xref: /openbmc/linux/arch/x86/kvm/emulate.c (revision e6e8c6c2)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3  * emulate.c
4  *
5  * Generic x86 (32-bit and 64-bit) instruction decoder and emulator.
6  *
7  * Copyright (c) 2005 Keir Fraser
8  *
9  * Linux coding style, mod r/m decoder, segment base fixes, real-mode
10  * privileged instructions:
11  *
12  * Copyright (C) 2006 Qumranet
13  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
14  *
15  *   Avi Kivity <avi@qumranet.com>
16  *   Yaniv Kamay <yaniv@qumranet.com>
17  *
18  * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4
19  */
20 
21 #include <linux/kvm_host.h>
22 #include "kvm_cache_regs.h"
23 #include "kvm_emulate.h"
24 #include <linux/stringify.h>
25 #include <asm/debugreg.h>
26 #include <asm/nospec-branch.h>
27 #include <asm/ibt.h>
28 
29 #include "x86.h"
30 #include "tss.h"
31 #include "mmu.h"
32 #include "pmu.h"
33 
34 /*
35  * Operand types
36  */
37 #define OpNone             0ull
38 #define OpImplicit         1ull  /* No generic decode */
39 #define OpReg              2ull  /* Register */
40 #define OpMem              3ull  /* Memory */
41 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
42 #define OpDI               5ull  /* ES:DI/EDI/RDI */
43 #define OpMem64            6ull  /* Memory, 64-bit */
44 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
45 #define OpDX               8ull  /* DX register */
46 #define OpCL               9ull  /* CL register (for shifts) */
47 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
48 #define OpOne             11ull  /* Implied 1 */
49 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
50 #define OpMem16           13ull  /* Memory operand (16-bit). */
51 #define OpMem32           14ull  /* Memory operand (32-bit). */
52 #define OpImmU            15ull  /* Immediate operand, zero extended */
53 #define OpSI              16ull  /* SI/ESI/RSI */
54 #define OpImmFAddr        17ull  /* Immediate far address */
55 #define OpMemFAddr        18ull  /* Far address in memory */
56 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
57 #define OpES              20ull  /* ES */
58 #define OpCS              21ull  /* CS */
59 #define OpSS              22ull  /* SS */
60 #define OpDS              23ull  /* DS */
61 #define OpFS              24ull  /* FS */
62 #define OpGS              25ull  /* GS */
63 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
64 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
65 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
66 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
67 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
68 
69 #define OpBits             5  /* Width of operand field */
70 #define OpMask             ((1ull << OpBits) - 1)
71 
72 /*
73  * Opcode effective-address decode tables.
74  * Note that we only emulate instructions that have at least one memory
75  * operand (excluding implicit stack references). We assume that stack
76  * references and instruction fetches will never occur in special memory
77  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
78  * not be handled.
79  */
80 
81 /* Operand sizes: 8-bit operands or specified/overridden size. */
82 #define ByteOp      (1<<0)	/* 8-bit operands. */
83 /* Destination operand type. */
84 #define DstShift    1
85 #define ImplicitOps (OpImplicit << DstShift)
86 #define DstReg      (OpReg << DstShift)
87 #define DstMem      (OpMem << DstShift)
88 #define DstAcc      (OpAcc << DstShift)
89 #define DstDI       (OpDI << DstShift)
90 #define DstMem64    (OpMem64 << DstShift)
91 #define DstMem16    (OpMem16 << DstShift)
92 #define DstImmUByte (OpImmUByte << DstShift)
93 #define DstDX       (OpDX << DstShift)
94 #define DstAccLo    (OpAccLo << DstShift)
95 #define DstMask     (OpMask << DstShift)
96 /* Source operand type. */
97 #define SrcShift    6
98 #define SrcNone     (OpNone << SrcShift)
99 #define SrcReg      (OpReg << SrcShift)
100 #define SrcMem      (OpMem << SrcShift)
101 #define SrcMem16    (OpMem16 << SrcShift)
102 #define SrcMem32    (OpMem32 << SrcShift)
103 #define SrcImm      (OpImm << SrcShift)
104 #define SrcImmByte  (OpImmByte << SrcShift)
105 #define SrcOne      (OpOne << SrcShift)
106 #define SrcImmUByte (OpImmUByte << SrcShift)
107 #define SrcImmU     (OpImmU << SrcShift)
108 #define SrcSI       (OpSI << SrcShift)
109 #define SrcXLat     (OpXLat << SrcShift)
110 #define SrcImmFAddr (OpImmFAddr << SrcShift)
111 #define SrcMemFAddr (OpMemFAddr << SrcShift)
112 #define SrcAcc      (OpAcc << SrcShift)
113 #define SrcImmU16   (OpImmU16 << SrcShift)
114 #define SrcImm64    (OpImm64 << SrcShift)
115 #define SrcDX       (OpDX << SrcShift)
116 #define SrcMem8     (OpMem8 << SrcShift)
117 #define SrcAccHi    (OpAccHi << SrcShift)
118 #define SrcMask     (OpMask << SrcShift)
119 #define BitOp       (1<<11)
120 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
121 #define String      (1<<13)     /* String instruction (rep capable) */
122 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
123 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
124 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
125 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
126 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
127 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
128 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
129 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
130 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
131 #define Sse         (1<<18)     /* SSE Vector instruction */
132 /* Generic ModRM decode. */
133 #define ModRM       (1<<19)
134 /* Destination is only written; never read. */
135 #define Mov         (1<<20)
136 /* Misc flags */
137 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
138 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
139 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
140 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
141 #define Undefined   (1<<25) /* No Such Instruction */
142 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
143 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
144 #define No64	    (1<<28)
145 #define PageTable   (1 << 29)   /* instruction used to write page table */
146 #define NotImpl     (1 << 30)   /* instruction is not implemented */
147 /* Source 2 operand type */
148 #define Src2Shift   (31)
149 #define Src2None    (OpNone << Src2Shift)
150 #define Src2Mem     (OpMem << Src2Shift)
151 #define Src2CL      (OpCL << Src2Shift)
152 #define Src2ImmByte (OpImmByte << Src2Shift)
153 #define Src2One     (OpOne << Src2Shift)
154 #define Src2Imm     (OpImm << Src2Shift)
155 #define Src2ES      (OpES << Src2Shift)
156 #define Src2CS      (OpCS << Src2Shift)
157 #define Src2SS      (OpSS << Src2Shift)
158 #define Src2DS      (OpDS << Src2Shift)
159 #define Src2FS      (OpFS << Src2Shift)
160 #define Src2GS      (OpGS << Src2Shift)
161 #define Src2Mask    (OpMask << Src2Shift)
162 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
163 #define AlignMask   ((u64)7 << 41)
164 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
165 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
166 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
167 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
168 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
169 #define NoWrite     ((u64)1 << 45)  /* No writeback */
170 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
171 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
172 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
173 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
174 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
175 #define NearBranch  ((u64)1 << 52)  /* Near branches */
176 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
177 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
178 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
179 #define IsBranch    ((u64)1 << 56)  /* Instruction is considered a branch. */
180 
181 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
182 
183 #define X2(x...) x, x
184 #define X3(x...) X2(x), x
185 #define X4(x...) X2(x), X2(x)
186 #define X5(x...) X4(x), x
187 #define X6(x...) X4(x), X2(x)
188 #define X7(x...) X4(x), X3(x)
189 #define X8(x...) X4(x), X4(x)
190 #define X16(x...) X8(x), X8(x)
191 
192 struct opcode {
193 	u64 flags;
194 	u8 intercept;
195 	u8 pad[7];
196 	union {
197 		int (*execute)(struct x86_emulate_ctxt *ctxt);
198 		const struct opcode *group;
199 		const struct group_dual *gdual;
200 		const struct gprefix *gprefix;
201 		const struct escape *esc;
202 		const struct instr_dual *idual;
203 		const struct mode_dual *mdual;
204 		void (*fastop)(struct fastop *fake);
205 	} u;
206 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
207 };
208 
209 struct group_dual {
210 	struct opcode mod012[8];
211 	struct opcode mod3[8];
212 };
213 
214 struct gprefix {
215 	struct opcode pfx_no;
216 	struct opcode pfx_66;
217 	struct opcode pfx_f2;
218 	struct opcode pfx_f3;
219 };
220 
221 struct escape {
222 	struct opcode op[8];
223 	struct opcode high[64];
224 };
225 
226 struct instr_dual {
227 	struct opcode mod012;
228 	struct opcode mod3;
229 };
230 
231 struct mode_dual {
232 	struct opcode mode32;
233 	struct opcode mode64;
234 };
235 
236 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
237 
238 enum x86_transfer_type {
239 	X86_TRANSFER_NONE,
240 	X86_TRANSFER_CALL_JMP,
241 	X86_TRANSFER_RET,
242 	X86_TRANSFER_TASK_SWITCH,
243 };
244 
245 static ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr)
246 {
247 	if (KVM_EMULATOR_BUG_ON(nr >= NR_EMULATOR_GPRS, ctxt))
248 		nr &= NR_EMULATOR_GPRS - 1;
249 
250 	if (!(ctxt->regs_valid & (1 << nr))) {
251 		ctxt->regs_valid |= 1 << nr;
252 		ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr);
253 	}
254 	return ctxt->_regs[nr];
255 }
256 
257 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr)
258 {
259 	if (KVM_EMULATOR_BUG_ON(nr >= NR_EMULATOR_GPRS, ctxt))
260 		nr &= NR_EMULATOR_GPRS - 1;
261 
262 	BUILD_BUG_ON(sizeof(ctxt->regs_dirty) * BITS_PER_BYTE < NR_EMULATOR_GPRS);
263 	BUILD_BUG_ON(sizeof(ctxt->regs_valid) * BITS_PER_BYTE < NR_EMULATOR_GPRS);
264 
265 	ctxt->regs_valid |= 1 << nr;
266 	ctxt->regs_dirty |= 1 << nr;
267 	return &ctxt->_regs[nr];
268 }
269 
270 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr)
271 {
272 	reg_read(ctxt, nr);
273 	return reg_write(ctxt, nr);
274 }
275 
276 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
277 {
278 	unsigned long dirty = ctxt->regs_dirty;
279 	unsigned reg;
280 
281 	for_each_set_bit(reg, &dirty, NR_EMULATOR_GPRS)
282 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
283 }
284 
285 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
286 {
287 	ctxt->regs_dirty = 0;
288 	ctxt->regs_valid = 0;
289 }
290 
291 /*
292  * These EFLAGS bits are restored from saved value during emulation, and
293  * any changes are written back to the saved value after emulation.
294  */
295 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
296 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
297 
298 #ifdef CONFIG_X86_64
299 #define ON64(x) x
300 #else
301 #define ON64(x)
302 #endif
303 
304 /*
305  * fastop functions have a special calling convention:
306  *
307  * dst:    rax        (in/out)
308  * src:    rdx        (in/out)
309  * src2:   rcx        (in)
310  * flags:  rflags     (in/out)
311  * ex:     rsi        (in:fastop pointer, out:zero if exception)
312  *
313  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
314  * different operand sizes can be reached by calculation, rather than a jump
315  * table (which would be bigger than the code).
316  *
317  * The 16 byte alignment, considering 5 bytes for the RET thunk, 3 for ENDBR
318  * and 1 for the straight line speculation INT3, leaves 7 bytes for the
319  * body of the function.  Currently none is larger than 4.
320  */
321 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
322 
323 #define FASTOP_SIZE	16
324 
325 #define __FOP_FUNC(name) \
326 	".align " __stringify(FASTOP_SIZE) " \n\t" \
327 	".type " name ", @function \n\t" \
328 	name ":\n\t" \
329 	ASM_ENDBR \
330 	IBT_NOSEAL(name)
331 
332 #define FOP_FUNC(name) \
333 	__FOP_FUNC(#name)
334 
335 #define __FOP_RET(name) \
336 	"11: " ASM_RET \
337 	".size " name ", .-" name "\n\t"
338 
339 #define FOP_RET(name) \
340 	__FOP_RET(#name)
341 
342 #define __FOP_START(op, align) \
343 	extern void em_##op(struct fastop *fake); \
344 	asm(".pushsection .text, \"ax\" \n\t" \
345 	    ".global em_" #op " \n\t" \
346 	    ".align " __stringify(align) " \n\t" \
347 	    "em_" #op ":\n\t"
348 
349 #define FOP_START(op) __FOP_START(op, FASTOP_SIZE)
350 
351 #define FOP_END \
352 	    ".popsection")
353 
354 #define __FOPNOP(name) \
355 	__FOP_FUNC(name) \
356 	__FOP_RET(name)
357 
358 #define FOPNOP() \
359 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
360 
361 #define FOP1E(op,  dst) \
362 	__FOP_FUNC(#op "_" #dst) \
363 	"10: " #op " %" #dst " \n\t" \
364 	__FOP_RET(#op "_" #dst)
365 
366 #define FOP1EEX(op,  dst) \
367 	FOP1E(op, dst) _ASM_EXTABLE_TYPE_REG(10b, 11b, EX_TYPE_ZERO_REG, %%esi)
368 
369 #define FASTOP1(op) \
370 	FOP_START(op) \
371 	FOP1E(op##b, al) \
372 	FOP1E(op##w, ax) \
373 	FOP1E(op##l, eax) \
374 	ON64(FOP1E(op##q, rax))	\
375 	FOP_END
376 
377 /* 1-operand, using src2 (for MUL/DIV r/m) */
378 #define FASTOP1SRC2(op, name) \
379 	FOP_START(name) \
380 	FOP1E(op, cl) \
381 	FOP1E(op, cx) \
382 	FOP1E(op, ecx) \
383 	ON64(FOP1E(op, rcx)) \
384 	FOP_END
385 
386 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
387 #define FASTOP1SRC2EX(op, name) \
388 	FOP_START(name) \
389 	FOP1EEX(op, cl) \
390 	FOP1EEX(op, cx) \
391 	FOP1EEX(op, ecx) \
392 	ON64(FOP1EEX(op, rcx)) \
393 	FOP_END
394 
395 #define FOP2E(op,  dst, src)	   \
396 	__FOP_FUNC(#op "_" #dst "_" #src) \
397 	#op " %" #src ", %" #dst " \n\t" \
398 	__FOP_RET(#op "_" #dst "_" #src)
399 
400 #define FASTOP2(op) \
401 	FOP_START(op) \
402 	FOP2E(op##b, al, dl) \
403 	FOP2E(op##w, ax, dx) \
404 	FOP2E(op##l, eax, edx) \
405 	ON64(FOP2E(op##q, rax, rdx)) \
406 	FOP_END
407 
408 /* 2 operand, word only */
409 #define FASTOP2W(op) \
410 	FOP_START(op) \
411 	FOPNOP() \
412 	FOP2E(op##w, ax, dx) \
413 	FOP2E(op##l, eax, edx) \
414 	ON64(FOP2E(op##q, rax, rdx)) \
415 	FOP_END
416 
417 /* 2 operand, src is CL */
418 #define FASTOP2CL(op) \
419 	FOP_START(op) \
420 	FOP2E(op##b, al, cl) \
421 	FOP2E(op##w, ax, cl) \
422 	FOP2E(op##l, eax, cl) \
423 	ON64(FOP2E(op##q, rax, cl)) \
424 	FOP_END
425 
426 /* 2 operand, src and dest are reversed */
427 #define FASTOP2R(op, name) \
428 	FOP_START(name) \
429 	FOP2E(op##b, dl, al) \
430 	FOP2E(op##w, dx, ax) \
431 	FOP2E(op##l, edx, eax) \
432 	ON64(FOP2E(op##q, rdx, rax)) \
433 	FOP_END
434 
435 #define FOP3E(op,  dst, src, src2) \
436 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
437 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
438 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
439 
440 /* 3-operand, word-only, src2=cl */
441 #define FASTOP3WCL(op) \
442 	FOP_START(op) \
443 	FOPNOP() \
444 	FOP3E(op##w, ax, dx, cl) \
445 	FOP3E(op##l, eax, edx, cl) \
446 	ON64(FOP3E(op##q, rax, rdx, cl)) \
447 	FOP_END
448 
449 /* Special case for SETcc - 1 instruction per cc */
450 #define FOP_SETCC(op) \
451 	FOP_FUNC(op) \
452 	#op " %al \n\t" \
453 	FOP_RET(op)
454 
455 FOP_START(setcc)
456 FOP_SETCC(seto)
457 FOP_SETCC(setno)
458 FOP_SETCC(setc)
459 FOP_SETCC(setnc)
460 FOP_SETCC(setz)
461 FOP_SETCC(setnz)
462 FOP_SETCC(setbe)
463 FOP_SETCC(setnbe)
464 FOP_SETCC(sets)
465 FOP_SETCC(setns)
466 FOP_SETCC(setp)
467 FOP_SETCC(setnp)
468 FOP_SETCC(setl)
469 FOP_SETCC(setnl)
470 FOP_SETCC(setle)
471 FOP_SETCC(setnle)
472 FOP_END;
473 
474 FOP_START(salc)
475 FOP_FUNC(salc)
476 "pushf; sbb %al, %al; popf \n\t"
477 FOP_RET(salc)
478 FOP_END;
479 
480 /*
481  * XXX: inoutclob user must know where the argument is being expanded.
482  *      Using asm goto would allow us to remove _fault.
483  */
484 #define asm_safe(insn, inoutclob...) \
485 ({ \
486 	int _fault = 0; \
487  \
488 	asm volatile("1:" insn "\n" \
489 	             "2:\n" \
490 		     _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_ONE_REG, %[_fault]) \
491 	             : [_fault] "+r"(_fault) inoutclob ); \
492  \
493 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
494 })
495 
496 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
497 				    enum x86_intercept intercept,
498 				    enum x86_intercept_stage stage)
499 {
500 	struct x86_instruction_info info = {
501 		.intercept  = intercept,
502 		.rep_prefix = ctxt->rep_prefix,
503 		.modrm_mod  = ctxt->modrm_mod,
504 		.modrm_reg  = ctxt->modrm_reg,
505 		.modrm_rm   = ctxt->modrm_rm,
506 		.src_val    = ctxt->src.val64,
507 		.dst_val    = ctxt->dst.val64,
508 		.src_bytes  = ctxt->src.bytes,
509 		.dst_bytes  = ctxt->dst.bytes,
510 		.ad_bytes   = ctxt->ad_bytes,
511 		.next_rip   = ctxt->eip,
512 	};
513 
514 	return ctxt->ops->intercept(ctxt, &info, stage);
515 }
516 
517 static void assign_masked(ulong *dest, ulong src, ulong mask)
518 {
519 	*dest = (*dest & ~mask) | (src & mask);
520 }
521 
522 static void assign_register(unsigned long *reg, u64 val, int bytes)
523 {
524 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
525 	switch (bytes) {
526 	case 1:
527 		*(u8 *)reg = (u8)val;
528 		break;
529 	case 2:
530 		*(u16 *)reg = (u16)val;
531 		break;
532 	case 4:
533 		*reg = (u32)val;
534 		break;	/* 64b: zero-extend */
535 	case 8:
536 		*reg = val;
537 		break;
538 	}
539 }
540 
541 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
542 {
543 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
544 }
545 
546 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
547 {
548 	u16 sel;
549 	struct desc_struct ss;
550 
551 	if (ctxt->mode == X86EMUL_MODE_PROT64)
552 		return ~0UL;
553 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
554 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
555 }
556 
557 static int stack_size(struct x86_emulate_ctxt *ctxt)
558 {
559 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
560 }
561 
562 /* Access/update address held in a register, based on addressing mode. */
563 static inline unsigned long
564 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
565 {
566 	if (ctxt->ad_bytes == sizeof(unsigned long))
567 		return reg;
568 	else
569 		return reg & ad_mask(ctxt);
570 }
571 
572 static inline unsigned long
573 register_address(struct x86_emulate_ctxt *ctxt, int reg)
574 {
575 	return address_mask(ctxt, reg_read(ctxt, reg));
576 }
577 
578 static void masked_increment(ulong *reg, ulong mask, int inc)
579 {
580 	assign_masked(reg, *reg + inc, mask);
581 }
582 
583 static inline void
584 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
585 {
586 	ulong *preg = reg_rmw(ctxt, reg);
587 
588 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
589 }
590 
591 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
592 {
593 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
594 }
595 
596 static u32 desc_limit_scaled(struct desc_struct *desc)
597 {
598 	u32 limit = get_desc_limit(desc);
599 
600 	return desc->g ? (limit << 12) | 0xfff : limit;
601 }
602 
603 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
604 {
605 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
606 		return 0;
607 
608 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
609 }
610 
611 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
612 			     u32 error, bool valid)
613 {
614 	if (KVM_EMULATOR_BUG_ON(vec > 0x1f, ctxt))
615 		return X86EMUL_UNHANDLEABLE;
616 
617 	ctxt->exception.vector = vec;
618 	ctxt->exception.error_code = error;
619 	ctxt->exception.error_code_valid = valid;
620 	return X86EMUL_PROPAGATE_FAULT;
621 }
622 
623 static int emulate_db(struct x86_emulate_ctxt *ctxt)
624 {
625 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
626 }
627 
628 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
629 {
630 	return emulate_exception(ctxt, GP_VECTOR, err, true);
631 }
632 
633 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
634 {
635 	return emulate_exception(ctxt, SS_VECTOR, err, true);
636 }
637 
638 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
639 {
640 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
641 }
642 
643 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
644 {
645 	return emulate_exception(ctxt, TS_VECTOR, err, true);
646 }
647 
648 static int emulate_de(struct x86_emulate_ctxt *ctxt)
649 {
650 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
651 }
652 
653 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
654 {
655 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
656 }
657 
658 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
659 {
660 	u16 selector;
661 	struct desc_struct desc;
662 
663 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
664 	return selector;
665 }
666 
667 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
668 				 unsigned seg)
669 {
670 	u16 dummy;
671 	u32 base3;
672 	struct desc_struct desc;
673 
674 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
675 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
676 }
677 
678 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
679 {
680 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
681 }
682 
683 static inline bool emul_is_noncanonical_address(u64 la,
684 						struct x86_emulate_ctxt *ctxt)
685 {
686 	return !__is_canonical_address(la, ctxt_virt_addr_bits(ctxt));
687 }
688 
689 /*
690  * x86 defines three classes of vector instructions: explicitly
691  * aligned, explicitly unaligned, and the rest, which change behaviour
692  * depending on whether they're AVX encoded or not.
693  *
694  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
695  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
696  * 512 bytes of data must be aligned to a 16 byte boundary.
697  */
698 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
699 {
700 	u64 alignment = ctxt->d & AlignMask;
701 
702 	if (likely(size < 16))
703 		return 1;
704 
705 	switch (alignment) {
706 	case Unaligned:
707 	case Avx:
708 		return 1;
709 	case Aligned16:
710 		return 16;
711 	case Aligned:
712 	default:
713 		return size;
714 	}
715 }
716 
717 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
718 				       struct segmented_address addr,
719 				       unsigned *max_size, unsigned size,
720 				       bool write, bool fetch,
721 				       enum x86emul_mode mode, ulong *linear)
722 {
723 	struct desc_struct desc;
724 	bool usable;
725 	ulong la;
726 	u32 lim;
727 	u16 sel;
728 	u8  va_bits;
729 
730 	la = seg_base(ctxt, addr.seg) + addr.ea;
731 	*max_size = 0;
732 	switch (mode) {
733 	case X86EMUL_MODE_PROT64:
734 		*linear = la;
735 		va_bits = ctxt_virt_addr_bits(ctxt);
736 		if (!__is_canonical_address(la, va_bits))
737 			goto bad;
738 
739 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
740 		if (size > *max_size)
741 			goto bad;
742 		break;
743 	default:
744 		*linear = la = (u32)la;
745 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
746 						addr.seg);
747 		if (!usable)
748 			goto bad;
749 		/* code segment in protected mode or read-only data segment */
750 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8))
751 					|| !(desc.type & 2)) && write)
752 			goto bad;
753 		/* unreadable code segment */
754 		if (!fetch && (desc.type & 8) && !(desc.type & 2))
755 			goto bad;
756 		lim = desc_limit_scaled(&desc);
757 		if (!(desc.type & 8) && (desc.type & 4)) {
758 			/* expand-down segment */
759 			if (addr.ea <= lim)
760 				goto bad;
761 			lim = desc.d ? 0xffffffff : 0xffff;
762 		}
763 		if (addr.ea > lim)
764 			goto bad;
765 		if (lim == 0xffffffff)
766 			*max_size = ~0u;
767 		else {
768 			*max_size = (u64)lim + 1 - addr.ea;
769 			if (size > *max_size)
770 				goto bad;
771 		}
772 		break;
773 	}
774 	if (la & (insn_alignment(ctxt, size) - 1))
775 		return emulate_gp(ctxt, 0);
776 	return X86EMUL_CONTINUE;
777 bad:
778 	if (addr.seg == VCPU_SREG_SS)
779 		return emulate_ss(ctxt, 0);
780 	else
781 		return emulate_gp(ctxt, 0);
782 }
783 
784 static int linearize(struct x86_emulate_ctxt *ctxt,
785 		     struct segmented_address addr,
786 		     unsigned size, bool write,
787 		     ulong *linear)
788 {
789 	unsigned max_size;
790 	return __linearize(ctxt, addr, &max_size, size, write, false,
791 			   ctxt->mode, linear);
792 }
793 
794 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst,
795 			     enum x86emul_mode mode)
796 {
797 	ulong linear;
798 	int rc;
799 	unsigned max_size;
800 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
801 					   .ea = dst };
802 
803 	if (ctxt->op_bytes != sizeof(unsigned long))
804 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
805 	rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear);
806 	if (rc == X86EMUL_CONTINUE)
807 		ctxt->_eip = addr.ea;
808 	return rc;
809 }
810 
811 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
812 {
813 	return assign_eip(ctxt, dst, ctxt->mode);
814 }
815 
816 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst,
817 			  const struct desc_struct *cs_desc)
818 {
819 	enum x86emul_mode mode = ctxt->mode;
820 	int rc;
821 
822 #ifdef CONFIG_X86_64
823 	if (ctxt->mode >= X86EMUL_MODE_PROT16) {
824 		if (cs_desc->l) {
825 			u64 efer = 0;
826 
827 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
828 			if (efer & EFER_LMA)
829 				mode = X86EMUL_MODE_PROT64;
830 		} else
831 			mode = X86EMUL_MODE_PROT32; /* temporary value */
832 	}
833 #endif
834 	if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32)
835 		mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
836 	rc = assign_eip(ctxt, dst, mode);
837 	if (rc == X86EMUL_CONTINUE)
838 		ctxt->mode = mode;
839 	return rc;
840 }
841 
842 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel)
843 {
844 	return assign_eip_near(ctxt, ctxt->_eip + rel);
845 }
846 
847 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear,
848 			      void *data, unsigned size)
849 {
850 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true);
851 }
852 
853 static int linear_write_system(struct x86_emulate_ctxt *ctxt,
854 			       ulong linear, void *data,
855 			       unsigned int size)
856 {
857 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true);
858 }
859 
860 static int segmented_read_std(struct x86_emulate_ctxt *ctxt,
861 			      struct segmented_address addr,
862 			      void *data,
863 			      unsigned size)
864 {
865 	int rc;
866 	ulong linear;
867 
868 	rc = linearize(ctxt, addr, size, false, &linear);
869 	if (rc != X86EMUL_CONTINUE)
870 		return rc;
871 	return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false);
872 }
873 
874 static int segmented_write_std(struct x86_emulate_ctxt *ctxt,
875 			       struct segmented_address addr,
876 			       void *data,
877 			       unsigned int size)
878 {
879 	int rc;
880 	ulong linear;
881 
882 	rc = linearize(ctxt, addr, size, true, &linear);
883 	if (rc != X86EMUL_CONTINUE)
884 		return rc;
885 	return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false);
886 }
887 
888 /*
889  * Prefetch the remaining bytes of the instruction without crossing page
890  * boundary if they are not in fetch_cache yet.
891  */
892 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size)
893 {
894 	int rc;
895 	unsigned size, max_size;
896 	unsigned long linear;
897 	int cur_size = ctxt->fetch.end - ctxt->fetch.data;
898 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
899 					   .ea = ctxt->eip + cur_size };
900 
901 	/*
902 	 * We do not know exactly how many bytes will be needed, and
903 	 * __linearize is expensive, so fetch as much as possible.  We
904 	 * just have to avoid going beyond the 15 byte limit, the end
905 	 * of the segment, or the end of the page.
906 	 *
907 	 * __linearize is called with size 0 so that it does not do any
908 	 * boundary check itself.  Instead, we use max_size to check
909 	 * against op_size.
910 	 */
911 	rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode,
912 			 &linear);
913 	if (unlikely(rc != X86EMUL_CONTINUE))
914 		return rc;
915 
916 	size = min_t(unsigned, 15UL ^ cur_size, max_size);
917 	size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear));
918 
919 	/*
920 	 * One instruction can only straddle two pages,
921 	 * and one has been loaded at the beginning of
922 	 * x86_decode_insn.  So, if not enough bytes
923 	 * still, we must have hit the 15-byte boundary.
924 	 */
925 	if (unlikely(size < op_size))
926 		return emulate_gp(ctxt, 0);
927 
928 	rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end,
929 			      size, &ctxt->exception);
930 	if (unlikely(rc != X86EMUL_CONTINUE))
931 		return rc;
932 	ctxt->fetch.end += size;
933 	return X86EMUL_CONTINUE;
934 }
935 
936 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt,
937 					       unsigned size)
938 {
939 	unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr;
940 
941 	if (unlikely(done_size < size))
942 		return __do_insn_fetch_bytes(ctxt, size - done_size);
943 	else
944 		return X86EMUL_CONTINUE;
945 }
946 
947 /* Fetch next part of the instruction being emulated. */
948 #define insn_fetch(_type, _ctxt)					\
949 ({	_type _x;							\
950 									\
951 	rc = do_insn_fetch_bytes(_ctxt, sizeof(_type));			\
952 	if (rc != X86EMUL_CONTINUE)					\
953 		goto done;						\
954 	ctxt->_eip += sizeof(_type);					\
955 	memcpy(&_x, ctxt->fetch.ptr, sizeof(_type));			\
956 	ctxt->fetch.ptr += sizeof(_type);				\
957 	_x;								\
958 })
959 
960 #define insn_fetch_arr(_arr, _size, _ctxt)				\
961 ({									\
962 	rc = do_insn_fetch_bytes(_ctxt, _size);				\
963 	if (rc != X86EMUL_CONTINUE)					\
964 		goto done;						\
965 	ctxt->_eip += (_size);						\
966 	memcpy(_arr, ctxt->fetch.ptr, _size);				\
967 	ctxt->fetch.ptr += (_size);					\
968 })
969 
970 /*
971  * Given the 'reg' portion of a ModRM byte, and a register block, return a
972  * pointer into the block that addresses the relevant register.
973  * @highbyte_regs specifies whether to decode AH,CH,DH,BH.
974  */
975 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg,
976 			     int byteop)
977 {
978 	void *p;
979 	int highbyte_regs = (ctxt->rex_prefix == 0) && byteop;
980 
981 	if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8)
982 		p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1;
983 	else
984 		p = reg_rmw(ctxt, modrm_reg);
985 	return p;
986 }
987 
988 static int read_descriptor(struct x86_emulate_ctxt *ctxt,
989 			   struct segmented_address addr,
990 			   u16 *size, unsigned long *address, int op_bytes)
991 {
992 	int rc;
993 
994 	if (op_bytes == 2)
995 		op_bytes = 3;
996 	*address = 0;
997 	rc = segmented_read_std(ctxt, addr, size, 2);
998 	if (rc != X86EMUL_CONTINUE)
999 		return rc;
1000 	addr.ea += 2;
1001 	rc = segmented_read_std(ctxt, addr, address, op_bytes);
1002 	return rc;
1003 }
1004 
1005 FASTOP2(add);
1006 FASTOP2(or);
1007 FASTOP2(adc);
1008 FASTOP2(sbb);
1009 FASTOP2(and);
1010 FASTOP2(sub);
1011 FASTOP2(xor);
1012 FASTOP2(cmp);
1013 FASTOP2(test);
1014 
1015 FASTOP1SRC2(mul, mul_ex);
1016 FASTOP1SRC2(imul, imul_ex);
1017 FASTOP1SRC2EX(div, div_ex);
1018 FASTOP1SRC2EX(idiv, idiv_ex);
1019 
1020 FASTOP3WCL(shld);
1021 FASTOP3WCL(shrd);
1022 
1023 FASTOP2W(imul);
1024 
1025 FASTOP1(not);
1026 FASTOP1(neg);
1027 FASTOP1(inc);
1028 FASTOP1(dec);
1029 
1030 FASTOP2CL(rol);
1031 FASTOP2CL(ror);
1032 FASTOP2CL(rcl);
1033 FASTOP2CL(rcr);
1034 FASTOP2CL(shl);
1035 FASTOP2CL(shr);
1036 FASTOP2CL(sar);
1037 
1038 FASTOP2W(bsf);
1039 FASTOP2W(bsr);
1040 FASTOP2W(bt);
1041 FASTOP2W(bts);
1042 FASTOP2W(btr);
1043 FASTOP2W(btc);
1044 
1045 FASTOP2(xadd);
1046 
1047 FASTOP2R(cmp, cmp_r);
1048 
1049 static int em_bsf_c(struct x86_emulate_ctxt *ctxt)
1050 {
1051 	/* If src is zero, do not writeback, but update flags */
1052 	if (ctxt->src.val == 0)
1053 		ctxt->dst.type = OP_NONE;
1054 	return fastop(ctxt, em_bsf);
1055 }
1056 
1057 static int em_bsr_c(struct x86_emulate_ctxt *ctxt)
1058 {
1059 	/* If src is zero, do not writeback, but update flags */
1060 	if (ctxt->src.val == 0)
1061 		ctxt->dst.type = OP_NONE;
1062 	return fastop(ctxt, em_bsr);
1063 }
1064 
1065 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
1066 {
1067 	u8 rc;
1068 	void (*fop)(void) = (void *)em_setcc + FASTOP_SIZE * (condition & 0xf);
1069 
1070 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
1071 	asm("push %[flags]; popf; " CALL_NOSPEC
1072 	    : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags));
1073 	return rc;
1074 }
1075 
1076 static void fetch_register_operand(struct operand *op)
1077 {
1078 	switch (op->bytes) {
1079 	case 1:
1080 		op->val = *(u8 *)op->addr.reg;
1081 		break;
1082 	case 2:
1083 		op->val = *(u16 *)op->addr.reg;
1084 		break;
1085 	case 4:
1086 		op->val = *(u32 *)op->addr.reg;
1087 		break;
1088 	case 8:
1089 		op->val = *(u64 *)op->addr.reg;
1090 		break;
1091 	}
1092 }
1093 
1094 static int em_fninit(struct x86_emulate_ctxt *ctxt)
1095 {
1096 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1097 		return emulate_nm(ctxt);
1098 
1099 	kvm_fpu_get();
1100 	asm volatile("fninit");
1101 	kvm_fpu_put();
1102 	return X86EMUL_CONTINUE;
1103 }
1104 
1105 static int em_fnstcw(struct x86_emulate_ctxt *ctxt)
1106 {
1107 	u16 fcw;
1108 
1109 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1110 		return emulate_nm(ctxt);
1111 
1112 	kvm_fpu_get();
1113 	asm volatile("fnstcw %0": "+m"(fcw));
1114 	kvm_fpu_put();
1115 
1116 	ctxt->dst.val = fcw;
1117 
1118 	return X86EMUL_CONTINUE;
1119 }
1120 
1121 static int em_fnstsw(struct x86_emulate_ctxt *ctxt)
1122 {
1123 	u16 fsw;
1124 
1125 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
1126 		return emulate_nm(ctxt);
1127 
1128 	kvm_fpu_get();
1129 	asm volatile("fnstsw %0": "+m"(fsw));
1130 	kvm_fpu_put();
1131 
1132 	ctxt->dst.val = fsw;
1133 
1134 	return X86EMUL_CONTINUE;
1135 }
1136 
1137 static void decode_register_operand(struct x86_emulate_ctxt *ctxt,
1138 				    struct operand *op)
1139 {
1140 	unsigned reg = ctxt->modrm_reg;
1141 
1142 	if (!(ctxt->d & ModRM))
1143 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1144 
1145 	if (ctxt->d & Sse) {
1146 		op->type = OP_XMM;
1147 		op->bytes = 16;
1148 		op->addr.xmm = reg;
1149 		kvm_read_sse_reg(reg, &op->vec_val);
1150 		return;
1151 	}
1152 	if (ctxt->d & Mmx) {
1153 		reg &= 7;
1154 		op->type = OP_MM;
1155 		op->bytes = 8;
1156 		op->addr.mm = reg;
1157 		return;
1158 	}
1159 
1160 	op->type = OP_REG;
1161 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1162 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1163 
1164 	fetch_register_operand(op);
1165 	op->orig_val = op->val;
1166 }
1167 
1168 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1169 {
1170 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1171 		ctxt->modrm_seg = VCPU_SREG_SS;
1172 }
1173 
1174 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1175 			struct operand *op)
1176 {
1177 	u8 sib;
1178 	int index_reg, base_reg, scale;
1179 	int rc = X86EMUL_CONTINUE;
1180 	ulong modrm_ea = 0;
1181 
1182 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1183 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1184 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1185 
1186 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1187 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1188 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1189 	ctxt->modrm_seg = VCPU_SREG_DS;
1190 
1191 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1192 		op->type = OP_REG;
1193 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1194 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1195 				ctxt->d & ByteOp);
1196 		if (ctxt->d & Sse) {
1197 			op->type = OP_XMM;
1198 			op->bytes = 16;
1199 			op->addr.xmm = ctxt->modrm_rm;
1200 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1201 			return rc;
1202 		}
1203 		if (ctxt->d & Mmx) {
1204 			op->type = OP_MM;
1205 			op->bytes = 8;
1206 			op->addr.mm = ctxt->modrm_rm & 7;
1207 			return rc;
1208 		}
1209 		fetch_register_operand(op);
1210 		return rc;
1211 	}
1212 
1213 	op->type = OP_MEM;
1214 
1215 	if (ctxt->ad_bytes == 2) {
1216 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1217 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1218 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1219 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1220 
1221 		/* 16-bit ModR/M decode. */
1222 		switch (ctxt->modrm_mod) {
1223 		case 0:
1224 			if (ctxt->modrm_rm == 6)
1225 				modrm_ea += insn_fetch(u16, ctxt);
1226 			break;
1227 		case 1:
1228 			modrm_ea += insn_fetch(s8, ctxt);
1229 			break;
1230 		case 2:
1231 			modrm_ea += insn_fetch(u16, ctxt);
1232 			break;
1233 		}
1234 		switch (ctxt->modrm_rm) {
1235 		case 0:
1236 			modrm_ea += bx + si;
1237 			break;
1238 		case 1:
1239 			modrm_ea += bx + di;
1240 			break;
1241 		case 2:
1242 			modrm_ea += bp + si;
1243 			break;
1244 		case 3:
1245 			modrm_ea += bp + di;
1246 			break;
1247 		case 4:
1248 			modrm_ea += si;
1249 			break;
1250 		case 5:
1251 			modrm_ea += di;
1252 			break;
1253 		case 6:
1254 			if (ctxt->modrm_mod != 0)
1255 				modrm_ea += bp;
1256 			break;
1257 		case 7:
1258 			modrm_ea += bx;
1259 			break;
1260 		}
1261 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1262 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1263 			ctxt->modrm_seg = VCPU_SREG_SS;
1264 		modrm_ea = (u16)modrm_ea;
1265 	} else {
1266 		/* 32/64-bit ModR/M decode. */
1267 		if ((ctxt->modrm_rm & 7) == 4) {
1268 			sib = insn_fetch(u8, ctxt);
1269 			index_reg |= (sib >> 3) & 7;
1270 			base_reg |= sib & 7;
1271 			scale = sib >> 6;
1272 
1273 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1274 				modrm_ea += insn_fetch(s32, ctxt);
1275 			else {
1276 				modrm_ea += reg_read(ctxt, base_reg);
1277 				adjust_modrm_seg(ctxt, base_reg);
1278 				/* Increment ESP on POP [ESP] */
1279 				if ((ctxt->d & IncSP) &&
1280 				    base_reg == VCPU_REGS_RSP)
1281 					modrm_ea += ctxt->op_bytes;
1282 			}
1283 			if (index_reg != 4)
1284 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1285 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1286 			modrm_ea += insn_fetch(s32, ctxt);
1287 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1288 				ctxt->rip_relative = 1;
1289 		} else {
1290 			base_reg = ctxt->modrm_rm;
1291 			modrm_ea += reg_read(ctxt, base_reg);
1292 			adjust_modrm_seg(ctxt, base_reg);
1293 		}
1294 		switch (ctxt->modrm_mod) {
1295 		case 1:
1296 			modrm_ea += insn_fetch(s8, ctxt);
1297 			break;
1298 		case 2:
1299 			modrm_ea += insn_fetch(s32, ctxt);
1300 			break;
1301 		}
1302 	}
1303 	op->addr.mem.ea = modrm_ea;
1304 	if (ctxt->ad_bytes != 8)
1305 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1306 
1307 done:
1308 	return rc;
1309 }
1310 
1311 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1312 		      struct operand *op)
1313 {
1314 	int rc = X86EMUL_CONTINUE;
1315 
1316 	op->type = OP_MEM;
1317 	switch (ctxt->ad_bytes) {
1318 	case 2:
1319 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1320 		break;
1321 	case 4:
1322 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1323 		break;
1324 	case 8:
1325 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1326 		break;
1327 	}
1328 done:
1329 	return rc;
1330 }
1331 
1332 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1333 {
1334 	long sv = 0, mask;
1335 
1336 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1337 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1338 
1339 		if (ctxt->src.bytes == 2)
1340 			sv = (s16)ctxt->src.val & (s16)mask;
1341 		else if (ctxt->src.bytes == 4)
1342 			sv = (s32)ctxt->src.val & (s32)mask;
1343 		else
1344 			sv = (s64)ctxt->src.val & (s64)mask;
1345 
1346 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1347 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1348 	}
1349 
1350 	/* only subword offset */
1351 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1352 }
1353 
1354 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1355 			 unsigned long addr, void *dest, unsigned size)
1356 {
1357 	int rc;
1358 	struct read_cache *mc = &ctxt->mem_read;
1359 
1360 	if (mc->pos < mc->end)
1361 		goto read_cached;
1362 
1363 	if (KVM_EMULATOR_BUG_ON((mc->end + size) >= sizeof(mc->data), ctxt))
1364 		return X86EMUL_UNHANDLEABLE;
1365 
1366 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1367 				      &ctxt->exception);
1368 	if (rc != X86EMUL_CONTINUE)
1369 		return rc;
1370 
1371 	mc->end += size;
1372 
1373 read_cached:
1374 	memcpy(dest, mc->data + mc->pos, size);
1375 	mc->pos += size;
1376 	return X86EMUL_CONTINUE;
1377 }
1378 
1379 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1380 			  struct segmented_address addr,
1381 			  void *data,
1382 			  unsigned size)
1383 {
1384 	int rc;
1385 	ulong linear;
1386 
1387 	rc = linearize(ctxt, addr, size, false, &linear);
1388 	if (rc != X86EMUL_CONTINUE)
1389 		return rc;
1390 	return read_emulated(ctxt, linear, data, size);
1391 }
1392 
1393 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1394 			   struct segmented_address addr,
1395 			   const void *data,
1396 			   unsigned size)
1397 {
1398 	int rc;
1399 	ulong linear;
1400 
1401 	rc = linearize(ctxt, addr, size, true, &linear);
1402 	if (rc != X86EMUL_CONTINUE)
1403 		return rc;
1404 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1405 					 &ctxt->exception);
1406 }
1407 
1408 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1409 			     struct segmented_address addr,
1410 			     const void *orig_data, const void *data,
1411 			     unsigned size)
1412 {
1413 	int rc;
1414 	ulong linear;
1415 
1416 	rc = linearize(ctxt, addr, size, true, &linear);
1417 	if (rc != X86EMUL_CONTINUE)
1418 		return rc;
1419 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1420 					   size, &ctxt->exception);
1421 }
1422 
1423 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1424 			   unsigned int size, unsigned short port,
1425 			   void *dest)
1426 {
1427 	struct read_cache *rc = &ctxt->io_read;
1428 
1429 	if (rc->pos == rc->end) { /* refill pio read ahead */
1430 		unsigned int in_page, n;
1431 		unsigned int count = ctxt->rep_prefix ?
1432 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1433 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1434 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1435 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1436 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1437 		if (n == 0)
1438 			n = 1;
1439 		rc->pos = rc->end = 0;
1440 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1441 			return 0;
1442 		rc->end = n * size;
1443 	}
1444 
1445 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1446 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1447 		ctxt->dst.data = rc->data + rc->pos;
1448 		ctxt->dst.type = OP_MEM_STR;
1449 		ctxt->dst.count = (rc->end - rc->pos) / size;
1450 		rc->pos = rc->end;
1451 	} else {
1452 		memcpy(dest, rc->data + rc->pos, size);
1453 		rc->pos += size;
1454 	}
1455 	return 1;
1456 }
1457 
1458 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1459 				     u16 index, struct desc_struct *desc)
1460 {
1461 	struct desc_ptr dt;
1462 	ulong addr;
1463 
1464 	ctxt->ops->get_idt(ctxt, &dt);
1465 
1466 	if (dt.size < index * 8 + 7)
1467 		return emulate_gp(ctxt, index << 3 | 0x2);
1468 
1469 	addr = dt.address + index * 8;
1470 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1471 }
1472 
1473 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1474 				     u16 selector, struct desc_ptr *dt)
1475 {
1476 	const struct x86_emulate_ops *ops = ctxt->ops;
1477 	u32 base3 = 0;
1478 
1479 	if (selector & 1 << 2) {
1480 		struct desc_struct desc;
1481 		u16 sel;
1482 
1483 		memset(dt, 0, sizeof(*dt));
1484 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1485 				      VCPU_SREG_LDTR))
1486 			return;
1487 
1488 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1489 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1490 	} else
1491 		ops->get_gdt(ctxt, dt);
1492 }
1493 
1494 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1495 			      u16 selector, ulong *desc_addr_p)
1496 {
1497 	struct desc_ptr dt;
1498 	u16 index = selector >> 3;
1499 	ulong addr;
1500 
1501 	get_descriptor_table_ptr(ctxt, selector, &dt);
1502 
1503 	if (dt.size < index * 8 + 7)
1504 		return emulate_gp(ctxt, selector & 0xfffc);
1505 
1506 	addr = dt.address + index * 8;
1507 
1508 #ifdef CONFIG_X86_64
1509 	if (addr >> 32 != 0) {
1510 		u64 efer = 0;
1511 
1512 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1513 		if (!(efer & EFER_LMA))
1514 			addr &= (u32)-1;
1515 	}
1516 #endif
1517 
1518 	*desc_addr_p = addr;
1519 	return X86EMUL_CONTINUE;
1520 }
1521 
1522 /* allowed just for 8 bytes segments */
1523 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1524 				   u16 selector, struct desc_struct *desc,
1525 				   ulong *desc_addr_p)
1526 {
1527 	int rc;
1528 
1529 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1530 	if (rc != X86EMUL_CONTINUE)
1531 		return rc;
1532 
1533 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1534 }
1535 
1536 /* allowed just for 8 bytes segments */
1537 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1538 				    u16 selector, struct desc_struct *desc)
1539 {
1540 	int rc;
1541 	ulong addr;
1542 
1543 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1544 	if (rc != X86EMUL_CONTINUE)
1545 		return rc;
1546 
1547 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1548 }
1549 
1550 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1551 				     u16 selector, int seg, u8 cpl,
1552 				     enum x86_transfer_type transfer,
1553 				     struct desc_struct *desc)
1554 {
1555 	struct desc_struct seg_desc, old_desc;
1556 	u8 dpl, rpl;
1557 	unsigned err_vec = GP_VECTOR;
1558 	u32 err_code = 0;
1559 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1560 	ulong desc_addr;
1561 	int ret;
1562 	u16 dummy;
1563 	u32 base3 = 0;
1564 
1565 	memset(&seg_desc, 0, sizeof(seg_desc));
1566 
1567 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1568 		/* set real mode segment descriptor (keep limit etc. for
1569 		 * unreal mode) */
1570 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1571 		set_desc_base(&seg_desc, selector << 4);
1572 		goto load;
1573 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1574 		/* VM86 needs a clean new segment descriptor */
1575 		set_desc_base(&seg_desc, selector << 4);
1576 		set_desc_limit(&seg_desc, 0xffff);
1577 		seg_desc.type = 3;
1578 		seg_desc.p = 1;
1579 		seg_desc.s = 1;
1580 		seg_desc.dpl = 3;
1581 		goto load;
1582 	}
1583 
1584 	rpl = selector & 3;
1585 
1586 	/* TR should be in GDT only */
1587 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1588 		goto exception;
1589 
1590 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1591 	if (null_selector) {
1592 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1593 			goto exception;
1594 
1595 		if (seg == VCPU_SREG_SS) {
1596 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1597 				goto exception;
1598 
1599 			/*
1600 			 * ctxt->ops->set_segment expects the CPL to be in
1601 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1602 			 */
1603 			seg_desc.type = 3;
1604 			seg_desc.p = 1;
1605 			seg_desc.s = 1;
1606 			seg_desc.dpl = cpl;
1607 			seg_desc.d = 1;
1608 			seg_desc.g = 1;
1609 		}
1610 
1611 		/* Skip all following checks */
1612 		goto load;
1613 	}
1614 
1615 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1616 	if (ret != X86EMUL_CONTINUE)
1617 		return ret;
1618 
1619 	err_code = selector & 0xfffc;
1620 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1621 							   GP_VECTOR;
1622 
1623 	/* can't load system descriptor into segment selector */
1624 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1625 		if (transfer == X86_TRANSFER_CALL_JMP)
1626 			return X86EMUL_UNHANDLEABLE;
1627 		goto exception;
1628 	}
1629 
1630 	dpl = seg_desc.dpl;
1631 
1632 	switch (seg) {
1633 	case VCPU_SREG_SS:
1634 		/*
1635 		 * segment is not a writable data segment or segment
1636 		 * selector's RPL != CPL or segment selector's RPL != CPL
1637 		 */
1638 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1639 			goto exception;
1640 		break;
1641 	case VCPU_SREG_CS:
1642 		if (!(seg_desc.type & 8))
1643 			goto exception;
1644 
1645 		if (transfer == X86_TRANSFER_RET) {
1646 			/* RET can never return to an inner privilege level. */
1647 			if (rpl < cpl)
1648 				goto exception;
1649 			/* Outer-privilege level return is not implemented */
1650 			if (rpl > cpl)
1651 				return X86EMUL_UNHANDLEABLE;
1652 		}
1653 		if (transfer == X86_TRANSFER_RET || transfer == X86_TRANSFER_TASK_SWITCH) {
1654 			if (seg_desc.type & 4) {
1655 				/* conforming */
1656 				if (dpl > rpl)
1657 					goto exception;
1658 			} else {
1659 				/* nonconforming */
1660 				if (dpl != rpl)
1661 					goto exception;
1662 			}
1663 		} else { /* X86_TRANSFER_CALL_JMP */
1664 			if (seg_desc.type & 4) {
1665 				/* conforming */
1666 				if (dpl > cpl)
1667 					goto exception;
1668 			} else {
1669 				/* nonconforming */
1670 				if (rpl > cpl || dpl != cpl)
1671 					goto exception;
1672 			}
1673 		}
1674 		/* in long-mode d/b must be clear if l is set */
1675 		if (seg_desc.d && seg_desc.l) {
1676 			u64 efer = 0;
1677 
1678 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1679 			if (efer & EFER_LMA)
1680 				goto exception;
1681 		}
1682 
1683 		/* CS(RPL) <- CPL */
1684 		selector = (selector & 0xfffc) | cpl;
1685 		break;
1686 	case VCPU_SREG_TR:
1687 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1688 			goto exception;
1689 		break;
1690 	case VCPU_SREG_LDTR:
1691 		if (seg_desc.s || seg_desc.type != 2)
1692 			goto exception;
1693 		break;
1694 	default: /*  DS, ES, FS, or GS */
1695 		/*
1696 		 * segment is not a data or readable code segment or
1697 		 * ((segment is a data or nonconforming code segment)
1698 		 * and (both RPL and CPL > DPL))
1699 		 */
1700 		if ((seg_desc.type & 0xa) == 0x8 ||
1701 		    (((seg_desc.type & 0xc) != 0xc) &&
1702 		     (rpl > dpl && cpl > dpl)))
1703 			goto exception;
1704 		break;
1705 	}
1706 
1707 	if (!seg_desc.p) {
1708 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1709 		goto exception;
1710 	}
1711 
1712 	if (seg_desc.s) {
1713 		/* mark segment as accessed */
1714 		if (!(seg_desc.type & 1)) {
1715 			seg_desc.type |= 1;
1716 			ret = write_segment_descriptor(ctxt, selector,
1717 						       &seg_desc);
1718 			if (ret != X86EMUL_CONTINUE)
1719 				return ret;
1720 		}
1721 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1722 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1723 		if (ret != X86EMUL_CONTINUE)
1724 			return ret;
1725 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1726 						 ((u64)base3 << 32), ctxt))
1727 			return emulate_gp(ctxt, err_code);
1728 	}
1729 
1730 	if (seg == VCPU_SREG_TR) {
1731 		old_desc = seg_desc;
1732 		seg_desc.type |= 2; /* busy */
1733 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1734 						  sizeof(seg_desc), &ctxt->exception);
1735 		if (ret != X86EMUL_CONTINUE)
1736 			return ret;
1737 	}
1738 load:
1739 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1740 	if (desc)
1741 		*desc = seg_desc;
1742 	return X86EMUL_CONTINUE;
1743 exception:
1744 	return emulate_exception(ctxt, err_vec, err_code, true);
1745 }
1746 
1747 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1748 				   u16 selector, int seg)
1749 {
1750 	u8 cpl = ctxt->ops->cpl(ctxt);
1751 
1752 	/*
1753 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1754 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1755 	 * but it's wrong).
1756 	 *
1757 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1758 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1759 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1760 	 * and only forbid it here.
1761 	 */
1762 	if (seg == VCPU_SREG_SS && selector == 3 &&
1763 	    ctxt->mode == X86EMUL_MODE_PROT64)
1764 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1765 
1766 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1767 					 X86_TRANSFER_NONE, NULL);
1768 }
1769 
1770 static void write_register_operand(struct operand *op)
1771 {
1772 	return assign_register(op->addr.reg, op->val, op->bytes);
1773 }
1774 
1775 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1776 {
1777 	switch (op->type) {
1778 	case OP_REG:
1779 		write_register_operand(op);
1780 		break;
1781 	case OP_MEM:
1782 		if (ctxt->lock_prefix)
1783 			return segmented_cmpxchg(ctxt,
1784 						 op->addr.mem,
1785 						 &op->orig_val,
1786 						 &op->val,
1787 						 op->bytes);
1788 		else
1789 			return segmented_write(ctxt,
1790 					       op->addr.mem,
1791 					       &op->val,
1792 					       op->bytes);
1793 		break;
1794 	case OP_MEM_STR:
1795 		return segmented_write(ctxt,
1796 				       op->addr.mem,
1797 				       op->data,
1798 				       op->bytes * op->count);
1799 		break;
1800 	case OP_XMM:
1801 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1802 		break;
1803 	case OP_MM:
1804 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1805 		break;
1806 	case OP_NONE:
1807 		/* no writeback */
1808 		break;
1809 	default:
1810 		break;
1811 	}
1812 	return X86EMUL_CONTINUE;
1813 }
1814 
1815 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes)
1816 {
1817 	struct segmented_address addr;
1818 
1819 	rsp_increment(ctxt, -bytes);
1820 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1821 	addr.seg = VCPU_SREG_SS;
1822 
1823 	return segmented_write(ctxt, addr, data, bytes);
1824 }
1825 
1826 static int em_push(struct x86_emulate_ctxt *ctxt)
1827 {
1828 	/* Disable writeback. */
1829 	ctxt->dst.type = OP_NONE;
1830 	return push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1831 }
1832 
1833 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1834 		       void *dest, int len)
1835 {
1836 	int rc;
1837 	struct segmented_address addr;
1838 
1839 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1840 	addr.seg = VCPU_SREG_SS;
1841 	rc = segmented_read(ctxt, addr, dest, len);
1842 	if (rc != X86EMUL_CONTINUE)
1843 		return rc;
1844 
1845 	rsp_increment(ctxt, len);
1846 	return rc;
1847 }
1848 
1849 static int em_pop(struct x86_emulate_ctxt *ctxt)
1850 {
1851 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1852 }
1853 
1854 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1855 			void *dest, int len)
1856 {
1857 	int rc;
1858 	unsigned long val, change_mask;
1859 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1860 	int cpl = ctxt->ops->cpl(ctxt);
1861 
1862 	rc = emulate_pop(ctxt, &val, len);
1863 	if (rc != X86EMUL_CONTINUE)
1864 		return rc;
1865 
1866 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1867 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1868 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1869 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1870 
1871 	switch(ctxt->mode) {
1872 	case X86EMUL_MODE_PROT64:
1873 	case X86EMUL_MODE_PROT32:
1874 	case X86EMUL_MODE_PROT16:
1875 		if (cpl == 0)
1876 			change_mask |= X86_EFLAGS_IOPL;
1877 		if (cpl <= iopl)
1878 			change_mask |= X86_EFLAGS_IF;
1879 		break;
1880 	case X86EMUL_MODE_VM86:
1881 		if (iopl < 3)
1882 			return emulate_gp(ctxt, 0);
1883 		change_mask |= X86_EFLAGS_IF;
1884 		break;
1885 	default: /* real mode */
1886 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1887 		break;
1888 	}
1889 
1890 	*(unsigned long *)dest =
1891 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1892 
1893 	return rc;
1894 }
1895 
1896 static int em_popf(struct x86_emulate_ctxt *ctxt)
1897 {
1898 	ctxt->dst.type = OP_REG;
1899 	ctxt->dst.addr.reg = &ctxt->eflags;
1900 	ctxt->dst.bytes = ctxt->op_bytes;
1901 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1902 }
1903 
1904 static int em_enter(struct x86_emulate_ctxt *ctxt)
1905 {
1906 	int rc;
1907 	unsigned frame_size = ctxt->src.val;
1908 	unsigned nesting_level = ctxt->src2.val & 31;
1909 	ulong rbp;
1910 
1911 	if (nesting_level)
1912 		return X86EMUL_UNHANDLEABLE;
1913 
1914 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1915 	rc = push(ctxt, &rbp, stack_size(ctxt));
1916 	if (rc != X86EMUL_CONTINUE)
1917 		return rc;
1918 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1919 		      stack_mask(ctxt));
1920 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1921 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1922 		      stack_mask(ctxt));
1923 	return X86EMUL_CONTINUE;
1924 }
1925 
1926 static int em_leave(struct x86_emulate_ctxt *ctxt)
1927 {
1928 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1929 		      stack_mask(ctxt));
1930 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1931 }
1932 
1933 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1934 {
1935 	int seg = ctxt->src2.val;
1936 
1937 	ctxt->src.val = get_segment_selector(ctxt, seg);
1938 	if (ctxt->op_bytes == 4) {
1939 		rsp_increment(ctxt, -2);
1940 		ctxt->op_bytes = 2;
1941 	}
1942 
1943 	return em_push(ctxt);
1944 }
1945 
1946 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1947 {
1948 	int seg = ctxt->src2.val;
1949 	unsigned long selector;
1950 	int rc;
1951 
1952 	rc = emulate_pop(ctxt, &selector, 2);
1953 	if (rc != X86EMUL_CONTINUE)
1954 		return rc;
1955 
1956 	if (ctxt->modrm_reg == VCPU_SREG_SS)
1957 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1958 	if (ctxt->op_bytes > 2)
1959 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1960 
1961 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1962 	return rc;
1963 }
1964 
1965 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1966 {
1967 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1968 	int rc = X86EMUL_CONTINUE;
1969 	int reg = VCPU_REGS_RAX;
1970 
1971 	while (reg <= VCPU_REGS_RDI) {
1972 		(reg == VCPU_REGS_RSP) ?
1973 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1974 
1975 		rc = em_push(ctxt);
1976 		if (rc != X86EMUL_CONTINUE)
1977 			return rc;
1978 
1979 		++reg;
1980 	}
1981 
1982 	return rc;
1983 }
1984 
1985 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1986 {
1987 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1988 	return em_push(ctxt);
1989 }
1990 
1991 static int em_popa(struct x86_emulate_ctxt *ctxt)
1992 {
1993 	int rc = X86EMUL_CONTINUE;
1994 	int reg = VCPU_REGS_RDI;
1995 	u32 val;
1996 
1997 	while (reg >= VCPU_REGS_RAX) {
1998 		if (reg == VCPU_REGS_RSP) {
1999 			rsp_increment(ctxt, ctxt->op_bytes);
2000 			--reg;
2001 		}
2002 
2003 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2004 		if (rc != X86EMUL_CONTINUE)
2005 			break;
2006 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2007 		--reg;
2008 	}
2009 	return rc;
2010 }
2011 
2012 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2013 {
2014 	const struct x86_emulate_ops *ops = ctxt->ops;
2015 	int rc;
2016 	struct desc_ptr dt;
2017 	gva_t cs_addr;
2018 	gva_t eip_addr;
2019 	u16 cs, eip;
2020 
2021 	/* TODO: Add limit checks */
2022 	ctxt->src.val = ctxt->eflags;
2023 	rc = em_push(ctxt);
2024 	if (rc != X86EMUL_CONTINUE)
2025 		return rc;
2026 
2027 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2028 
2029 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2030 	rc = em_push(ctxt);
2031 	if (rc != X86EMUL_CONTINUE)
2032 		return rc;
2033 
2034 	ctxt->src.val = ctxt->_eip;
2035 	rc = em_push(ctxt);
2036 	if (rc != X86EMUL_CONTINUE)
2037 		return rc;
2038 
2039 	ops->get_idt(ctxt, &dt);
2040 
2041 	eip_addr = dt.address + (irq << 2);
2042 	cs_addr = dt.address + (irq << 2) + 2;
2043 
2044 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2045 	if (rc != X86EMUL_CONTINUE)
2046 		return rc;
2047 
2048 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2049 	if (rc != X86EMUL_CONTINUE)
2050 		return rc;
2051 
2052 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2053 	if (rc != X86EMUL_CONTINUE)
2054 		return rc;
2055 
2056 	ctxt->_eip = eip;
2057 
2058 	return rc;
2059 }
2060 
2061 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2062 {
2063 	int rc;
2064 
2065 	invalidate_registers(ctxt);
2066 	rc = __emulate_int_real(ctxt, irq);
2067 	if (rc == X86EMUL_CONTINUE)
2068 		writeback_registers(ctxt);
2069 	return rc;
2070 }
2071 
2072 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2073 {
2074 	switch(ctxt->mode) {
2075 	case X86EMUL_MODE_REAL:
2076 		return __emulate_int_real(ctxt, irq);
2077 	case X86EMUL_MODE_VM86:
2078 	case X86EMUL_MODE_PROT16:
2079 	case X86EMUL_MODE_PROT32:
2080 	case X86EMUL_MODE_PROT64:
2081 	default:
2082 		/* Protected mode interrupts unimplemented yet */
2083 		return X86EMUL_UNHANDLEABLE;
2084 	}
2085 }
2086 
2087 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2088 {
2089 	int rc = X86EMUL_CONTINUE;
2090 	unsigned long temp_eip = 0;
2091 	unsigned long temp_eflags = 0;
2092 	unsigned long cs = 0;
2093 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2094 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2095 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2096 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2097 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2098 			     X86_EFLAGS_FIXED;
2099 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2100 				  X86_EFLAGS_VIP;
2101 
2102 	/* TODO: Add stack limit check */
2103 
2104 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2105 
2106 	if (rc != X86EMUL_CONTINUE)
2107 		return rc;
2108 
2109 	if (temp_eip & ~0xffff)
2110 		return emulate_gp(ctxt, 0);
2111 
2112 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2113 
2114 	if (rc != X86EMUL_CONTINUE)
2115 		return rc;
2116 
2117 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2118 
2119 	if (rc != X86EMUL_CONTINUE)
2120 		return rc;
2121 
2122 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2123 
2124 	if (rc != X86EMUL_CONTINUE)
2125 		return rc;
2126 
2127 	ctxt->_eip = temp_eip;
2128 
2129 	if (ctxt->op_bytes == 4)
2130 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2131 	else if (ctxt->op_bytes == 2) {
2132 		ctxt->eflags &= ~0xffff;
2133 		ctxt->eflags |= temp_eflags;
2134 	}
2135 
2136 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2137 	ctxt->eflags |= X86_EFLAGS_FIXED;
2138 	ctxt->ops->set_nmi_mask(ctxt, false);
2139 
2140 	return rc;
2141 }
2142 
2143 static int em_iret(struct x86_emulate_ctxt *ctxt)
2144 {
2145 	switch(ctxt->mode) {
2146 	case X86EMUL_MODE_REAL:
2147 		return emulate_iret_real(ctxt);
2148 	case X86EMUL_MODE_VM86:
2149 	case X86EMUL_MODE_PROT16:
2150 	case X86EMUL_MODE_PROT32:
2151 	case X86EMUL_MODE_PROT64:
2152 	default:
2153 		/* iret from protected mode unimplemented yet */
2154 		return X86EMUL_UNHANDLEABLE;
2155 	}
2156 }
2157 
2158 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2159 {
2160 	int rc;
2161 	unsigned short sel;
2162 	struct desc_struct new_desc;
2163 	u8 cpl = ctxt->ops->cpl(ctxt);
2164 
2165 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2166 
2167 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2168 				       X86_TRANSFER_CALL_JMP,
2169 				       &new_desc);
2170 	if (rc != X86EMUL_CONTINUE)
2171 		return rc;
2172 
2173 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
2174 	/* Error handling is not implemented. */
2175 	if (rc != X86EMUL_CONTINUE)
2176 		return X86EMUL_UNHANDLEABLE;
2177 
2178 	return rc;
2179 }
2180 
2181 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2182 {
2183 	return assign_eip_near(ctxt, ctxt->src.val);
2184 }
2185 
2186 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2187 {
2188 	int rc;
2189 	long int old_eip;
2190 
2191 	old_eip = ctxt->_eip;
2192 	rc = assign_eip_near(ctxt, ctxt->src.val);
2193 	if (rc != X86EMUL_CONTINUE)
2194 		return rc;
2195 	ctxt->src.val = old_eip;
2196 	rc = em_push(ctxt);
2197 	return rc;
2198 }
2199 
2200 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2201 {
2202 	u64 old = ctxt->dst.orig_val64;
2203 
2204 	if (ctxt->dst.bytes == 16)
2205 		return X86EMUL_UNHANDLEABLE;
2206 
2207 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2208 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2209 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2210 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2211 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2212 	} else {
2213 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2214 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2215 
2216 		ctxt->eflags |= X86_EFLAGS_ZF;
2217 	}
2218 	return X86EMUL_CONTINUE;
2219 }
2220 
2221 static int em_ret(struct x86_emulate_ctxt *ctxt)
2222 {
2223 	int rc;
2224 	unsigned long eip;
2225 
2226 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2227 	if (rc != X86EMUL_CONTINUE)
2228 		return rc;
2229 
2230 	return assign_eip_near(ctxt, eip);
2231 }
2232 
2233 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2234 {
2235 	int rc;
2236 	unsigned long eip, cs;
2237 	int cpl = ctxt->ops->cpl(ctxt);
2238 	struct desc_struct new_desc;
2239 
2240 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2241 	if (rc != X86EMUL_CONTINUE)
2242 		return rc;
2243 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2244 	if (rc != X86EMUL_CONTINUE)
2245 		return rc;
2246 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2247 				       X86_TRANSFER_RET,
2248 				       &new_desc);
2249 	if (rc != X86EMUL_CONTINUE)
2250 		return rc;
2251 	rc = assign_eip_far(ctxt, eip, &new_desc);
2252 	/* Error handling is not implemented. */
2253 	if (rc != X86EMUL_CONTINUE)
2254 		return X86EMUL_UNHANDLEABLE;
2255 
2256 	return rc;
2257 }
2258 
2259 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2260 {
2261         int rc;
2262 
2263         rc = em_ret_far(ctxt);
2264         if (rc != X86EMUL_CONTINUE)
2265                 return rc;
2266         rsp_increment(ctxt, ctxt->src.val);
2267         return X86EMUL_CONTINUE;
2268 }
2269 
2270 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2271 {
2272 	/* Save real source value, then compare EAX against destination. */
2273 	ctxt->dst.orig_val = ctxt->dst.val;
2274 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2275 	ctxt->src.orig_val = ctxt->src.val;
2276 	ctxt->src.val = ctxt->dst.orig_val;
2277 	fastop(ctxt, em_cmp);
2278 
2279 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2280 		/* Success: write back to memory; no update of EAX */
2281 		ctxt->src.type = OP_NONE;
2282 		ctxt->dst.val = ctxt->src.orig_val;
2283 	} else {
2284 		/* Failure: write the value we saw to EAX. */
2285 		ctxt->src.type = OP_REG;
2286 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2287 		ctxt->src.val = ctxt->dst.orig_val;
2288 		/* Create write-cycle to dest by writing the same value */
2289 		ctxt->dst.val = ctxt->dst.orig_val;
2290 	}
2291 	return X86EMUL_CONTINUE;
2292 }
2293 
2294 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2295 {
2296 	int seg = ctxt->src2.val;
2297 	unsigned short sel;
2298 	int rc;
2299 
2300 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2301 
2302 	rc = load_segment_descriptor(ctxt, sel, seg);
2303 	if (rc != X86EMUL_CONTINUE)
2304 		return rc;
2305 
2306 	ctxt->dst.val = ctxt->src.val;
2307 	return rc;
2308 }
2309 
2310 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt)
2311 {
2312 #ifdef CONFIG_X86_64
2313 	return ctxt->ops->guest_has_long_mode(ctxt);
2314 #else
2315 	return false;
2316 #endif
2317 }
2318 
2319 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags)
2320 {
2321 	desc->g    = (flags >> 23) & 1;
2322 	desc->d    = (flags >> 22) & 1;
2323 	desc->l    = (flags >> 21) & 1;
2324 	desc->avl  = (flags >> 20) & 1;
2325 	desc->p    = (flags >> 15) & 1;
2326 	desc->dpl  = (flags >> 13) & 3;
2327 	desc->s    = (flags >> 12) & 1;
2328 	desc->type = (flags >>  8) & 15;
2329 }
2330 
2331 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, const char *smstate,
2332 			   int n)
2333 {
2334 	struct desc_struct desc;
2335 	int offset;
2336 	u16 selector;
2337 
2338 	selector = GET_SMSTATE(u32, smstate, 0x7fa8 + n * 4);
2339 
2340 	if (n < 3)
2341 		offset = 0x7f84 + n * 12;
2342 	else
2343 		offset = 0x7f2c + (n - 3) * 12;
2344 
2345 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2346 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2347 	rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, offset));
2348 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, n);
2349 	return X86EMUL_CONTINUE;
2350 }
2351 
2352 #ifdef CONFIG_X86_64
2353 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, const char *smstate,
2354 			   int n)
2355 {
2356 	struct desc_struct desc;
2357 	int offset;
2358 	u16 selector;
2359 	u32 base3;
2360 
2361 	offset = 0x7e00 + n * 16;
2362 
2363 	selector =                GET_SMSTATE(u16, smstate, offset);
2364 	rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smstate, offset + 2) << 8);
2365 	set_desc_limit(&desc,     GET_SMSTATE(u32, smstate, offset + 4));
2366 	set_desc_base(&desc,      GET_SMSTATE(u32, smstate, offset + 8));
2367 	base3 =                   GET_SMSTATE(u32, smstate, offset + 12);
2368 
2369 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, n);
2370 	return X86EMUL_CONTINUE;
2371 }
2372 #endif
2373 
2374 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt,
2375 				    u64 cr0, u64 cr3, u64 cr4)
2376 {
2377 	int bad;
2378 	u64 pcid;
2379 
2380 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
2381 	pcid = 0;
2382 	if (cr4 & X86_CR4_PCIDE) {
2383 		pcid = cr3 & 0xfff;
2384 		cr3 &= ~0xfff;
2385 	}
2386 
2387 	bad = ctxt->ops->set_cr(ctxt, 3, cr3);
2388 	if (bad)
2389 		return X86EMUL_UNHANDLEABLE;
2390 
2391 	/*
2392 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
2393 	 * Then enable protected mode.	However, PCID cannot be enabled
2394 	 * if EFER.LMA=0, so set it separately.
2395 	 */
2396 	bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2397 	if (bad)
2398 		return X86EMUL_UNHANDLEABLE;
2399 
2400 	bad = ctxt->ops->set_cr(ctxt, 0, cr0);
2401 	if (bad)
2402 		return X86EMUL_UNHANDLEABLE;
2403 
2404 	if (cr4 & X86_CR4_PCIDE) {
2405 		bad = ctxt->ops->set_cr(ctxt, 4, cr4);
2406 		if (bad)
2407 			return X86EMUL_UNHANDLEABLE;
2408 		if (pcid) {
2409 			bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid);
2410 			if (bad)
2411 				return X86EMUL_UNHANDLEABLE;
2412 		}
2413 
2414 	}
2415 
2416 	return X86EMUL_CONTINUE;
2417 }
2418 
2419 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
2420 			     const char *smstate)
2421 {
2422 	struct desc_struct desc;
2423 	struct desc_ptr dt;
2424 	u16 selector;
2425 	u32 val, cr0, cr3, cr4;
2426 	int i;
2427 
2428 	cr0 =                      GET_SMSTATE(u32, smstate, 0x7ffc);
2429 	cr3 =                      GET_SMSTATE(u32, smstate, 0x7ff8);
2430 	ctxt->eflags =             GET_SMSTATE(u32, smstate, 0x7ff4) | X86_EFLAGS_FIXED;
2431 	ctxt->_eip =               GET_SMSTATE(u32, smstate, 0x7ff0);
2432 
2433 	for (i = 0; i < NR_EMULATOR_GPRS; i++)
2434 		*reg_write(ctxt, i) = GET_SMSTATE(u32, smstate, 0x7fd0 + i * 4);
2435 
2436 	val = GET_SMSTATE(u32, smstate, 0x7fcc);
2437 
2438 	if (ctxt->ops->set_dr(ctxt, 6, val))
2439 		return X86EMUL_UNHANDLEABLE;
2440 
2441 	val = GET_SMSTATE(u32, smstate, 0x7fc8);
2442 
2443 	if (ctxt->ops->set_dr(ctxt, 7, val))
2444 		return X86EMUL_UNHANDLEABLE;
2445 
2446 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc4);
2447 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f64));
2448 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f60));
2449 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f5c));
2450 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR);
2451 
2452 	selector =                 GET_SMSTATE(u32, smstate, 0x7fc0);
2453 	set_desc_base(&desc,       GET_SMSTATE(u32, smstate, 0x7f80));
2454 	set_desc_limit(&desc,      GET_SMSTATE(u32, smstate, 0x7f7c));
2455 	rsm_set_desc_flags(&desc,  GET_SMSTATE(u32, smstate, 0x7f78));
2456 	ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR);
2457 
2458 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f74);
2459 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f70);
2460 	ctxt->ops->set_gdt(ctxt, &dt);
2461 
2462 	dt.address =               GET_SMSTATE(u32, smstate, 0x7f58);
2463 	dt.size =                  GET_SMSTATE(u32, smstate, 0x7f54);
2464 	ctxt->ops->set_idt(ctxt, &dt);
2465 
2466 	for (i = 0; i < 6; i++) {
2467 		int r = rsm_load_seg_32(ctxt, smstate, i);
2468 		if (r != X86EMUL_CONTINUE)
2469 			return r;
2470 	}
2471 
2472 	cr4 = GET_SMSTATE(u32, smstate, 0x7f14);
2473 
2474 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7ef8));
2475 
2476 	return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2477 }
2478 
2479 #ifdef CONFIG_X86_64
2480 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
2481 			     const char *smstate)
2482 {
2483 	struct desc_struct desc;
2484 	struct desc_ptr dt;
2485 	u64 val, cr0, cr3, cr4;
2486 	u32 base3;
2487 	u16 selector;
2488 	int i, r;
2489 
2490 	for (i = 0; i < NR_EMULATOR_GPRS; i++)
2491 		*reg_write(ctxt, i) = GET_SMSTATE(u64, smstate, 0x7ff8 - i * 8);
2492 
2493 	ctxt->_eip   = GET_SMSTATE(u64, smstate, 0x7f78);
2494 	ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7f70) | X86_EFLAGS_FIXED;
2495 
2496 	val = GET_SMSTATE(u64, smstate, 0x7f68);
2497 
2498 	if (ctxt->ops->set_dr(ctxt, 6, val))
2499 		return X86EMUL_UNHANDLEABLE;
2500 
2501 	val = GET_SMSTATE(u64, smstate, 0x7f60);
2502 
2503 	if (ctxt->ops->set_dr(ctxt, 7, val))
2504 		return X86EMUL_UNHANDLEABLE;
2505 
2506 	cr0 =                       GET_SMSTATE(u64, smstate, 0x7f58);
2507 	cr3 =                       GET_SMSTATE(u64, smstate, 0x7f50);
2508 	cr4 =                       GET_SMSTATE(u64, smstate, 0x7f48);
2509 	ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7f00));
2510 	val =                       GET_SMSTATE(u64, smstate, 0x7ed0);
2511 
2512 	if (ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA))
2513 		return X86EMUL_UNHANDLEABLE;
2514 
2515 	selector =                  GET_SMSTATE(u32, smstate, 0x7e90);
2516 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e92) << 8);
2517 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e94));
2518 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e98));
2519 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e9c);
2520 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR);
2521 
2522 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e84);
2523 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e88);
2524 	ctxt->ops->set_idt(ctxt, &dt);
2525 
2526 	selector =                  GET_SMSTATE(u32, smstate, 0x7e70);
2527 	rsm_set_desc_flags(&desc,   GET_SMSTATE(u32, smstate, 0x7e72) << 8);
2528 	set_desc_limit(&desc,       GET_SMSTATE(u32, smstate, 0x7e74));
2529 	set_desc_base(&desc,        GET_SMSTATE(u32, smstate, 0x7e78));
2530 	base3 =                     GET_SMSTATE(u32, smstate, 0x7e7c);
2531 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR);
2532 
2533 	dt.size =                   GET_SMSTATE(u32, smstate, 0x7e64);
2534 	dt.address =                GET_SMSTATE(u64, smstate, 0x7e68);
2535 	ctxt->ops->set_gdt(ctxt, &dt);
2536 
2537 	r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4);
2538 	if (r != X86EMUL_CONTINUE)
2539 		return r;
2540 
2541 	for (i = 0; i < 6; i++) {
2542 		r = rsm_load_seg_64(ctxt, smstate, i);
2543 		if (r != X86EMUL_CONTINUE)
2544 			return r;
2545 	}
2546 
2547 	return X86EMUL_CONTINUE;
2548 }
2549 #endif
2550 
2551 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2552 {
2553 	unsigned long cr0, cr4, efer;
2554 	char buf[512];
2555 	u64 smbase;
2556 	int ret;
2557 
2558 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0)
2559 		return emulate_ud(ctxt);
2560 
2561 	smbase = ctxt->ops->get_smbase(ctxt);
2562 
2563 	ret = ctxt->ops->read_phys(ctxt, smbase + 0xfe00, buf, sizeof(buf));
2564 	if (ret != X86EMUL_CONTINUE)
2565 		return X86EMUL_UNHANDLEABLE;
2566 
2567 	if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0)
2568 		ctxt->ops->set_nmi_mask(ctxt, false);
2569 
2570 	ctxt->ops->exiting_smm(ctxt);
2571 
2572 	/*
2573 	 * Get back to real mode, to prepare a safe state in which to load
2574 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
2575 	 * supports long mode.
2576 	 */
2577 	if (emulator_has_longmode(ctxt)) {
2578 		struct desc_struct cs_desc;
2579 
2580 		/* Zero CR4.PCIDE before CR0.PG.  */
2581 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2582 		if (cr4 & X86_CR4_PCIDE)
2583 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
2584 
2585 		/* A 32-bit code segment is required to clear EFER.LMA.  */
2586 		memset(&cs_desc, 0, sizeof(cs_desc));
2587 		cs_desc.type = 0xb;
2588 		cs_desc.s = cs_desc.g = cs_desc.p = 1;
2589 		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
2590 	}
2591 
2592 	/* For the 64-bit case, this will clear EFER.LMA.  */
2593 	cr0 = ctxt->ops->get_cr(ctxt, 0);
2594 	if (cr0 & X86_CR0_PE)
2595 		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
2596 
2597 	if (emulator_has_longmode(ctxt)) {
2598 		/* Clear CR4.PAE before clearing EFER.LME. */
2599 		cr4 = ctxt->ops->get_cr(ctxt, 4);
2600 		if (cr4 & X86_CR4_PAE)
2601 			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
2602 
2603 		/* And finally go back to 32-bit mode.  */
2604 		efer = 0;
2605 		ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
2606 	}
2607 
2608 	/*
2609 	 * Give leave_smm() a chance to make ISA-specific changes to the vCPU
2610 	 * state (e.g. enter guest mode) before loading state from the SMM
2611 	 * state-save area.
2612 	 */
2613 	if (ctxt->ops->leave_smm(ctxt, buf))
2614 		goto emulate_shutdown;
2615 
2616 #ifdef CONFIG_X86_64
2617 	if (emulator_has_longmode(ctxt))
2618 		ret = rsm_load_state_64(ctxt, buf);
2619 	else
2620 #endif
2621 		ret = rsm_load_state_32(ctxt, buf);
2622 
2623 	if (ret != X86EMUL_CONTINUE)
2624 		goto emulate_shutdown;
2625 
2626 	/*
2627 	 * Note, the ctxt->ops callbacks are responsible for handling side
2628 	 * effects when writing MSRs and CRs, e.g. MMU context resets, CPUID
2629 	 * runtime updates, etc...  If that changes, e.g. this flow is moved
2630 	 * out of the emulator to make it look more like enter_smm(), then
2631 	 * those side effects need to be explicitly handled for both success
2632 	 * and shutdown.
2633 	 */
2634 	return X86EMUL_CONTINUE;
2635 
2636 emulate_shutdown:
2637 	ctxt->ops->triple_fault(ctxt);
2638 	return X86EMUL_CONTINUE;
2639 }
2640 
2641 static void
2642 setup_syscalls_segments(struct desc_struct *cs, struct desc_struct *ss)
2643 {
2644 	cs->l = 0;		/* will be adjusted later */
2645 	set_desc_base(cs, 0);	/* flat segment */
2646 	cs->g = 1;		/* 4kb granularity */
2647 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2648 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2649 	cs->s = 1;
2650 	cs->dpl = 0;		/* will be adjusted later */
2651 	cs->p = 1;
2652 	cs->d = 1;
2653 	cs->avl = 0;
2654 
2655 	set_desc_base(ss, 0);	/* flat segment */
2656 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2657 	ss->g = 1;		/* 4kb granularity */
2658 	ss->s = 1;
2659 	ss->type = 0x03;	/* Read/Write, Accessed */
2660 	ss->d = 1;		/* 32bit stack segment */
2661 	ss->dpl = 0;
2662 	ss->p = 1;
2663 	ss->l = 0;
2664 	ss->avl = 0;
2665 }
2666 
2667 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2668 {
2669 	u32 eax, ebx, ecx, edx;
2670 
2671 	eax = ecx = 0;
2672 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2673 	return is_guest_vendor_intel(ebx, ecx, edx);
2674 }
2675 
2676 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2677 {
2678 	const struct x86_emulate_ops *ops = ctxt->ops;
2679 	u32 eax, ebx, ecx, edx;
2680 
2681 	/*
2682 	 * syscall should always be enabled in longmode - so only become
2683 	 * vendor specific (cpuid) if other modes are active...
2684 	 */
2685 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2686 		return true;
2687 
2688 	eax = 0x00000000;
2689 	ecx = 0x00000000;
2690 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2691 	/*
2692 	 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a
2693 	 * 64bit guest with a 32bit compat-app running will #UD !! While this
2694 	 * behaviour can be fixed (by emulating) into AMD response - CPUs of
2695 	 * AMD can't behave like Intel.
2696 	 */
2697 	if (is_guest_vendor_intel(ebx, ecx, edx))
2698 		return false;
2699 
2700 	if (is_guest_vendor_amd(ebx, ecx, edx) ||
2701 	    is_guest_vendor_hygon(ebx, ecx, edx))
2702 		return true;
2703 
2704 	/*
2705 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2706 	 * stricter rules...
2707 	 */
2708 	return false;
2709 }
2710 
2711 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2712 {
2713 	const struct x86_emulate_ops *ops = ctxt->ops;
2714 	struct desc_struct cs, ss;
2715 	u64 msr_data;
2716 	u16 cs_sel, ss_sel;
2717 	u64 efer = 0;
2718 
2719 	/* syscall is not available in real mode */
2720 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2721 	    ctxt->mode == X86EMUL_MODE_VM86)
2722 		return emulate_ud(ctxt);
2723 
2724 	if (!(em_syscall_is_enabled(ctxt)))
2725 		return emulate_ud(ctxt);
2726 
2727 	ops->get_msr(ctxt, MSR_EFER, &efer);
2728 	if (!(efer & EFER_SCE))
2729 		return emulate_ud(ctxt);
2730 
2731 	setup_syscalls_segments(&cs, &ss);
2732 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2733 	msr_data >>= 32;
2734 	cs_sel = (u16)(msr_data & 0xfffc);
2735 	ss_sel = (u16)(msr_data + 8);
2736 
2737 	if (efer & EFER_LMA) {
2738 		cs.d = 0;
2739 		cs.l = 1;
2740 	}
2741 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2742 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2743 
2744 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2745 	if (efer & EFER_LMA) {
2746 #ifdef CONFIG_X86_64
2747 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2748 
2749 		ops->get_msr(ctxt,
2750 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2751 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2752 		ctxt->_eip = msr_data;
2753 
2754 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2755 		ctxt->eflags &= ~msr_data;
2756 		ctxt->eflags |= X86_EFLAGS_FIXED;
2757 #endif
2758 	} else {
2759 		/* legacy mode */
2760 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2761 		ctxt->_eip = (u32)msr_data;
2762 
2763 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2764 	}
2765 
2766 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2767 	return X86EMUL_CONTINUE;
2768 }
2769 
2770 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2771 {
2772 	const struct x86_emulate_ops *ops = ctxt->ops;
2773 	struct desc_struct cs, ss;
2774 	u64 msr_data;
2775 	u16 cs_sel, ss_sel;
2776 	u64 efer = 0;
2777 
2778 	ops->get_msr(ctxt, MSR_EFER, &efer);
2779 	/* inject #GP if in real mode */
2780 	if (ctxt->mode == X86EMUL_MODE_REAL)
2781 		return emulate_gp(ctxt, 0);
2782 
2783 	/*
2784 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2785 	 * mode).
2786 	 */
2787 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2788 	    && !vendor_intel(ctxt))
2789 		return emulate_ud(ctxt);
2790 
2791 	/* sysenter/sysexit have not been tested in 64bit mode. */
2792 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2793 		return X86EMUL_UNHANDLEABLE;
2794 
2795 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2796 	if ((msr_data & 0xfffc) == 0x0)
2797 		return emulate_gp(ctxt, 0);
2798 
2799 	setup_syscalls_segments(&cs, &ss);
2800 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2801 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2802 	ss_sel = cs_sel + 8;
2803 	if (efer & EFER_LMA) {
2804 		cs.d = 0;
2805 		cs.l = 1;
2806 	}
2807 
2808 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2809 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2810 
2811 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2812 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2813 
2814 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2815 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2816 							      (u32)msr_data;
2817 	if (efer & EFER_LMA)
2818 		ctxt->mode = X86EMUL_MODE_PROT64;
2819 
2820 	return X86EMUL_CONTINUE;
2821 }
2822 
2823 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2824 {
2825 	const struct x86_emulate_ops *ops = ctxt->ops;
2826 	struct desc_struct cs, ss;
2827 	u64 msr_data, rcx, rdx;
2828 	int usermode;
2829 	u16 cs_sel = 0, ss_sel = 0;
2830 
2831 	/* inject #GP if in real mode or Virtual 8086 mode */
2832 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2833 	    ctxt->mode == X86EMUL_MODE_VM86)
2834 		return emulate_gp(ctxt, 0);
2835 
2836 	setup_syscalls_segments(&cs, &ss);
2837 
2838 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2839 		usermode = X86EMUL_MODE_PROT64;
2840 	else
2841 		usermode = X86EMUL_MODE_PROT32;
2842 
2843 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2844 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2845 
2846 	cs.dpl = 3;
2847 	ss.dpl = 3;
2848 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2849 	switch (usermode) {
2850 	case X86EMUL_MODE_PROT32:
2851 		cs_sel = (u16)(msr_data + 16);
2852 		if ((msr_data & 0xfffc) == 0x0)
2853 			return emulate_gp(ctxt, 0);
2854 		ss_sel = (u16)(msr_data + 24);
2855 		rcx = (u32)rcx;
2856 		rdx = (u32)rdx;
2857 		break;
2858 	case X86EMUL_MODE_PROT64:
2859 		cs_sel = (u16)(msr_data + 32);
2860 		if (msr_data == 0x0)
2861 			return emulate_gp(ctxt, 0);
2862 		ss_sel = cs_sel + 8;
2863 		cs.d = 0;
2864 		cs.l = 1;
2865 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2866 		    emul_is_noncanonical_address(rdx, ctxt))
2867 			return emulate_gp(ctxt, 0);
2868 		break;
2869 	}
2870 	cs_sel |= SEGMENT_RPL_MASK;
2871 	ss_sel |= SEGMENT_RPL_MASK;
2872 
2873 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2874 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2875 
2876 	ctxt->_eip = rdx;
2877 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2878 
2879 	return X86EMUL_CONTINUE;
2880 }
2881 
2882 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2883 {
2884 	int iopl;
2885 	if (ctxt->mode == X86EMUL_MODE_REAL)
2886 		return false;
2887 	if (ctxt->mode == X86EMUL_MODE_VM86)
2888 		return true;
2889 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2890 	return ctxt->ops->cpl(ctxt) > iopl;
2891 }
2892 
2893 #define VMWARE_PORT_VMPORT	(0x5658)
2894 #define VMWARE_PORT_VMRPC	(0x5659)
2895 
2896 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2897 					    u16 port, u16 len)
2898 {
2899 	const struct x86_emulate_ops *ops = ctxt->ops;
2900 	struct desc_struct tr_seg;
2901 	u32 base3;
2902 	int r;
2903 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2904 	unsigned mask = (1 << len) - 1;
2905 	unsigned long base;
2906 
2907 	/*
2908 	 * VMware allows access to these ports even if denied
2909 	 * by TSS I/O permission bitmap. Mimic behavior.
2910 	 */
2911 	if (enable_vmware_backdoor &&
2912 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2913 		return true;
2914 
2915 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2916 	if (!tr_seg.p)
2917 		return false;
2918 	if (desc_limit_scaled(&tr_seg) < 103)
2919 		return false;
2920 	base = get_desc_base(&tr_seg);
2921 #ifdef CONFIG_X86_64
2922 	base |= ((u64)base3) << 32;
2923 #endif
2924 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2925 	if (r != X86EMUL_CONTINUE)
2926 		return false;
2927 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2928 		return false;
2929 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2930 	if (r != X86EMUL_CONTINUE)
2931 		return false;
2932 	if ((perm >> bit_idx) & mask)
2933 		return false;
2934 	return true;
2935 }
2936 
2937 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt,
2938 				 u16 port, u16 len)
2939 {
2940 	if (ctxt->perm_ok)
2941 		return true;
2942 
2943 	if (emulator_bad_iopl(ctxt))
2944 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2945 			return false;
2946 
2947 	ctxt->perm_ok = true;
2948 
2949 	return true;
2950 }
2951 
2952 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2953 {
2954 	/*
2955 	 * Intel CPUs mask the counter and pointers in quite strange
2956 	 * manner when ECX is zero due to REP-string optimizations.
2957 	 */
2958 #ifdef CONFIG_X86_64
2959 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2960 		return;
2961 
2962 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2963 
2964 	switch (ctxt->b) {
2965 	case 0xa4:	/* movsb */
2966 	case 0xa5:	/* movsd/w */
2967 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2968 		fallthrough;
2969 	case 0xaa:	/* stosb */
2970 	case 0xab:	/* stosd/w */
2971 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2972 	}
2973 #endif
2974 }
2975 
2976 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2977 				struct tss_segment_16 *tss)
2978 {
2979 	tss->ip = ctxt->_eip;
2980 	tss->flag = ctxt->eflags;
2981 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2982 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2983 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2984 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2985 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2986 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2987 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2988 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2989 
2990 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2991 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2992 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2993 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2994 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2995 }
2996 
2997 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2998 				 struct tss_segment_16 *tss)
2999 {
3000 	int ret;
3001 	u8 cpl;
3002 
3003 	ctxt->_eip = tss->ip;
3004 	ctxt->eflags = tss->flag | 2;
3005 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
3006 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
3007 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
3008 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
3009 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
3010 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
3011 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
3012 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
3013 
3014 	/*
3015 	 * SDM says that segment selectors are loaded before segment
3016 	 * descriptors
3017 	 */
3018 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
3019 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3020 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3021 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3022 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3023 
3024 	cpl = tss->cs & 3;
3025 
3026 	/*
3027 	 * Now load segment descriptors. If fault happens at this stage
3028 	 * it is handled in a context of new task
3029 	 */
3030 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
3031 					X86_TRANSFER_TASK_SWITCH, NULL);
3032 	if (ret != X86EMUL_CONTINUE)
3033 		return ret;
3034 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3035 					X86_TRANSFER_TASK_SWITCH, NULL);
3036 	if (ret != X86EMUL_CONTINUE)
3037 		return ret;
3038 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3039 					X86_TRANSFER_TASK_SWITCH, NULL);
3040 	if (ret != X86EMUL_CONTINUE)
3041 		return ret;
3042 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3043 					X86_TRANSFER_TASK_SWITCH, NULL);
3044 	if (ret != X86EMUL_CONTINUE)
3045 		return ret;
3046 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3047 					X86_TRANSFER_TASK_SWITCH, NULL);
3048 	if (ret != X86EMUL_CONTINUE)
3049 		return ret;
3050 
3051 	return X86EMUL_CONTINUE;
3052 }
3053 
3054 static int task_switch_16(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
3055 			  ulong old_tss_base, struct desc_struct *new_desc)
3056 {
3057 	struct tss_segment_16 tss_seg;
3058 	int ret;
3059 	u32 new_tss_base = get_desc_base(new_desc);
3060 
3061 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3062 	if (ret != X86EMUL_CONTINUE)
3063 		return ret;
3064 
3065 	save_state_to_tss16(ctxt, &tss_seg);
3066 
3067 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3068 	if (ret != X86EMUL_CONTINUE)
3069 		return ret;
3070 
3071 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3072 	if (ret != X86EMUL_CONTINUE)
3073 		return ret;
3074 
3075 	if (old_tss_sel != 0xffff) {
3076 		tss_seg.prev_task_link = old_tss_sel;
3077 
3078 		ret = linear_write_system(ctxt, new_tss_base,
3079 					  &tss_seg.prev_task_link,
3080 					  sizeof(tss_seg.prev_task_link));
3081 		if (ret != X86EMUL_CONTINUE)
3082 			return ret;
3083 	}
3084 
3085 	return load_state_from_tss16(ctxt, &tss_seg);
3086 }
3087 
3088 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
3089 				struct tss_segment_32 *tss)
3090 {
3091 	/* CR3 and ldt selector are not saved intentionally */
3092 	tss->eip = ctxt->_eip;
3093 	tss->eflags = ctxt->eflags;
3094 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
3095 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
3096 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
3097 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
3098 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
3099 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
3100 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
3101 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
3102 
3103 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
3104 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
3105 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
3106 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
3107 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
3108 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
3109 }
3110 
3111 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
3112 				 struct tss_segment_32 *tss)
3113 {
3114 	int ret;
3115 	u8 cpl;
3116 
3117 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
3118 		return emulate_gp(ctxt, 0);
3119 	ctxt->_eip = tss->eip;
3120 	ctxt->eflags = tss->eflags | 2;
3121 
3122 	/* General purpose registers */
3123 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
3124 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
3125 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
3126 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
3127 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
3128 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
3129 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
3130 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
3131 
3132 	/*
3133 	 * SDM says that segment selectors are loaded before segment
3134 	 * descriptors.  This is important because CPL checks will
3135 	 * use CS.RPL.
3136 	 */
3137 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
3138 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
3139 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
3140 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
3141 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
3142 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
3143 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
3144 
3145 	/*
3146 	 * If we're switching between Protected Mode and VM86, we need to make
3147 	 * sure to update the mode before loading the segment descriptors so
3148 	 * that the selectors are interpreted correctly.
3149 	 */
3150 	if (ctxt->eflags & X86_EFLAGS_VM) {
3151 		ctxt->mode = X86EMUL_MODE_VM86;
3152 		cpl = 3;
3153 	} else {
3154 		ctxt->mode = X86EMUL_MODE_PROT32;
3155 		cpl = tss->cs & 3;
3156 	}
3157 
3158 	/*
3159 	 * Now load segment descriptors. If fault happens at this stage
3160 	 * it is handled in a context of new task
3161 	 */
3162 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
3163 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
3164 	if (ret != X86EMUL_CONTINUE)
3165 		return ret;
3166 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
3167 					X86_TRANSFER_TASK_SWITCH, NULL);
3168 	if (ret != X86EMUL_CONTINUE)
3169 		return ret;
3170 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
3171 					X86_TRANSFER_TASK_SWITCH, NULL);
3172 	if (ret != X86EMUL_CONTINUE)
3173 		return ret;
3174 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
3175 					X86_TRANSFER_TASK_SWITCH, NULL);
3176 	if (ret != X86EMUL_CONTINUE)
3177 		return ret;
3178 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
3179 					X86_TRANSFER_TASK_SWITCH, NULL);
3180 	if (ret != X86EMUL_CONTINUE)
3181 		return ret;
3182 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
3183 					X86_TRANSFER_TASK_SWITCH, NULL);
3184 	if (ret != X86EMUL_CONTINUE)
3185 		return ret;
3186 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
3187 					X86_TRANSFER_TASK_SWITCH, NULL);
3188 
3189 	return ret;
3190 }
3191 
3192 static int task_switch_32(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
3193 			  ulong old_tss_base, struct desc_struct *new_desc)
3194 {
3195 	struct tss_segment_32 tss_seg;
3196 	int ret;
3197 	u32 new_tss_base = get_desc_base(new_desc);
3198 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
3199 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
3200 
3201 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
3202 	if (ret != X86EMUL_CONTINUE)
3203 		return ret;
3204 
3205 	save_state_to_tss32(ctxt, &tss_seg);
3206 
3207 	/* Only GP registers and segment selectors are saved */
3208 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
3209 				  ldt_sel_offset - eip_offset);
3210 	if (ret != X86EMUL_CONTINUE)
3211 		return ret;
3212 
3213 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
3214 	if (ret != X86EMUL_CONTINUE)
3215 		return ret;
3216 
3217 	if (old_tss_sel != 0xffff) {
3218 		tss_seg.prev_task_link = old_tss_sel;
3219 
3220 		ret = linear_write_system(ctxt, new_tss_base,
3221 					  &tss_seg.prev_task_link,
3222 					  sizeof(tss_seg.prev_task_link));
3223 		if (ret != X86EMUL_CONTINUE)
3224 			return ret;
3225 	}
3226 
3227 	return load_state_from_tss32(ctxt, &tss_seg);
3228 }
3229 
3230 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
3231 				   u16 tss_selector, int idt_index, int reason,
3232 				   bool has_error_code, u32 error_code)
3233 {
3234 	const struct x86_emulate_ops *ops = ctxt->ops;
3235 	struct desc_struct curr_tss_desc, next_tss_desc;
3236 	int ret;
3237 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
3238 	ulong old_tss_base =
3239 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
3240 	u32 desc_limit;
3241 	ulong desc_addr, dr7;
3242 
3243 	/* FIXME: old_tss_base == ~0 ? */
3244 
3245 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
3246 	if (ret != X86EMUL_CONTINUE)
3247 		return ret;
3248 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
3249 	if (ret != X86EMUL_CONTINUE)
3250 		return ret;
3251 
3252 	/* FIXME: check that next_tss_desc is tss */
3253 
3254 	/*
3255 	 * Check privileges. The three cases are task switch caused by...
3256 	 *
3257 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
3258 	 * 2. Exception/IRQ/iret: No check is performed
3259 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
3260 	 *    hardware checks it before exiting.
3261 	 */
3262 	if (reason == TASK_SWITCH_GATE) {
3263 		if (idt_index != -1) {
3264 			/* Software interrupts */
3265 			struct desc_struct task_gate_desc;
3266 			int dpl;
3267 
3268 			ret = read_interrupt_descriptor(ctxt, idt_index,
3269 							&task_gate_desc);
3270 			if (ret != X86EMUL_CONTINUE)
3271 				return ret;
3272 
3273 			dpl = task_gate_desc.dpl;
3274 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
3275 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
3276 		}
3277 	}
3278 
3279 	desc_limit = desc_limit_scaled(&next_tss_desc);
3280 	if (!next_tss_desc.p ||
3281 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
3282 	     desc_limit < 0x2b)) {
3283 		return emulate_ts(ctxt, tss_selector & 0xfffc);
3284 	}
3285 
3286 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
3287 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
3288 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
3289 	}
3290 
3291 	if (reason == TASK_SWITCH_IRET)
3292 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
3293 
3294 	/* set back link to prev task only if NT bit is set in eflags
3295 	   note that old_tss_sel is not used after this point */
3296 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
3297 		old_tss_sel = 0xffff;
3298 
3299 	if (next_tss_desc.type & 8)
3300 		ret = task_switch_32(ctxt, old_tss_sel, old_tss_base, &next_tss_desc);
3301 	else
3302 		ret = task_switch_16(ctxt, old_tss_sel,
3303 				     old_tss_base, &next_tss_desc);
3304 	if (ret != X86EMUL_CONTINUE)
3305 		return ret;
3306 
3307 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
3308 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3309 
3310 	if (reason != TASK_SWITCH_IRET) {
3311 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3312 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3313 	}
3314 
3315 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3316 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3317 
3318 	if (has_error_code) {
3319 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3320 		ctxt->lock_prefix = 0;
3321 		ctxt->src.val = (unsigned long) error_code;
3322 		ret = em_push(ctxt);
3323 	}
3324 
3325 	ops->get_dr(ctxt, 7, &dr7);
3326 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3327 
3328 	return ret;
3329 }
3330 
3331 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3332 			 u16 tss_selector, int idt_index, int reason,
3333 			 bool has_error_code, u32 error_code)
3334 {
3335 	int rc;
3336 
3337 	invalidate_registers(ctxt);
3338 	ctxt->_eip = ctxt->eip;
3339 	ctxt->dst.type = OP_NONE;
3340 
3341 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3342 				     has_error_code, error_code);
3343 
3344 	if (rc == X86EMUL_CONTINUE) {
3345 		ctxt->eip = ctxt->_eip;
3346 		writeback_registers(ctxt);
3347 	}
3348 
3349 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3350 }
3351 
3352 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3353 		struct operand *op)
3354 {
3355 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3356 
3357 	register_address_increment(ctxt, reg, df * op->bytes);
3358 	op->addr.mem.ea = register_address(ctxt, reg);
3359 }
3360 
3361 static int em_das(struct x86_emulate_ctxt *ctxt)
3362 {
3363 	u8 al, old_al;
3364 	bool af, cf, old_cf;
3365 
3366 	cf = ctxt->eflags & X86_EFLAGS_CF;
3367 	al = ctxt->dst.val;
3368 
3369 	old_al = al;
3370 	old_cf = cf;
3371 	cf = false;
3372 	af = ctxt->eflags & X86_EFLAGS_AF;
3373 	if ((al & 0x0f) > 9 || af) {
3374 		al -= 6;
3375 		cf = old_cf | (al >= 250);
3376 		af = true;
3377 	} else {
3378 		af = false;
3379 	}
3380 	if (old_al > 0x99 || old_cf) {
3381 		al -= 0x60;
3382 		cf = true;
3383 	}
3384 
3385 	ctxt->dst.val = al;
3386 	/* Set PF, ZF, SF */
3387 	ctxt->src.type = OP_IMM;
3388 	ctxt->src.val = 0;
3389 	ctxt->src.bytes = 1;
3390 	fastop(ctxt, em_or);
3391 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3392 	if (cf)
3393 		ctxt->eflags |= X86_EFLAGS_CF;
3394 	if (af)
3395 		ctxt->eflags |= X86_EFLAGS_AF;
3396 	return X86EMUL_CONTINUE;
3397 }
3398 
3399 static int em_aam(struct x86_emulate_ctxt *ctxt)
3400 {
3401 	u8 al, ah;
3402 
3403 	if (ctxt->src.val == 0)
3404 		return emulate_de(ctxt);
3405 
3406 	al = ctxt->dst.val & 0xff;
3407 	ah = al / ctxt->src.val;
3408 	al %= ctxt->src.val;
3409 
3410 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3411 
3412 	/* Set PF, ZF, SF */
3413 	ctxt->src.type = OP_IMM;
3414 	ctxt->src.val = 0;
3415 	ctxt->src.bytes = 1;
3416 	fastop(ctxt, em_or);
3417 
3418 	return X86EMUL_CONTINUE;
3419 }
3420 
3421 static int em_aad(struct x86_emulate_ctxt *ctxt)
3422 {
3423 	u8 al = ctxt->dst.val & 0xff;
3424 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3425 
3426 	al = (al + (ah * ctxt->src.val)) & 0xff;
3427 
3428 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3429 
3430 	/* Set PF, ZF, SF */
3431 	ctxt->src.type = OP_IMM;
3432 	ctxt->src.val = 0;
3433 	ctxt->src.bytes = 1;
3434 	fastop(ctxt, em_or);
3435 
3436 	return X86EMUL_CONTINUE;
3437 }
3438 
3439 static int em_call(struct x86_emulate_ctxt *ctxt)
3440 {
3441 	int rc;
3442 	long rel = ctxt->src.val;
3443 
3444 	ctxt->src.val = (unsigned long)ctxt->_eip;
3445 	rc = jmp_rel(ctxt, rel);
3446 	if (rc != X86EMUL_CONTINUE)
3447 		return rc;
3448 	return em_push(ctxt);
3449 }
3450 
3451 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3452 {
3453 	u16 sel, old_cs;
3454 	ulong old_eip;
3455 	int rc;
3456 	struct desc_struct old_desc, new_desc;
3457 	const struct x86_emulate_ops *ops = ctxt->ops;
3458 	int cpl = ctxt->ops->cpl(ctxt);
3459 	enum x86emul_mode prev_mode = ctxt->mode;
3460 
3461 	old_eip = ctxt->_eip;
3462 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3463 
3464 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3465 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3466 				       X86_TRANSFER_CALL_JMP, &new_desc);
3467 	if (rc != X86EMUL_CONTINUE)
3468 		return rc;
3469 
3470 	rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc);
3471 	if (rc != X86EMUL_CONTINUE)
3472 		goto fail;
3473 
3474 	ctxt->src.val = old_cs;
3475 	rc = em_push(ctxt);
3476 	if (rc != X86EMUL_CONTINUE)
3477 		goto fail;
3478 
3479 	ctxt->src.val = old_eip;
3480 	rc = em_push(ctxt);
3481 	/* If we failed, we tainted the memory, but the very least we should
3482 	   restore cs */
3483 	if (rc != X86EMUL_CONTINUE) {
3484 		pr_warn_once("faulting far call emulation tainted memory\n");
3485 		goto fail;
3486 	}
3487 	return rc;
3488 fail:
3489 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3490 	ctxt->mode = prev_mode;
3491 	return rc;
3492 
3493 }
3494 
3495 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3496 {
3497 	int rc;
3498 	unsigned long eip;
3499 
3500 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3501 	if (rc != X86EMUL_CONTINUE)
3502 		return rc;
3503 	rc = assign_eip_near(ctxt, eip);
3504 	if (rc != X86EMUL_CONTINUE)
3505 		return rc;
3506 	rsp_increment(ctxt, ctxt->src.val);
3507 	return X86EMUL_CONTINUE;
3508 }
3509 
3510 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3511 {
3512 	/* Write back the register source. */
3513 	ctxt->src.val = ctxt->dst.val;
3514 	write_register_operand(&ctxt->src);
3515 
3516 	/* Write back the memory destination with implicit LOCK prefix. */
3517 	ctxt->dst.val = ctxt->src.orig_val;
3518 	ctxt->lock_prefix = 1;
3519 	return X86EMUL_CONTINUE;
3520 }
3521 
3522 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3523 {
3524 	ctxt->dst.val = ctxt->src2.val;
3525 	return fastop(ctxt, em_imul);
3526 }
3527 
3528 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3529 {
3530 	ctxt->dst.type = OP_REG;
3531 	ctxt->dst.bytes = ctxt->src.bytes;
3532 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3533 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3534 
3535 	return X86EMUL_CONTINUE;
3536 }
3537 
3538 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3539 {
3540 	u64 tsc_aux = 0;
3541 
3542 	if (!ctxt->ops->guest_has_rdpid(ctxt))
3543 		return emulate_ud(ctxt);
3544 
3545 	ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux);
3546 	ctxt->dst.val = tsc_aux;
3547 	return X86EMUL_CONTINUE;
3548 }
3549 
3550 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3551 {
3552 	u64 tsc = 0;
3553 
3554 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3555 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3556 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3557 	return X86EMUL_CONTINUE;
3558 }
3559 
3560 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3561 {
3562 	u64 pmc;
3563 
3564 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3565 		return emulate_gp(ctxt, 0);
3566 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3567 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3568 	return X86EMUL_CONTINUE;
3569 }
3570 
3571 static int em_mov(struct x86_emulate_ctxt *ctxt)
3572 {
3573 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3574 	return X86EMUL_CONTINUE;
3575 }
3576 
3577 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3578 {
3579 	u16 tmp;
3580 
3581 	if (!ctxt->ops->guest_has_movbe(ctxt))
3582 		return emulate_ud(ctxt);
3583 
3584 	switch (ctxt->op_bytes) {
3585 	case 2:
3586 		/*
3587 		 * From MOVBE definition: "...When the operand size is 16 bits,
3588 		 * the upper word of the destination register remains unchanged
3589 		 * ..."
3590 		 *
3591 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3592 		 * rules so we have to do the operation almost per hand.
3593 		 */
3594 		tmp = (u16)ctxt->src.val;
3595 		ctxt->dst.val &= ~0xffffUL;
3596 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3597 		break;
3598 	case 4:
3599 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3600 		break;
3601 	case 8:
3602 		ctxt->dst.val = swab64(ctxt->src.val);
3603 		break;
3604 	default:
3605 		BUG();
3606 	}
3607 	return X86EMUL_CONTINUE;
3608 }
3609 
3610 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3611 {
3612 	if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
3613 		return emulate_gp(ctxt, 0);
3614 
3615 	/* Disable writeback. */
3616 	ctxt->dst.type = OP_NONE;
3617 	return X86EMUL_CONTINUE;
3618 }
3619 
3620 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3621 {
3622 	unsigned long val;
3623 
3624 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3625 		val = ctxt->src.val & ~0ULL;
3626 	else
3627 		val = ctxt->src.val & ~0U;
3628 
3629 	/* #UD condition is already handled. */
3630 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3631 		return emulate_gp(ctxt, 0);
3632 
3633 	/* Disable writeback. */
3634 	ctxt->dst.type = OP_NONE;
3635 	return X86EMUL_CONTINUE;
3636 }
3637 
3638 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3639 {
3640 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3641 	u64 msr_data;
3642 	int r;
3643 
3644 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3645 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3646 	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);
3647 
3648 	if (r == X86EMUL_IO_NEEDED)
3649 		return r;
3650 
3651 	if (r > 0)
3652 		return emulate_gp(ctxt, 0);
3653 
3654 	return r < 0 ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
3655 }
3656 
3657 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3658 {
3659 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3660 	u64 msr_data;
3661 	int r;
3662 
3663 	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);
3664 
3665 	if (r == X86EMUL_IO_NEEDED)
3666 		return r;
3667 
3668 	if (r)
3669 		return emulate_gp(ctxt, 0);
3670 
3671 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3672 	*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3673 	return X86EMUL_CONTINUE;
3674 }
3675 
3676 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3677 {
3678 	if (segment > VCPU_SREG_GS &&
3679 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3680 	    ctxt->ops->cpl(ctxt) > 0)
3681 		return emulate_gp(ctxt, 0);
3682 
3683 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3684 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3685 		ctxt->dst.bytes = 2;
3686 	return X86EMUL_CONTINUE;
3687 }
3688 
3689 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3690 {
3691 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3692 		return emulate_ud(ctxt);
3693 
3694 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3695 }
3696 
3697 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3698 {
3699 	u16 sel = ctxt->src.val;
3700 
3701 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3702 		return emulate_ud(ctxt);
3703 
3704 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3705 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3706 
3707 	/* Disable writeback. */
3708 	ctxt->dst.type = OP_NONE;
3709 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3710 }
3711 
3712 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3713 {
3714 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3715 }
3716 
3717 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3718 {
3719 	u16 sel = ctxt->src.val;
3720 
3721 	/* Disable writeback. */
3722 	ctxt->dst.type = OP_NONE;
3723 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3724 }
3725 
3726 static int em_str(struct x86_emulate_ctxt *ctxt)
3727 {
3728 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3729 }
3730 
3731 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3732 {
3733 	u16 sel = ctxt->src.val;
3734 
3735 	/* Disable writeback. */
3736 	ctxt->dst.type = OP_NONE;
3737 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3738 }
3739 
3740 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3741 {
3742 	int rc;
3743 	ulong linear;
3744 
3745 	rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear);
3746 	if (rc == X86EMUL_CONTINUE)
3747 		ctxt->ops->invlpg(ctxt, linear);
3748 	/* Disable writeback. */
3749 	ctxt->dst.type = OP_NONE;
3750 	return X86EMUL_CONTINUE;
3751 }
3752 
3753 static int em_clts(struct x86_emulate_ctxt *ctxt)
3754 {
3755 	ulong cr0;
3756 
3757 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3758 	cr0 &= ~X86_CR0_TS;
3759 	ctxt->ops->set_cr(ctxt, 0, cr0);
3760 	return X86EMUL_CONTINUE;
3761 }
3762 
3763 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3764 {
3765 	int rc = ctxt->ops->fix_hypercall(ctxt);
3766 
3767 	if (rc != X86EMUL_CONTINUE)
3768 		return rc;
3769 
3770 	/* Let the processor re-execute the fixed hypercall */
3771 	ctxt->_eip = ctxt->eip;
3772 	/* Disable writeback. */
3773 	ctxt->dst.type = OP_NONE;
3774 	return X86EMUL_CONTINUE;
3775 }
3776 
3777 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3778 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3779 					      struct desc_ptr *ptr))
3780 {
3781 	struct desc_ptr desc_ptr;
3782 
3783 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3784 	    ctxt->ops->cpl(ctxt) > 0)
3785 		return emulate_gp(ctxt, 0);
3786 
3787 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3788 		ctxt->op_bytes = 8;
3789 	get(ctxt, &desc_ptr);
3790 	if (ctxt->op_bytes == 2) {
3791 		ctxt->op_bytes = 4;
3792 		desc_ptr.address &= 0x00ffffff;
3793 	}
3794 	/* Disable writeback. */
3795 	ctxt->dst.type = OP_NONE;
3796 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3797 				   &desc_ptr, 2 + ctxt->op_bytes);
3798 }
3799 
3800 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3801 {
3802 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3803 }
3804 
3805 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3806 {
3807 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3808 }
3809 
3810 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3811 {
3812 	struct desc_ptr desc_ptr;
3813 	int rc;
3814 
3815 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3816 		ctxt->op_bytes = 8;
3817 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3818 			     &desc_ptr.size, &desc_ptr.address,
3819 			     ctxt->op_bytes);
3820 	if (rc != X86EMUL_CONTINUE)
3821 		return rc;
3822 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3823 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3824 		return emulate_gp(ctxt, 0);
3825 	if (lgdt)
3826 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3827 	else
3828 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3829 	/* Disable writeback. */
3830 	ctxt->dst.type = OP_NONE;
3831 	return X86EMUL_CONTINUE;
3832 }
3833 
3834 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3835 {
3836 	return em_lgdt_lidt(ctxt, true);
3837 }
3838 
3839 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3840 {
3841 	return em_lgdt_lidt(ctxt, false);
3842 }
3843 
3844 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3845 {
3846 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3847 	    ctxt->ops->cpl(ctxt) > 0)
3848 		return emulate_gp(ctxt, 0);
3849 
3850 	if (ctxt->dst.type == OP_MEM)
3851 		ctxt->dst.bytes = 2;
3852 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3853 	return X86EMUL_CONTINUE;
3854 }
3855 
3856 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3857 {
3858 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3859 			  | (ctxt->src.val & 0x0f));
3860 	ctxt->dst.type = OP_NONE;
3861 	return X86EMUL_CONTINUE;
3862 }
3863 
3864 static int em_loop(struct x86_emulate_ctxt *ctxt)
3865 {
3866 	int rc = X86EMUL_CONTINUE;
3867 
3868 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3869 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3870 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3871 		rc = jmp_rel(ctxt, ctxt->src.val);
3872 
3873 	return rc;
3874 }
3875 
3876 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3877 {
3878 	int rc = X86EMUL_CONTINUE;
3879 
3880 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3881 		rc = jmp_rel(ctxt, ctxt->src.val);
3882 
3883 	return rc;
3884 }
3885 
3886 static int em_in(struct x86_emulate_ctxt *ctxt)
3887 {
3888 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3889 			     &ctxt->dst.val))
3890 		return X86EMUL_IO_NEEDED;
3891 
3892 	return X86EMUL_CONTINUE;
3893 }
3894 
3895 static int em_out(struct x86_emulate_ctxt *ctxt)
3896 {
3897 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3898 				    &ctxt->src.val, 1);
3899 	/* Disable writeback. */
3900 	ctxt->dst.type = OP_NONE;
3901 	return X86EMUL_CONTINUE;
3902 }
3903 
3904 static int em_cli(struct x86_emulate_ctxt *ctxt)
3905 {
3906 	if (emulator_bad_iopl(ctxt))
3907 		return emulate_gp(ctxt, 0);
3908 
3909 	ctxt->eflags &= ~X86_EFLAGS_IF;
3910 	return X86EMUL_CONTINUE;
3911 }
3912 
3913 static int em_sti(struct x86_emulate_ctxt *ctxt)
3914 {
3915 	if (emulator_bad_iopl(ctxt))
3916 		return emulate_gp(ctxt, 0);
3917 
3918 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3919 	ctxt->eflags |= X86_EFLAGS_IF;
3920 	return X86EMUL_CONTINUE;
3921 }
3922 
3923 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3924 {
3925 	u32 eax, ebx, ecx, edx;
3926 	u64 msr = 0;
3927 
3928 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3929 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3930 	    ctxt->ops->cpl(ctxt)) {
3931 		return emulate_gp(ctxt, 0);
3932 	}
3933 
3934 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3935 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3936 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3937 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3938 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3939 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3940 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3941 	return X86EMUL_CONTINUE;
3942 }
3943 
3944 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3945 {
3946 	u32 flags;
3947 
3948 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3949 		X86_EFLAGS_SF;
3950 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3951 
3952 	ctxt->eflags &= ~0xffUL;
3953 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3954 	return X86EMUL_CONTINUE;
3955 }
3956 
3957 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3958 {
3959 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3960 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3961 	return X86EMUL_CONTINUE;
3962 }
3963 
3964 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3965 {
3966 	switch (ctxt->op_bytes) {
3967 #ifdef CONFIG_X86_64
3968 	case 8:
3969 		asm("bswap %0" : "+r"(ctxt->dst.val));
3970 		break;
3971 #endif
3972 	default:
3973 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3974 		break;
3975 	}
3976 	return X86EMUL_CONTINUE;
3977 }
3978 
3979 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3980 {
3981 	/* emulating clflush regardless of cpuid */
3982 	return X86EMUL_CONTINUE;
3983 }
3984 
3985 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
3986 {
3987 	/* emulating clflushopt regardless of cpuid */
3988 	return X86EMUL_CONTINUE;
3989 }
3990 
3991 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3992 {
3993 	ctxt->dst.val = (s32) ctxt->src.val;
3994 	return X86EMUL_CONTINUE;
3995 }
3996 
3997 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3998 {
3999 	if (!ctxt->ops->guest_has_fxsr(ctxt))
4000 		return emulate_ud(ctxt);
4001 
4002 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
4003 		return emulate_nm(ctxt);
4004 
4005 	/*
4006 	 * Don't emulate a case that should never be hit, instead of working
4007 	 * around a lack of fxsave64/fxrstor64 on old compilers.
4008 	 */
4009 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
4010 		return X86EMUL_UNHANDLEABLE;
4011 
4012 	return X86EMUL_CONTINUE;
4013 }
4014 
4015 /*
4016  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
4017  * and restore MXCSR.
4018  */
4019 static size_t __fxstate_size(int nregs)
4020 {
4021 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
4022 }
4023 
4024 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
4025 {
4026 	bool cr4_osfxsr;
4027 	if (ctxt->mode == X86EMUL_MODE_PROT64)
4028 		return __fxstate_size(16);
4029 
4030 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
4031 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
4032 }
4033 
4034 /*
4035  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
4036  *  1) 16 bit mode
4037  *  2) 32 bit mode
4038  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
4039  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
4040  *       save and restore
4041  *  3) 64-bit mode with REX.W prefix
4042  *     - like (2), but XMM 8-15 are being saved and restored
4043  *  4) 64-bit mode without REX.W prefix
4044  *     - like (3), but FIP and FDP are 64 bit
4045  *
4046  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
4047  * desired result.  (4) is not emulated.
4048  *
4049  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
4050  * and FPU DS) should match.
4051  */
4052 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
4053 {
4054 	struct fxregs_state fx_state;
4055 	int rc;
4056 
4057 	rc = check_fxsr(ctxt);
4058 	if (rc != X86EMUL_CONTINUE)
4059 		return rc;
4060 
4061 	kvm_fpu_get();
4062 
4063 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
4064 
4065 	kvm_fpu_put();
4066 
4067 	if (rc != X86EMUL_CONTINUE)
4068 		return rc;
4069 
4070 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
4071 		                   fxstate_size(ctxt));
4072 }
4073 
4074 /*
4075  * FXRSTOR might restore XMM registers not provided by the guest. Fill
4076  * in the host registers (via FXSAVE) instead, so they won't be modified.
4077  * (preemption has to stay disabled until FXRSTOR).
4078  *
4079  * Use noinline to keep the stack for other functions called by callers small.
4080  */
4081 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
4082 				 const size_t used_size)
4083 {
4084 	struct fxregs_state fx_tmp;
4085 	int rc;
4086 
4087 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
4088 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
4089 	       __fxstate_size(16) - used_size);
4090 
4091 	return rc;
4092 }
4093 
4094 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
4095 {
4096 	struct fxregs_state fx_state;
4097 	int rc;
4098 	size_t size;
4099 
4100 	rc = check_fxsr(ctxt);
4101 	if (rc != X86EMUL_CONTINUE)
4102 		return rc;
4103 
4104 	size = fxstate_size(ctxt);
4105 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
4106 	if (rc != X86EMUL_CONTINUE)
4107 		return rc;
4108 
4109 	kvm_fpu_get();
4110 
4111 	if (size < __fxstate_size(16)) {
4112 		rc = fxregs_fixup(&fx_state, size);
4113 		if (rc != X86EMUL_CONTINUE)
4114 			goto out;
4115 	}
4116 
4117 	if (fx_state.mxcsr >> 16) {
4118 		rc = emulate_gp(ctxt, 0);
4119 		goto out;
4120 	}
4121 
4122 	if (rc == X86EMUL_CONTINUE)
4123 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
4124 
4125 out:
4126 	kvm_fpu_put();
4127 
4128 	return rc;
4129 }
4130 
4131 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
4132 {
4133 	u32 eax, ecx, edx;
4134 
4135 	eax = reg_read(ctxt, VCPU_REGS_RAX);
4136 	edx = reg_read(ctxt, VCPU_REGS_RDX);
4137 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
4138 
4139 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
4140 		return emulate_gp(ctxt, 0);
4141 
4142 	return X86EMUL_CONTINUE;
4143 }
4144 
4145 static bool valid_cr(int nr)
4146 {
4147 	switch (nr) {
4148 	case 0:
4149 	case 2 ... 4:
4150 	case 8:
4151 		return true;
4152 	default:
4153 		return false;
4154 	}
4155 }
4156 
4157 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
4158 {
4159 	if (!valid_cr(ctxt->modrm_reg))
4160 		return emulate_ud(ctxt);
4161 
4162 	return X86EMUL_CONTINUE;
4163 }
4164 
4165 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt)
4166 {
4167 	unsigned long dr7;
4168 
4169 	ctxt->ops->get_dr(ctxt, 7, &dr7);
4170 
4171 	/* Check if DR7.Global_Enable is set */
4172 	return dr7 & (1 << 13);
4173 }
4174 
4175 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
4176 {
4177 	int dr = ctxt->modrm_reg;
4178 	u64 cr4;
4179 
4180 	if (dr > 7)
4181 		return emulate_ud(ctxt);
4182 
4183 	cr4 = ctxt->ops->get_cr(ctxt, 4);
4184 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
4185 		return emulate_ud(ctxt);
4186 
4187 	if (check_dr7_gd(ctxt)) {
4188 		ulong dr6;
4189 
4190 		ctxt->ops->get_dr(ctxt, 6, &dr6);
4191 		dr6 &= ~DR_TRAP_BITS;
4192 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
4193 		ctxt->ops->set_dr(ctxt, 6, dr6);
4194 		return emulate_db(ctxt);
4195 	}
4196 
4197 	return X86EMUL_CONTINUE;
4198 }
4199 
4200 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
4201 {
4202 	u64 new_val = ctxt->src.val64;
4203 	int dr = ctxt->modrm_reg;
4204 
4205 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
4206 		return emulate_gp(ctxt, 0);
4207 
4208 	return check_dr_read(ctxt);
4209 }
4210 
4211 static int check_svme(struct x86_emulate_ctxt *ctxt)
4212 {
4213 	u64 efer = 0;
4214 
4215 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
4216 
4217 	if (!(efer & EFER_SVME))
4218 		return emulate_ud(ctxt);
4219 
4220 	return X86EMUL_CONTINUE;
4221 }
4222 
4223 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
4224 {
4225 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
4226 
4227 	/* Valid physical address? */
4228 	if (rax & 0xffff000000000000ULL)
4229 		return emulate_gp(ctxt, 0);
4230 
4231 	return check_svme(ctxt);
4232 }
4233 
4234 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
4235 {
4236 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4237 
4238 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
4239 		return emulate_gp(ctxt, 0);
4240 
4241 	return X86EMUL_CONTINUE;
4242 }
4243 
4244 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
4245 {
4246 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
4247 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
4248 
4249 	/*
4250 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
4251 	 * in Ring3 when CR4.PCE=0.
4252 	 */
4253 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
4254 		return X86EMUL_CONTINUE;
4255 
4256 	/*
4257 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
4258 	 * check however is unnecessary because CPL is always 0 outside
4259 	 * protected mode.
4260 	 */
4261 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
4262 	    ctxt->ops->check_pmc(ctxt, rcx))
4263 		return emulate_gp(ctxt, 0);
4264 
4265 	return X86EMUL_CONTINUE;
4266 }
4267 
4268 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
4269 {
4270 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
4271 	if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes))
4272 		return emulate_gp(ctxt, 0);
4273 
4274 	return X86EMUL_CONTINUE;
4275 }
4276 
4277 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
4278 {
4279 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
4280 	if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes))
4281 		return emulate_gp(ctxt, 0);
4282 
4283 	return X86EMUL_CONTINUE;
4284 }
4285 
4286 #define D(_y) { .flags = (_y) }
4287 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
4288 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
4289 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
4290 #define N    D(NotImpl)
4291 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
4292 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
4293 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
4294 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
4295 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
4296 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
4297 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
4298 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
4299 #define II(_f, _e, _i) \
4300 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
4301 #define IIP(_f, _e, _i, _p) \
4302 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
4303 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4304 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4305 
4306 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4307 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4308 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4309 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4310 #define I2bvIP(_f, _e, _i, _p) \
4311 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4312 
4313 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4314 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4315 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4316 
4317 static const struct opcode group7_rm0[] = {
4318 	N,
4319 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4320 	N, N, N, N, N, N,
4321 };
4322 
4323 static const struct opcode group7_rm1[] = {
4324 	DI(SrcNone | Priv, monitor),
4325 	DI(SrcNone | Priv, mwait),
4326 	N, N, N, N, N, N,
4327 };
4328 
4329 static const struct opcode group7_rm2[] = {
4330 	N,
4331 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4332 	N, N, N, N, N, N,
4333 };
4334 
4335 static const struct opcode group7_rm3[] = {
4336 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4337 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4338 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4339 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4340 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4341 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4342 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4343 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4344 };
4345 
4346 static const struct opcode group7_rm7[] = {
4347 	N,
4348 	DIP(SrcNone, rdtscp, check_rdtsc),
4349 	N, N, N, N, N, N,
4350 };
4351 
4352 static const struct opcode group1[] = {
4353 	F(Lock, em_add),
4354 	F(Lock | PageTable, em_or),
4355 	F(Lock, em_adc),
4356 	F(Lock, em_sbb),
4357 	F(Lock | PageTable, em_and),
4358 	F(Lock, em_sub),
4359 	F(Lock, em_xor),
4360 	F(NoWrite, em_cmp),
4361 };
4362 
4363 static const struct opcode group1A[] = {
4364 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4365 };
4366 
4367 static const struct opcode group2[] = {
4368 	F(DstMem | ModRM, em_rol),
4369 	F(DstMem | ModRM, em_ror),
4370 	F(DstMem | ModRM, em_rcl),
4371 	F(DstMem | ModRM, em_rcr),
4372 	F(DstMem | ModRM, em_shl),
4373 	F(DstMem | ModRM, em_shr),
4374 	F(DstMem | ModRM, em_shl),
4375 	F(DstMem | ModRM, em_sar),
4376 };
4377 
4378 static const struct opcode group3[] = {
4379 	F(DstMem | SrcImm | NoWrite, em_test),
4380 	F(DstMem | SrcImm | NoWrite, em_test),
4381 	F(DstMem | SrcNone | Lock, em_not),
4382 	F(DstMem | SrcNone | Lock, em_neg),
4383 	F(DstXacc | Src2Mem, em_mul_ex),
4384 	F(DstXacc | Src2Mem, em_imul_ex),
4385 	F(DstXacc | Src2Mem, em_div_ex),
4386 	F(DstXacc | Src2Mem, em_idiv_ex),
4387 };
4388 
4389 static const struct opcode group4[] = {
4390 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4391 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4392 	N, N, N, N, N, N,
4393 };
4394 
4395 static const struct opcode group5[] = {
4396 	F(DstMem | SrcNone | Lock,		em_inc),
4397 	F(DstMem | SrcNone | Lock,		em_dec),
4398 	I(SrcMem | NearBranch | IsBranch,       em_call_near_abs),
4399 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_call_far),
4400 	I(SrcMem | NearBranch | IsBranch,       em_jmp_abs),
4401 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_jmp_far),
4402 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4403 };
4404 
4405 static const struct opcode group6[] = {
4406 	II(Prot | DstMem,	   em_sldt, sldt),
4407 	II(Prot | DstMem,	   em_str, str),
4408 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4409 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4410 	N, N, N, N,
4411 };
4412 
4413 static const struct group_dual group7 = { {
4414 	II(Mov | DstMem,			em_sgdt, sgdt),
4415 	II(Mov | DstMem,			em_sidt, sidt),
4416 	II(SrcMem | Priv,			em_lgdt, lgdt),
4417 	II(SrcMem | Priv,			em_lidt, lidt),
4418 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4419 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4420 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4421 }, {
4422 	EXT(0, group7_rm0),
4423 	EXT(0, group7_rm1),
4424 	EXT(0, group7_rm2),
4425 	EXT(0, group7_rm3),
4426 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4427 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4428 	EXT(0, group7_rm7),
4429 } };
4430 
4431 static const struct opcode group8[] = {
4432 	N, N, N, N,
4433 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4434 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4435 	F(DstMem | SrcImmByte | Lock,			em_btr),
4436 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4437 };
4438 
4439 /*
4440  * The "memory" destination is actually always a register, since we come
4441  * from the register case of group9.
4442  */
4443 static const struct gprefix pfx_0f_c7_7 = {
4444 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4445 };
4446 
4447 
4448 static const struct group_dual group9 = { {
4449 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4450 }, {
4451 	N, N, N, N, N, N, N,
4452 	GP(0, &pfx_0f_c7_7),
4453 } };
4454 
4455 static const struct opcode group11[] = {
4456 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4457 	X7(D(Undefined)),
4458 };
4459 
4460 static const struct gprefix pfx_0f_ae_7 = {
4461 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4462 };
4463 
4464 static const struct group_dual group15 = { {
4465 	I(ModRM | Aligned16, em_fxsave),
4466 	I(ModRM | Aligned16, em_fxrstor),
4467 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4468 }, {
4469 	N, N, N, N, N, N, N, N,
4470 } };
4471 
4472 static const struct gprefix pfx_0f_6f_0f_7f = {
4473 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4474 };
4475 
4476 static const struct instr_dual instr_dual_0f_2b = {
4477 	I(0, em_mov), N
4478 };
4479 
4480 static const struct gprefix pfx_0f_2b = {
4481 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4482 };
4483 
4484 static const struct gprefix pfx_0f_10_0f_11 = {
4485 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4486 };
4487 
4488 static const struct gprefix pfx_0f_28_0f_29 = {
4489 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4490 };
4491 
4492 static const struct gprefix pfx_0f_e7 = {
4493 	N, I(Sse, em_mov), N, N,
4494 };
4495 
4496 static const struct escape escape_d9 = { {
4497 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4498 }, {
4499 	/* 0xC0 - 0xC7 */
4500 	N, N, N, N, N, N, N, N,
4501 	/* 0xC8 - 0xCF */
4502 	N, N, N, N, N, N, N, N,
4503 	/* 0xD0 - 0xC7 */
4504 	N, N, N, N, N, N, N, N,
4505 	/* 0xD8 - 0xDF */
4506 	N, N, N, N, N, N, N, N,
4507 	/* 0xE0 - 0xE7 */
4508 	N, N, N, N, N, N, N, N,
4509 	/* 0xE8 - 0xEF */
4510 	N, N, N, N, N, N, N, N,
4511 	/* 0xF0 - 0xF7 */
4512 	N, N, N, N, N, N, N, N,
4513 	/* 0xF8 - 0xFF */
4514 	N, N, N, N, N, N, N, N,
4515 } };
4516 
4517 static const struct escape escape_db = { {
4518 	N, N, N, N, N, N, N, N,
4519 }, {
4520 	/* 0xC0 - 0xC7 */
4521 	N, N, N, N, N, N, N, N,
4522 	/* 0xC8 - 0xCF */
4523 	N, N, N, N, N, N, N, N,
4524 	/* 0xD0 - 0xC7 */
4525 	N, N, N, N, N, N, N, N,
4526 	/* 0xD8 - 0xDF */
4527 	N, N, N, N, N, N, N, N,
4528 	/* 0xE0 - 0xE7 */
4529 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4530 	/* 0xE8 - 0xEF */
4531 	N, N, N, N, N, N, N, N,
4532 	/* 0xF0 - 0xF7 */
4533 	N, N, N, N, N, N, N, N,
4534 	/* 0xF8 - 0xFF */
4535 	N, N, N, N, N, N, N, N,
4536 } };
4537 
4538 static const struct escape escape_dd = { {
4539 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4540 }, {
4541 	/* 0xC0 - 0xC7 */
4542 	N, N, N, N, N, N, N, N,
4543 	/* 0xC8 - 0xCF */
4544 	N, N, N, N, N, N, N, N,
4545 	/* 0xD0 - 0xC7 */
4546 	N, N, N, N, N, N, N, N,
4547 	/* 0xD8 - 0xDF */
4548 	N, N, N, N, N, N, N, N,
4549 	/* 0xE0 - 0xE7 */
4550 	N, N, N, N, N, N, N, N,
4551 	/* 0xE8 - 0xEF */
4552 	N, N, N, N, N, N, N, N,
4553 	/* 0xF0 - 0xF7 */
4554 	N, N, N, N, N, N, N, N,
4555 	/* 0xF8 - 0xFF */
4556 	N, N, N, N, N, N, N, N,
4557 } };
4558 
4559 static const struct instr_dual instr_dual_0f_c3 = {
4560 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4561 };
4562 
4563 static const struct mode_dual mode_dual_63 = {
4564 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4565 };
4566 
4567 static const struct instr_dual instr_dual_8d = {
4568 	D(DstReg | SrcMem | ModRM | NoAccess), N
4569 };
4570 
4571 static const struct opcode opcode_table[256] = {
4572 	/* 0x00 - 0x07 */
4573 	F6ALU(Lock, em_add),
4574 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4575 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4576 	/* 0x08 - 0x0F */
4577 	F6ALU(Lock | PageTable, em_or),
4578 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4579 	N,
4580 	/* 0x10 - 0x17 */
4581 	F6ALU(Lock, em_adc),
4582 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4583 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4584 	/* 0x18 - 0x1F */
4585 	F6ALU(Lock, em_sbb),
4586 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4587 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4588 	/* 0x20 - 0x27 */
4589 	F6ALU(Lock | PageTable, em_and), N, N,
4590 	/* 0x28 - 0x2F */
4591 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4592 	/* 0x30 - 0x37 */
4593 	F6ALU(Lock, em_xor), N, N,
4594 	/* 0x38 - 0x3F */
4595 	F6ALU(NoWrite, em_cmp), N, N,
4596 	/* 0x40 - 0x4F */
4597 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4598 	/* 0x50 - 0x57 */
4599 	X8(I(SrcReg | Stack, em_push)),
4600 	/* 0x58 - 0x5F */
4601 	X8(I(DstReg | Stack, em_pop)),
4602 	/* 0x60 - 0x67 */
4603 	I(ImplicitOps | Stack | No64, em_pusha),
4604 	I(ImplicitOps | Stack | No64, em_popa),
4605 	N, MD(ModRM, &mode_dual_63),
4606 	N, N, N, N,
4607 	/* 0x68 - 0x6F */
4608 	I(SrcImm | Mov | Stack, em_push),
4609 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4610 	I(SrcImmByte | Mov | Stack, em_push),
4611 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4612 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4613 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4614 	/* 0x70 - 0x7F */
4615 	X16(D(SrcImmByte | NearBranch | IsBranch)),
4616 	/* 0x80 - 0x87 */
4617 	G(ByteOp | DstMem | SrcImm, group1),
4618 	G(DstMem | SrcImm, group1),
4619 	G(ByteOp | DstMem | SrcImm | No64, group1),
4620 	G(DstMem | SrcImmByte, group1),
4621 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4622 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4623 	/* 0x88 - 0x8F */
4624 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4625 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4626 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4627 	ID(0, &instr_dual_8d),
4628 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4629 	G(0, group1A),
4630 	/* 0x90 - 0x97 */
4631 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4632 	/* 0x98 - 0x9F */
4633 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4634 	I(SrcImmFAddr | No64 | IsBranch, em_call_far), N,
4635 	II(ImplicitOps | Stack, em_pushf, pushf),
4636 	II(ImplicitOps | Stack, em_popf, popf),
4637 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4638 	/* 0xA0 - 0xA7 */
4639 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4640 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4641 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4642 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4643 	/* 0xA8 - 0xAF */
4644 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4645 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4646 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4647 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4648 	/* 0xB0 - 0xB7 */
4649 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4650 	/* 0xB8 - 0xBF */
4651 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4652 	/* 0xC0 - 0xC7 */
4653 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4654 	I(ImplicitOps | NearBranch | SrcImmU16 | IsBranch, em_ret_near_imm),
4655 	I(ImplicitOps | NearBranch | IsBranch, em_ret),
4656 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4657 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4658 	G(ByteOp, group11), G(0, group11),
4659 	/* 0xC8 - 0xCF */
4660 	I(Stack | SrcImmU16 | Src2ImmByte | IsBranch, em_enter),
4661 	I(Stack | IsBranch, em_leave),
4662 	I(ImplicitOps | SrcImmU16 | IsBranch, em_ret_far_imm),
4663 	I(ImplicitOps | IsBranch, em_ret_far),
4664 	D(ImplicitOps | IsBranch), DI(SrcImmByte | IsBranch, intn),
4665 	D(ImplicitOps | No64 | IsBranch),
4666 	II(ImplicitOps | IsBranch, em_iret, iret),
4667 	/* 0xD0 - 0xD7 */
4668 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4669 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4670 	I(DstAcc | SrcImmUByte | No64, em_aam),
4671 	I(DstAcc | SrcImmUByte | No64, em_aad),
4672 	F(DstAcc | ByteOp | No64, em_salc),
4673 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4674 	/* 0xD8 - 0xDF */
4675 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4676 	/* 0xE0 - 0xE7 */
4677 	X3(I(SrcImmByte | NearBranch | IsBranch, em_loop)),
4678 	I(SrcImmByte | NearBranch | IsBranch, em_jcxz),
4679 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4680 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4681 	/* 0xE8 - 0xEF */
4682 	I(SrcImm | NearBranch | IsBranch, em_call),
4683 	D(SrcImm | ImplicitOps | NearBranch | IsBranch),
4684 	I(SrcImmFAddr | No64 | IsBranch, em_jmp_far),
4685 	D(SrcImmByte | ImplicitOps | NearBranch | IsBranch),
4686 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4687 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4688 	/* 0xF0 - 0xF7 */
4689 	N, DI(ImplicitOps, icebp), N, N,
4690 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4691 	G(ByteOp, group3), G(0, group3),
4692 	/* 0xF8 - 0xFF */
4693 	D(ImplicitOps), D(ImplicitOps),
4694 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4695 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4696 };
4697 
4698 static const struct opcode twobyte_table[256] = {
4699 	/* 0x00 - 0x0F */
4700 	G(0, group6), GD(0, &group7), N, N,
4701 	N, I(ImplicitOps | EmulateOnUD | IsBranch, em_syscall),
4702 	II(ImplicitOps | Priv, em_clts, clts), N,
4703 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4704 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4705 	/* 0x10 - 0x1F */
4706 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4707 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4708 	N, N, N, N, N, N,
4709 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4710 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4711 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4712 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4713 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4714 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4715 	/* 0x20 - 0x2F */
4716 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4717 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4718 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4719 						check_cr_access),
4720 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4721 						check_dr_write),
4722 	N, N, N, N,
4723 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4724 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4725 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4726 	N, N, N, N,
4727 	/* 0x30 - 0x3F */
4728 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4729 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4730 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4731 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4732 	I(ImplicitOps | EmulateOnUD | IsBranch, em_sysenter),
4733 	I(ImplicitOps | Priv | EmulateOnUD | IsBranch, em_sysexit),
4734 	N, N,
4735 	N, N, N, N, N, N, N, N,
4736 	/* 0x40 - 0x4F */
4737 	X16(D(DstReg | SrcMem | ModRM)),
4738 	/* 0x50 - 0x5F */
4739 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4740 	/* 0x60 - 0x6F */
4741 	N, N, N, N,
4742 	N, N, N, N,
4743 	N, N, N, N,
4744 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4745 	/* 0x70 - 0x7F */
4746 	N, N, N, N,
4747 	N, N, N, N,
4748 	N, N, N, N,
4749 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4750 	/* 0x80 - 0x8F */
4751 	X16(D(SrcImm | NearBranch | IsBranch)),
4752 	/* 0x90 - 0x9F */
4753 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4754 	/* 0xA0 - 0xA7 */
4755 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4756 	II(ImplicitOps, em_cpuid, cpuid),
4757 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4758 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4759 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4760 	/* 0xA8 - 0xAF */
4761 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4762 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4763 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4764 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4765 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4766 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4767 	/* 0xB0 - 0xB7 */
4768 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4769 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4770 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4771 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4772 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4773 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4774 	/* 0xB8 - 0xBF */
4775 	N, N,
4776 	G(BitOp, group8),
4777 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4778 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4779 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4780 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4781 	/* 0xC0 - 0xC7 */
4782 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4783 	N, ID(0, &instr_dual_0f_c3),
4784 	N, N, N, GD(0, &group9),
4785 	/* 0xC8 - 0xCF */
4786 	X8(I(DstReg, em_bswap)),
4787 	/* 0xD0 - 0xDF */
4788 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4789 	/* 0xE0 - 0xEF */
4790 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4791 	N, N, N, N, N, N, N, N,
4792 	/* 0xF0 - 0xFF */
4793 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4794 };
4795 
4796 static const struct instr_dual instr_dual_0f_38_f0 = {
4797 	I(DstReg | SrcMem | Mov, em_movbe), N
4798 };
4799 
4800 static const struct instr_dual instr_dual_0f_38_f1 = {
4801 	I(DstMem | SrcReg | Mov, em_movbe), N
4802 };
4803 
4804 static const struct gprefix three_byte_0f_38_f0 = {
4805 	ID(0, &instr_dual_0f_38_f0), N, N, N
4806 };
4807 
4808 static const struct gprefix three_byte_0f_38_f1 = {
4809 	ID(0, &instr_dual_0f_38_f1), N, N, N
4810 };
4811 
4812 /*
4813  * Insns below are selected by the prefix which indexed by the third opcode
4814  * byte.
4815  */
4816 static const struct opcode opcode_map_0f_38[256] = {
4817 	/* 0x00 - 0x7f */
4818 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4819 	/* 0x80 - 0xef */
4820 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4821 	/* 0xf0 - 0xf1 */
4822 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4823 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4824 	/* 0xf2 - 0xff */
4825 	N, N, X4(N), X8(N)
4826 };
4827 
4828 #undef D
4829 #undef N
4830 #undef G
4831 #undef GD
4832 #undef I
4833 #undef GP
4834 #undef EXT
4835 #undef MD
4836 #undef ID
4837 
4838 #undef D2bv
4839 #undef D2bvIP
4840 #undef I2bv
4841 #undef I2bvIP
4842 #undef I6ALU
4843 
4844 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4845 {
4846 	unsigned size;
4847 
4848 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4849 	if (size == 8)
4850 		size = 4;
4851 	return size;
4852 }
4853 
4854 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4855 		      unsigned size, bool sign_extension)
4856 {
4857 	int rc = X86EMUL_CONTINUE;
4858 
4859 	op->type = OP_IMM;
4860 	op->bytes = size;
4861 	op->addr.mem.ea = ctxt->_eip;
4862 	/* NB. Immediates are sign-extended as necessary. */
4863 	switch (op->bytes) {
4864 	case 1:
4865 		op->val = insn_fetch(s8, ctxt);
4866 		break;
4867 	case 2:
4868 		op->val = insn_fetch(s16, ctxt);
4869 		break;
4870 	case 4:
4871 		op->val = insn_fetch(s32, ctxt);
4872 		break;
4873 	case 8:
4874 		op->val = insn_fetch(s64, ctxt);
4875 		break;
4876 	}
4877 	if (!sign_extension) {
4878 		switch (op->bytes) {
4879 		case 1:
4880 			op->val &= 0xff;
4881 			break;
4882 		case 2:
4883 			op->val &= 0xffff;
4884 			break;
4885 		case 4:
4886 			op->val &= 0xffffffff;
4887 			break;
4888 		}
4889 	}
4890 done:
4891 	return rc;
4892 }
4893 
4894 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4895 			  unsigned d)
4896 {
4897 	int rc = X86EMUL_CONTINUE;
4898 
4899 	switch (d) {
4900 	case OpReg:
4901 		decode_register_operand(ctxt, op);
4902 		break;
4903 	case OpImmUByte:
4904 		rc = decode_imm(ctxt, op, 1, false);
4905 		break;
4906 	case OpMem:
4907 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4908 	mem_common:
4909 		*op = ctxt->memop;
4910 		ctxt->memopp = op;
4911 		if (ctxt->d & BitOp)
4912 			fetch_bit_operand(ctxt);
4913 		op->orig_val = op->val;
4914 		break;
4915 	case OpMem64:
4916 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4917 		goto mem_common;
4918 	case OpAcc:
4919 		op->type = OP_REG;
4920 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4921 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4922 		fetch_register_operand(op);
4923 		op->orig_val = op->val;
4924 		break;
4925 	case OpAccLo:
4926 		op->type = OP_REG;
4927 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4928 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4929 		fetch_register_operand(op);
4930 		op->orig_val = op->val;
4931 		break;
4932 	case OpAccHi:
4933 		if (ctxt->d & ByteOp) {
4934 			op->type = OP_NONE;
4935 			break;
4936 		}
4937 		op->type = OP_REG;
4938 		op->bytes = ctxt->op_bytes;
4939 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4940 		fetch_register_operand(op);
4941 		op->orig_val = op->val;
4942 		break;
4943 	case OpDI:
4944 		op->type = OP_MEM;
4945 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4946 		op->addr.mem.ea =
4947 			register_address(ctxt, VCPU_REGS_RDI);
4948 		op->addr.mem.seg = VCPU_SREG_ES;
4949 		op->val = 0;
4950 		op->count = 1;
4951 		break;
4952 	case OpDX:
4953 		op->type = OP_REG;
4954 		op->bytes = 2;
4955 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4956 		fetch_register_operand(op);
4957 		break;
4958 	case OpCL:
4959 		op->type = OP_IMM;
4960 		op->bytes = 1;
4961 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4962 		break;
4963 	case OpImmByte:
4964 		rc = decode_imm(ctxt, op, 1, true);
4965 		break;
4966 	case OpOne:
4967 		op->type = OP_IMM;
4968 		op->bytes = 1;
4969 		op->val = 1;
4970 		break;
4971 	case OpImm:
4972 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4973 		break;
4974 	case OpImm64:
4975 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4976 		break;
4977 	case OpMem8:
4978 		ctxt->memop.bytes = 1;
4979 		if (ctxt->memop.type == OP_REG) {
4980 			ctxt->memop.addr.reg = decode_register(ctxt,
4981 					ctxt->modrm_rm, true);
4982 			fetch_register_operand(&ctxt->memop);
4983 		}
4984 		goto mem_common;
4985 	case OpMem16:
4986 		ctxt->memop.bytes = 2;
4987 		goto mem_common;
4988 	case OpMem32:
4989 		ctxt->memop.bytes = 4;
4990 		goto mem_common;
4991 	case OpImmU16:
4992 		rc = decode_imm(ctxt, op, 2, false);
4993 		break;
4994 	case OpImmU:
4995 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4996 		break;
4997 	case OpSI:
4998 		op->type = OP_MEM;
4999 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5000 		op->addr.mem.ea =
5001 			register_address(ctxt, VCPU_REGS_RSI);
5002 		op->addr.mem.seg = ctxt->seg_override;
5003 		op->val = 0;
5004 		op->count = 1;
5005 		break;
5006 	case OpXLat:
5007 		op->type = OP_MEM;
5008 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
5009 		op->addr.mem.ea =
5010 			address_mask(ctxt,
5011 				reg_read(ctxt, VCPU_REGS_RBX) +
5012 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
5013 		op->addr.mem.seg = ctxt->seg_override;
5014 		op->val = 0;
5015 		break;
5016 	case OpImmFAddr:
5017 		op->type = OP_IMM;
5018 		op->addr.mem.ea = ctxt->_eip;
5019 		op->bytes = ctxt->op_bytes + 2;
5020 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
5021 		break;
5022 	case OpMemFAddr:
5023 		ctxt->memop.bytes = ctxt->op_bytes + 2;
5024 		goto mem_common;
5025 	case OpES:
5026 		op->type = OP_IMM;
5027 		op->val = VCPU_SREG_ES;
5028 		break;
5029 	case OpCS:
5030 		op->type = OP_IMM;
5031 		op->val = VCPU_SREG_CS;
5032 		break;
5033 	case OpSS:
5034 		op->type = OP_IMM;
5035 		op->val = VCPU_SREG_SS;
5036 		break;
5037 	case OpDS:
5038 		op->type = OP_IMM;
5039 		op->val = VCPU_SREG_DS;
5040 		break;
5041 	case OpFS:
5042 		op->type = OP_IMM;
5043 		op->val = VCPU_SREG_FS;
5044 		break;
5045 	case OpGS:
5046 		op->type = OP_IMM;
5047 		op->val = VCPU_SREG_GS;
5048 		break;
5049 	case OpImplicit:
5050 		/* Special instructions do their own operand decoding. */
5051 	default:
5052 		op->type = OP_NONE; /* Disable writeback. */
5053 		break;
5054 	}
5055 
5056 done:
5057 	return rc;
5058 }
5059 
5060 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
5061 {
5062 	int rc = X86EMUL_CONTINUE;
5063 	int mode = ctxt->mode;
5064 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
5065 	bool op_prefix = false;
5066 	bool has_seg_override = false;
5067 	struct opcode opcode;
5068 	u16 dummy;
5069 	struct desc_struct desc;
5070 
5071 	ctxt->memop.type = OP_NONE;
5072 	ctxt->memopp = NULL;
5073 	ctxt->_eip = ctxt->eip;
5074 	ctxt->fetch.ptr = ctxt->fetch.data;
5075 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
5076 	ctxt->opcode_len = 1;
5077 	ctxt->intercept = x86_intercept_none;
5078 	if (insn_len > 0)
5079 		memcpy(ctxt->fetch.data, insn, insn_len);
5080 	else {
5081 		rc = __do_insn_fetch_bytes(ctxt, 1);
5082 		if (rc != X86EMUL_CONTINUE)
5083 			goto done;
5084 	}
5085 
5086 	switch (mode) {
5087 	case X86EMUL_MODE_REAL:
5088 	case X86EMUL_MODE_VM86:
5089 		def_op_bytes = def_ad_bytes = 2;
5090 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
5091 		if (desc.d)
5092 			def_op_bytes = def_ad_bytes = 4;
5093 		break;
5094 	case X86EMUL_MODE_PROT16:
5095 		def_op_bytes = def_ad_bytes = 2;
5096 		break;
5097 	case X86EMUL_MODE_PROT32:
5098 		def_op_bytes = def_ad_bytes = 4;
5099 		break;
5100 #ifdef CONFIG_X86_64
5101 	case X86EMUL_MODE_PROT64:
5102 		def_op_bytes = 4;
5103 		def_ad_bytes = 8;
5104 		break;
5105 #endif
5106 	default:
5107 		return EMULATION_FAILED;
5108 	}
5109 
5110 	ctxt->op_bytes = def_op_bytes;
5111 	ctxt->ad_bytes = def_ad_bytes;
5112 
5113 	/* Legacy prefixes. */
5114 	for (;;) {
5115 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
5116 		case 0x66:	/* operand-size override */
5117 			op_prefix = true;
5118 			/* switch between 2/4 bytes */
5119 			ctxt->op_bytes = def_op_bytes ^ 6;
5120 			break;
5121 		case 0x67:	/* address-size override */
5122 			if (mode == X86EMUL_MODE_PROT64)
5123 				/* switch between 4/8 bytes */
5124 				ctxt->ad_bytes = def_ad_bytes ^ 12;
5125 			else
5126 				/* switch between 2/4 bytes */
5127 				ctxt->ad_bytes = def_ad_bytes ^ 6;
5128 			break;
5129 		case 0x26:	/* ES override */
5130 			has_seg_override = true;
5131 			ctxt->seg_override = VCPU_SREG_ES;
5132 			break;
5133 		case 0x2e:	/* CS override */
5134 			has_seg_override = true;
5135 			ctxt->seg_override = VCPU_SREG_CS;
5136 			break;
5137 		case 0x36:	/* SS override */
5138 			has_seg_override = true;
5139 			ctxt->seg_override = VCPU_SREG_SS;
5140 			break;
5141 		case 0x3e:	/* DS override */
5142 			has_seg_override = true;
5143 			ctxt->seg_override = VCPU_SREG_DS;
5144 			break;
5145 		case 0x64:	/* FS override */
5146 			has_seg_override = true;
5147 			ctxt->seg_override = VCPU_SREG_FS;
5148 			break;
5149 		case 0x65:	/* GS override */
5150 			has_seg_override = true;
5151 			ctxt->seg_override = VCPU_SREG_GS;
5152 			break;
5153 		case 0x40 ... 0x4f: /* REX */
5154 			if (mode != X86EMUL_MODE_PROT64)
5155 				goto done_prefixes;
5156 			ctxt->rex_prefix = ctxt->b;
5157 			continue;
5158 		case 0xf0:	/* LOCK */
5159 			ctxt->lock_prefix = 1;
5160 			break;
5161 		case 0xf2:	/* REPNE/REPNZ */
5162 		case 0xf3:	/* REP/REPE/REPZ */
5163 			ctxt->rep_prefix = ctxt->b;
5164 			break;
5165 		default:
5166 			goto done_prefixes;
5167 		}
5168 
5169 		/* Any legacy prefix after a REX prefix nullifies its effect. */
5170 
5171 		ctxt->rex_prefix = 0;
5172 	}
5173 
5174 done_prefixes:
5175 
5176 	/* REX prefix. */
5177 	if (ctxt->rex_prefix & 8)
5178 		ctxt->op_bytes = 8;	/* REX.W */
5179 
5180 	/* Opcode byte(s). */
5181 	opcode = opcode_table[ctxt->b];
5182 	/* Two-byte opcode? */
5183 	if (ctxt->b == 0x0f) {
5184 		ctxt->opcode_len = 2;
5185 		ctxt->b = insn_fetch(u8, ctxt);
5186 		opcode = twobyte_table[ctxt->b];
5187 
5188 		/* 0F_38 opcode map */
5189 		if (ctxt->b == 0x38) {
5190 			ctxt->opcode_len = 3;
5191 			ctxt->b = insn_fetch(u8, ctxt);
5192 			opcode = opcode_map_0f_38[ctxt->b];
5193 		}
5194 	}
5195 	ctxt->d = opcode.flags;
5196 
5197 	if (ctxt->d & ModRM)
5198 		ctxt->modrm = insn_fetch(u8, ctxt);
5199 
5200 	/* vex-prefix instructions are not implemented */
5201 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
5202 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
5203 		ctxt->d = NotImpl;
5204 	}
5205 
5206 	while (ctxt->d & GroupMask) {
5207 		switch (ctxt->d & GroupMask) {
5208 		case Group:
5209 			goffset = (ctxt->modrm >> 3) & 7;
5210 			opcode = opcode.u.group[goffset];
5211 			break;
5212 		case GroupDual:
5213 			goffset = (ctxt->modrm >> 3) & 7;
5214 			if ((ctxt->modrm >> 6) == 3)
5215 				opcode = opcode.u.gdual->mod3[goffset];
5216 			else
5217 				opcode = opcode.u.gdual->mod012[goffset];
5218 			break;
5219 		case RMExt:
5220 			goffset = ctxt->modrm & 7;
5221 			opcode = opcode.u.group[goffset];
5222 			break;
5223 		case Prefix:
5224 			if (ctxt->rep_prefix && op_prefix)
5225 				return EMULATION_FAILED;
5226 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
5227 			switch (simd_prefix) {
5228 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
5229 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
5230 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
5231 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
5232 			}
5233 			break;
5234 		case Escape:
5235 			if (ctxt->modrm > 0xbf) {
5236 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
5237 				u32 index = array_index_nospec(
5238 					ctxt->modrm - 0xc0, size);
5239 
5240 				opcode = opcode.u.esc->high[index];
5241 			} else {
5242 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
5243 			}
5244 			break;
5245 		case InstrDual:
5246 			if ((ctxt->modrm >> 6) == 3)
5247 				opcode = opcode.u.idual->mod3;
5248 			else
5249 				opcode = opcode.u.idual->mod012;
5250 			break;
5251 		case ModeDual:
5252 			if (ctxt->mode == X86EMUL_MODE_PROT64)
5253 				opcode = opcode.u.mdual->mode64;
5254 			else
5255 				opcode = opcode.u.mdual->mode32;
5256 			break;
5257 		default:
5258 			return EMULATION_FAILED;
5259 		}
5260 
5261 		ctxt->d &= ~(u64)GroupMask;
5262 		ctxt->d |= opcode.flags;
5263 	}
5264 
5265 	ctxt->is_branch = opcode.flags & IsBranch;
5266 
5267 	/* Unrecognised? */
5268 	if (ctxt->d == 0)
5269 		return EMULATION_FAILED;
5270 
5271 	ctxt->execute = opcode.u.execute;
5272 
5273 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
5274 	    likely(!(ctxt->d & EmulateOnUD)))
5275 		return EMULATION_FAILED;
5276 
5277 	if (unlikely(ctxt->d &
5278 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
5279 	     No16))) {
5280 		/*
5281 		 * These are copied unconditionally here, and checked unconditionally
5282 		 * in x86_emulate_insn.
5283 		 */
5284 		ctxt->check_perm = opcode.check_perm;
5285 		ctxt->intercept = opcode.intercept;
5286 
5287 		if (ctxt->d & NotImpl)
5288 			return EMULATION_FAILED;
5289 
5290 		if (mode == X86EMUL_MODE_PROT64) {
5291 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
5292 				ctxt->op_bytes = 8;
5293 			else if (ctxt->d & NearBranch)
5294 				ctxt->op_bytes = 8;
5295 		}
5296 
5297 		if (ctxt->d & Op3264) {
5298 			if (mode == X86EMUL_MODE_PROT64)
5299 				ctxt->op_bytes = 8;
5300 			else
5301 				ctxt->op_bytes = 4;
5302 		}
5303 
5304 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5305 			ctxt->op_bytes = 4;
5306 
5307 		if (ctxt->d & Sse)
5308 			ctxt->op_bytes = 16;
5309 		else if (ctxt->d & Mmx)
5310 			ctxt->op_bytes = 8;
5311 	}
5312 
5313 	/* ModRM and SIB bytes. */
5314 	if (ctxt->d & ModRM) {
5315 		rc = decode_modrm(ctxt, &ctxt->memop);
5316 		if (!has_seg_override) {
5317 			has_seg_override = true;
5318 			ctxt->seg_override = ctxt->modrm_seg;
5319 		}
5320 	} else if (ctxt->d & MemAbs)
5321 		rc = decode_abs(ctxt, &ctxt->memop);
5322 	if (rc != X86EMUL_CONTINUE)
5323 		goto done;
5324 
5325 	if (!has_seg_override)
5326 		ctxt->seg_override = VCPU_SREG_DS;
5327 
5328 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5329 
5330 	/*
5331 	 * Decode and fetch the source operand: register, memory
5332 	 * or immediate.
5333 	 */
5334 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5335 	if (rc != X86EMUL_CONTINUE)
5336 		goto done;
5337 
5338 	/*
5339 	 * Decode and fetch the second source operand: register, memory
5340 	 * or immediate.
5341 	 */
5342 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5343 	if (rc != X86EMUL_CONTINUE)
5344 		goto done;
5345 
5346 	/* Decode and fetch the destination operand: register or memory. */
5347 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5348 
5349 	if (ctxt->rip_relative && likely(ctxt->memopp))
5350 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5351 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5352 
5353 done:
5354 	if (rc == X86EMUL_PROPAGATE_FAULT)
5355 		ctxt->have_exception = true;
5356 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5357 }
5358 
5359 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5360 {
5361 	return ctxt->d & PageTable;
5362 }
5363 
5364 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5365 {
5366 	/* The second termination condition only applies for REPE
5367 	 * and REPNE. Test if the repeat string operation prefix is
5368 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5369 	 * corresponding termination condition according to:
5370 	 * 	- if REPE/REPZ and ZF = 0 then done
5371 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5372 	 */
5373 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5374 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5375 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5376 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5377 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5378 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5379 		return true;
5380 
5381 	return false;
5382 }
5383 
5384 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5385 {
5386 	int rc;
5387 
5388 	kvm_fpu_get();
5389 	rc = asm_safe("fwait");
5390 	kvm_fpu_put();
5391 
5392 	if (unlikely(rc != X86EMUL_CONTINUE))
5393 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5394 
5395 	return X86EMUL_CONTINUE;
5396 }
5397 
5398 static void fetch_possible_mmx_operand(struct operand *op)
5399 {
5400 	if (op->type == OP_MM)
5401 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5402 }
5403 
5404 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5405 {
5406 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5407 
5408 	if (!(ctxt->d & ByteOp))
5409 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5410 
5411 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5412 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5413 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5414 	    : "c"(ctxt->src2.val));
5415 
5416 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5417 	if (!fop) /* exception is returned in fop variable */
5418 		return emulate_de(ctxt);
5419 	return X86EMUL_CONTINUE;
5420 }
5421 
5422 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5423 {
5424 	/* Clear fields that are set conditionally but read without a guard. */
5425 	ctxt->rip_relative = false;
5426 	ctxt->rex_prefix = 0;
5427 	ctxt->lock_prefix = 0;
5428 	ctxt->rep_prefix = 0;
5429 	ctxt->regs_valid = 0;
5430 	ctxt->regs_dirty = 0;
5431 
5432 	ctxt->io_read.pos = 0;
5433 	ctxt->io_read.end = 0;
5434 	ctxt->mem_read.end = 0;
5435 }
5436 
5437 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5438 {
5439 	const struct x86_emulate_ops *ops = ctxt->ops;
5440 	int rc = X86EMUL_CONTINUE;
5441 	int saved_dst_type = ctxt->dst.type;
5442 	unsigned emul_flags;
5443 
5444 	ctxt->mem_read.pos = 0;
5445 
5446 	/* LOCK prefix is allowed only with some instructions */
5447 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5448 		rc = emulate_ud(ctxt);
5449 		goto done;
5450 	}
5451 
5452 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5453 		rc = emulate_ud(ctxt);
5454 		goto done;
5455 	}
5456 
5457 	emul_flags = ctxt->ops->get_hflags(ctxt);
5458 	if (unlikely(ctxt->d &
5459 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5460 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5461 				(ctxt->d & Undefined)) {
5462 			rc = emulate_ud(ctxt);
5463 			goto done;
5464 		}
5465 
5466 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5467 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5468 			rc = emulate_ud(ctxt);
5469 			goto done;
5470 		}
5471 
5472 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5473 			rc = emulate_nm(ctxt);
5474 			goto done;
5475 		}
5476 
5477 		if (ctxt->d & Mmx) {
5478 			rc = flush_pending_x87_faults(ctxt);
5479 			if (rc != X86EMUL_CONTINUE)
5480 				goto done;
5481 			/*
5482 			 * Now that we know the fpu is exception safe, we can fetch
5483 			 * operands from it.
5484 			 */
5485 			fetch_possible_mmx_operand(&ctxt->src);
5486 			fetch_possible_mmx_operand(&ctxt->src2);
5487 			if (!(ctxt->d & Mov))
5488 				fetch_possible_mmx_operand(&ctxt->dst);
5489 		}
5490 
5491 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) {
5492 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5493 						      X86_ICPT_PRE_EXCEPT);
5494 			if (rc != X86EMUL_CONTINUE)
5495 				goto done;
5496 		}
5497 
5498 		/* Instruction can only be executed in protected mode */
5499 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5500 			rc = emulate_ud(ctxt);
5501 			goto done;
5502 		}
5503 
5504 		/* Privileged instruction can be executed only in CPL=0 */
5505 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5506 			if (ctxt->d & PrivUD)
5507 				rc = emulate_ud(ctxt);
5508 			else
5509 				rc = emulate_gp(ctxt, 0);
5510 			goto done;
5511 		}
5512 
5513 		/* Do instruction specific permission checks */
5514 		if (ctxt->d & CheckPerm) {
5515 			rc = ctxt->check_perm(ctxt);
5516 			if (rc != X86EMUL_CONTINUE)
5517 				goto done;
5518 		}
5519 
5520 		if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5521 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5522 						      X86_ICPT_POST_EXCEPT);
5523 			if (rc != X86EMUL_CONTINUE)
5524 				goto done;
5525 		}
5526 
5527 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5528 			/* All REP prefixes have the same first termination condition */
5529 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5530 				string_registers_quirk(ctxt);
5531 				ctxt->eip = ctxt->_eip;
5532 				ctxt->eflags &= ~X86_EFLAGS_RF;
5533 				goto done;
5534 			}
5535 		}
5536 	}
5537 
5538 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5539 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5540 				    ctxt->src.valptr, ctxt->src.bytes);
5541 		if (rc != X86EMUL_CONTINUE)
5542 			goto done;
5543 		ctxt->src.orig_val64 = ctxt->src.val64;
5544 	}
5545 
5546 	if (ctxt->src2.type == OP_MEM) {
5547 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5548 				    &ctxt->src2.val, ctxt->src2.bytes);
5549 		if (rc != X86EMUL_CONTINUE)
5550 			goto done;
5551 	}
5552 
5553 	if ((ctxt->d & DstMask) == ImplicitOps)
5554 		goto special_insn;
5555 
5556 
5557 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5558 		/* optimisation - avoid slow emulated read if Mov */
5559 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5560 				   &ctxt->dst.val, ctxt->dst.bytes);
5561 		if (rc != X86EMUL_CONTINUE) {
5562 			if (!(ctxt->d & NoWrite) &&
5563 			    rc == X86EMUL_PROPAGATE_FAULT &&
5564 			    ctxt->exception.vector == PF_VECTOR)
5565 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5566 			goto done;
5567 		}
5568 	}
5569 	/* Copy full 64-bit value for CMPXCHG8B.  */
5570 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5571 
5572 special_insn:
5573 
5574 	if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) {
5575 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5576 					      X86_ICPT_POST_MEMACCESS);
5577 		if (rc != X86EMUL_CONTINUE)
5578 			goto done;
5579 	}
5580 
5581 	if (ctxt->rep_prefix && (ctxt->d & String))
5582 		ctxt->eflags |= X86_EFLAGS_RF;
5583 	else
5584 		ctxt->eflags &= ~X86_EFLAGS_RF;
5585 
5586 	if (ctxt->execute) {
5587 		if (ctxt->d & Fastop)
5588 			rc = fastop(ctxt, ctxt->fop);
5589 		else
5590 			rc = ctxt->execute(ctxt);
5591 		if (rc != X86EMUL_CONTINUE)
5592 			goto done;
5593 		goto writeback;
5594 	}
5595 
5596 	if (ctxt->opcode_len == 2)
5597 		goto twobyte_insn;
5598 	else if (ctxt->opcode_len == 3)
5599 		goto threebyte_insn;
5600 
5601 	switch (ctxt->b) {
5602 	case 0x70 ... 0x7f: /* jcc (short) */
5603 		if (test_cc(ctxt->b, ctxt->eflags))
5604 			rc = jmp_rel(ctxt, ctxt->src.val);
5605 		break;
5606 	case 0x8d: /* lea r16/r32, m */
5607 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5608 		break;
5609 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5610 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5611 			ctxt->dst.type = OP_NONE;
5612 		else
5613 			rc = em_xchg(ctxt);
5614 		break;
5615 	case 0x98: /* cbw/cwde/cdqe */
5616 		switch (ctxt->op_bytes) {
5617 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5618 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5619 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5620 		}
5621 		break;
5622 	case 0xcc:		/* int3 */
5623 		rc = emulate_int(ctxt, 3);
5624 		break;
5625 	case 0xcd:		/* int n */
5626 		rc = emulate_int(ctxt, ctxt->src.val);
5627 		break;
5628 	case 0xce:		/* into */
5629 		if (ctxt->eflags & X86_EFLAGS_OF)
5630 			rc = emulate_int(ctxt, 4);
5631 		break;
5632 	case 0xe9: /* jmp rel */
5633 	case 0xeb: /* jmp rel short */
5634 		rc = jmp_rel(ctxt, ctxt->src.val);
5635 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5636 		break;
5637 	case 0xf4:              /* hlt */
5638 		ctxt->ops->halt(ctxt);
5639 		break;
5640 	case 0xf5:	/* cmc */
5641 		/* complement carry flag from eflags reg */
5642 		ctxt->eflags ^= X86_EFLAGS_CF;
5643 		break;
5644 	case 0xf8: /* clc */
5645 		ctxt->eflags &= ~X86_EFLAGS_CF;
5646 		break;
5647 	case 0xf9: /* stc */
5648 		ctxt->eflags |= X86_EFLAGS_CF;
5649 		break;
5650 	case 0xfc: /* cld */
5651 		ctxt->eflags &= ~X86_EFLAGS_DF;
5652 		break;
5653 	case 0xfd: /* std */
5654 		ctxt->eflags |= X86_EFLAGS_DF;
5655 		break;
5656 	default:
5657 		goto cannot_emulate;
5658 	}
5659 
5660 	if (rc != X86EMUL_CONTINUE)
5661 		goto done;
5662 
5663 writeback:
5664 	if (ctxt->d & SrcWrite) {
5665 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5666 		rc = writeback(ctxt, &ctxt->src);
5667 		if (rc != X86EMUL_CONTINUE)
5668 			goto done;
5669 	}
5670 	if (!(ctxt->d & NoWrite)) {
5671 		rc = writeback(ctxt, &ctxt->dst);
5672 		if (rc != X86EMUL_CONTINUE)
5673 			goto done;
5674 	}
5675 
5676 	/*
5677 	 * restore dst type in case the decoding will be reused
5678 	 * (happens for string instruction )
5679 	 */
5680 	ctxt->dst.type = saved_dst_type;
5681 
5682 	if ((ctxt->d & SrcMask) == SrcSI)
5683 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5684 
5685 	if ((ctxt->d & DstMask) == DstDI)
5686 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5687 
5688 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5689 		unsigned int count;
5690 		struct read_cache *r = &ctxt->io_read;
5691 		if ((ctxt->d & SrcMask) == SrcSI)
5692 			count = ctxt->src.count;
5693 		else
5694 			count = ctxt->dst.count;
5695 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5696 
5697 		if (!string_insn_completed(ctxt)) {
5698 			/*
5699 			 * Re-enter guest when pio read ahead buffer is empty
5700 			 * or, if it is not used, after each 1024 iteration.
5701 			 */
5702 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5703 			    (r->end == 0 || r->end != r->pos)) {
5704 				/*
5705 				 * Reset read cache. Usually happens before
5706 				 * decode, but since instruction is restarted
5707 				 * we have to do it here.
5708 				 */
5709 				ctxt->mem_read.end = 0;
5710 				writeback_registers(ctxt);
5711 				return EMULATION_RESTART;
5712 			}
5713 			goto done; /* skip rip writeback */
5714 		}
5715 		ctxt->eflags &= ~X86_EFLAGS_RF;
5716 	}
5717 
5718 	ctxt->eip = ctxt->_eip;
5719 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5720 		ctxt->eip = (u32)ctxt->_eip;
5721 
5722 done:
5723 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5724 		if (KVM_EMULATOR_BUG_ON(ctxt->exception.vector > 0x1f, ctxt))
5725 			return EMULATION_FAILED;
5726 		ctxt->have_exception = true;
5727 	}
5728 	if (rc == X86EMUL_INTERCEPTED)
5729 		return EMULATION_INTERCEPTED;
5730 
5731 	if (rc == X86EMUL_CONTINUE)
5732 		writeback_registers(ctxt);
5733 
5734 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5735 
5736 twobyte_insn:
5737 	switch (ctxt->b) {
5738 	case 0x09:		/* wbinvd */
5739 		(ctxt->ops->wbinvd)(ctxt);
5740 		break;
5741 	case 0x08:		/* invd */
5742 	case 0x0d:		/* GrpP (prefetch) */
5743 	case 0x18:		/* Grp16 (prefetch/nop) */
5744 	case 0x1f:		/* nop */
5745 		break;
5746 	case 0x20: /* mov cr, reg */
5747 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5748 		break;
5749 	case 0x21: /* mov from dr to reg */
5750 		ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
5751 		break;
5752 	case 0x40 ... 0x4f:	/* cmov */
5753 		if (test_cc(ctxt->b, ctxt->eflags))
5754 			ctxt->dst.val = ctxt->src.val;
5755 		else if (ctxt->op_bytes != 4)
5756 			ctxt->dst.type = OP_NONE; /* no writeback */
5757 		break;
5758 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5759 		if (test_cc(ctxt->b, ctxt->eflags))
5760 			rc = jmp_rel(ctxt, ctxt->src.val);
5761 		break;
5762 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5763 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5764 		break;
5765 	case 0xb6 ... 0xb7:	/* movzx */
5766 		ctxt->dst.bytes = ctxt->op_bytes;
5767 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5768 						       : (u16) ctxt->src.val;
5769 		break;
5770 	case 0xbe ... 0xbf:	/* movsx */
5771 		ctxt->dst.bytes = ctxt->op_bytes;
5772 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5773 							(s16) ctxt->src.val;
5774 		break;
5775 	default:
5776 		goto cannot_emulate;
5777 	}
5778 
5779 threebyte_insn:
5780 
5781 	if (rc != X86EMUL_CONTINUE)
5782 		goto done;
5783 
5784 	goto writeback;
5785 
5786 cannot_emulate:
5787 	return EMULATION_FAILED;
5788 }
5789 
5790 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5791 {
5792 	invalidate_registers(ctxt);
5793 }
5794 
5795 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5796 {
5797 	writeback_registers(ctxt);
5798 }
5799 
5800 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5801 {
5802 	if (ctxt->rep_prefix && (ctxt->d & String))
5803 		return false;
5804 
5805 	if (ctxt->d & TwoMemOp)
5806 		return false;
5807 
5808 	return true;
5809 }
5810