 		else
5567 			rc = em_xchg(ctxt);
5568 		break;
5569 	case 0x98: /* cbw/cwde/cdqe */
5570 		switch (ctxt->op_bytes) {
5571 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5572 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5573 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5574 		}
5575 		break;
5576 	case 0xcc:		/* int3 */
5577 		rc = emulate_int(ctxt, 3);
5578 		break;
5579 	case 0xcd:		/* int n */
5580 		rc = emulate_int(ctxt, ctxt->src.val);
5581 		break;
5582 	case 0xce:		/* into */
5583 		if (ctxt->eflags & X86_EFLAGS_OF)
5584 			rc = emulate_int(ctxt, 4);
5585 		break;
5586 	case 0xe9: /* jmp rel */
5587 	case 0xeb: /* jmp rel short */
5588 		rc = jmp_rel(ctxt, ctxt->src.val);
5589 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5590 		break;
5591 	case 0xf4:              /* hlt */
5592 		ctxt->ops->halt(ctxt);
5593 		break;
5594 	case 0xf5:	/* cmc */
5595 		/* complement carry flag from eflags reg */
5596 		ctxt->eflags ^= X86_EFLAGS_CF;
5597 		break;
5598 	case 0xf8: /* clc */
5599 		ctxt->eflags &= ~X86_EFLAGS_CF;
5600 		break;
5601 	case 0xf9: /* stc */
5602 		ctxt->eflags |= X86_EFLAGS_CF;
5603 		break;
5604 	case 0xfc: /* cld */
5605 		ctxt->eflags &= ~X86_EFLAGS_DF;
5606 		break;
5607 	case 0xfd: /* std */
5608 		ctxt->eflags |= X86_EFLAGS_DF;
5609 		break;
5610 	default:
5611 		goto cannot_emulate;
5612 	}
5613 
5614 	if (rc != X86EMUL_CONTINUE)
5615 		goto done;
5616 
5617 writeback:
5618 	if (ctxt->d & SrcWrite) {
5619 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5620 		rc = writeback(ctxt, &ctxt->src);
5621 		if (rc != X86EMUL_CONTINUE)
5622 			goto done;
5623 	}
5624 	if (!(ctxt->d & NoWrite)) {
5625 		rc = writeback(ctxt, &ctxt->dst);
5626 		if (rc != X86EMUL_CONTINUE)
5627 			goto done;
5628 	}
5629 
5630 	/*
5631 	 * restore dst type in case the decoding will be reused
5632 	 * (happens for string instruction )
5633 	 */
5634 	ctxt->dst.type = saved_dst_type;
5635 
5636 	if ((ctxt->d & SrcMask) == SrcSI)
5637 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5638 
5639 	if ((ctxt->d & DstMask) == DstDI)
5640 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5641 
5642 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5643 		unsigned int count;
5644 		struct read_cache *r = &ctxt->io_read;
5645 		if ((ctxt->d & SrcMask) == SrcSI)
5646 			count = ctxt->src.count;
5647 		else
5648 			count = ctxt->dst.count;
5649 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5650 
5651 		if (!string_insn_completed(ctxt)) {
5652 			/*
5653 			 * Re-enter guest when pio read ahead buffer is empty
5654 			 * or, if it is not used, after each 1024 iteration.
5655 			 */
5656 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5657 			    (r->end == 0 || r->end != r->pos)) {
5658 				/*
5659 				 * Reset read cache. Usually happens before
5660 				 * decode, but since instruction is restarted
5661 				 * we have to do it here.
5662 				 */
5663 				ctxt->mem_read.end = 0;
5664 				writeback_registers(ctxt);
5665 				return EMULATION_RESTART;
5666 			}
5667 			goto done; /* skip rip writeback */
5668 		}
5669 		ctxt->eflags &= ~X86_EFLAGS_RF;
5670 	}
5671 
5672 	ctxt->eip = ctxt->_eip;
5673 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5674 		ctxt->eip = (u32)ctxt->_eip;
5675 
5676 done:
5677 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5678 		WARN_ON(ctxt->exception.vector > 0x1f);
5679 		ctxt->have_exception = true;
5680 	}
5681 	if (rc == X86EMUL_INTERCEPTED)
5682 		return EMULATION_INTERCEPTED;
5683 
5684 	if (rc == X86EMUL_CONTINUE)
5685 		writeback_registers(ctxt);
5686 
5687 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5688 
5689 twobyte_insn:
5690 	switch (ctxt->b) {
5691 	case 0x09:		/* wbinvd */
5692 		(ctxt->ops->wbinvd)(ctxt);
5693 		break;
5694 	case 0x08:		/* invd */
5695 	case 0x0d:		/* GrpP (prefetch) */
5696 	case 0x18:		/* Grp16 (prefetch/nop) */
5697 	case 0x1f:		/* nop */
5698 		break;
5699 	case 0x20: /* mov cr, reg */
5700 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5701 		break;
5702 	case 0x21: /* mov from dr to reg */
5703 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5704 		break;
5705 	case 0x40 ... 0x4f:	/* cmov */
5706 		if (test_cc(ctxt->b, ctxt->eflags))
5707 			ctxt->dst.val = ctxt->src.val;
5708 		else if (ctxt->op_bytes != 4)
5709 			ctxt->dst.type = OP_NONE; /* no writeback */
5710 		break;
5711 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5712 		if (test_cc(ctxt->b, ctxt->eflags))
5713 			rc = jmp_rel(ctxt, ctxt->src.val);
5714 		break;
5715 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5716 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5717 		break;
5718 	case 0xb6 ... 0xb7:	/* movzx */
5719 		ctxt->dst.bytes = ctxt->op_bytes;
5720 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5721 						       : (u16) ctxt->src.val;
5722 		break;
5723 	case 0xbe ... 0xbf:	/* movsx */
5724 		ctxt->dst.bytes = ctxt->op_bytes;
5725 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5726 							(s16) ctxt->src.val;
5727 		break;
5728 	default:
5729 		goto cannot_emulate;
5730 	}
5731 
5732 threebyte_insn:
5733 
5734 	if (rc != X86EMUL_CONTINUE)
5735 		goto done;
5736 
5737 	goto writeback;
5738 
5739 cannot_emulate:
5740 	return EMULATION_FAILED;
5741 }
5742 
5743 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5744 {
5745 	invalidate_registers(ctxt);
5746 }
5747 
5748 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5749 {
5750 	writeback_registers(ctxt);
5751 }
5752 
5753 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5754 {
5755 	if (ctxt->rep_prefix && (ctxt->d & String))
5756 		return false;
5757 
5758 	if (ctxt->d & TwoMemOp)
5759 		return false;
5760 
5761 	return true;
5762 }
5763