xref: /linux/arch/x86/kvm/emulate.c (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include <linux/kvm_host.h>
23 #include "kvm_cache_regs.h"
24 #include "kvm_emulate.h"
25 #include <linux/stringify.h>
26 #include <asm/debugreg.h>
27 #include <asm/nospec-branch.h>
28 #include <asm/ibt.h>
29 
30 #include "x86.h"
31 #include "tss.h"
32 #include "mmu.h"
33 #include "pmu.h"
34 
35 /*
36  * Operand types
37  */
38 #define OpNone             0ull
39 #define OpImplicit         1ull  /* No generic decode */
40 #define OpReg              2ull  /* Register */
41 #define OpMem              3ull  /* Memory */
42 #define OpAcc              4ull  /* Accumulator: AL/AX/EAX/RAX */
43 #define OpDI               5ull  /* ES:DI/EDI/RDI */
44 #define OpMem64            6ull  /* Memory, 64-bit */
45 #define OpImmUByte         7ull  /* Zero-extended 8-bit immediate */
46 #define OpDX               8ull  /* DX register */
47 #define OpCL               9ull  /* CL register (for shifts) */
48 #define OpImmByte         10ull  /* 8-bit sign extended immediate */
49 #define OpOne             11ull  /* Implied 1 */
50 #define OpImm             12ull  /* Sign extended up to 32-bit immediate */
51 #define OpMem16           13ull  /* Memory operand (16-bit). */
52 #define OpMem32           14ull  /* Memory operand (32-bit). */
53 #define OpImmU            15ull  /* Immediate operand, zero extended */
54 #define OpSI              16ull  /* SI/ESI/RSI */
55 #define OpImmFAddr        17ull  /* Immediate far address */
56 #define OpMemFAddr        18ull  /* Far address in memory */
57 #define OpImmU16          19ull  /* Immediate operand, 16 bits, zero extended */
58 #define OpES              20ull  /* ES */
59 #define OpCS              21ull  /* CS */
60 #define OpSS              22ull  /* SS */
61 #define OpDS              23ull  /* DS */
62 #define OpFS              24ull  /* FS */
63 #define OpGS              25ull  /* GS */
64 #define OpMem8            26ull  /* 8-bit zero extended memory operand */
65 #define OpImm64           27ull  /* Sign extended 16/32/64-bit immediate */
66 #define OpXLat            28ull  /* memory at BX/EBX/RBX + zero-extended AL */
67 #define OpAccLo           29ull  /* Low part of extended acc (AX/AX/EAX/RAX) */
68 #define OpAccHi           30ull  /* High part of extended acc (-/DX/EDX/RDX) */
69 
70 #define OpBits             5  /* Width of operand field */
71 #define OpMask             ((1ull << OpBits) - 1)
72 
73 /*
74  * Opcode effective-address decode tables.
75  * Note that we only emulate instructions that have at least one memory
76  * operand (excluding implicit stack references). We assume that stack
77  * references and instruction fetches will never occur in special memory
78  * areas that require emulation. So, for example, 'mov <imm>,<reg>' need
79  * not be handled.
80  */
81 
82 /* Operand sizes: 8-bit operands or specified/overridden size. */
83 #define ByteOp      (1<<0)	/* 8-bit operands. */
84 /* Destination operand type. */
85 #define DstShift    1
86 #define ImplicitOps (OpImplicit << DstShift)
87 #define DstReg      (OpReg << DstShift)
88 #define DstMem      (OpMem << DstShift)
89 #define DstAcc      (OpAcc << DstShift)
90 #define DstDI       (OpDI << DstShift)
91 #define DstMem64    (OpMem64 << DstShift)
92 #define DstMem16    (OpMem16 << DstShift)
93 #define DstImmUByte (OpImmUByte << DstShift)
94 #define DstDX       (OpDX << DstShift)
95 #define DstAccLo    (OpAccLo << DstShift)
96 #define DstMask     (OpMask << DstShift)
97 /* Source operand type. */
98 #define SrcShift    6
99 #define SrcNone     (OpNone << SrcShift)
100 #define SrcReg      (OpReg << SrcShift)
101 #define SrcMem      (OpMem << SrcShift)
102 #define SrcMem16    (OpMem16 << SrcShift)
103 #define SrcMem32    (OpMem32 << SrcShift)
104 #define SrcImm      (OpImm << SrcShift)
105 #define SrcImmByte  (OpImmByte << SrcShift)
106 #define SrcOne      (OpOne << SrcShift)
107 #define SrcImmUByte (OpImmUByte << SrcShift)
108 #define SrcImmU     (OpImmU << SrcShift)
109 #define SrcSI       (OpSI << SrcShift)
110 #define SrcXLat     (OpXLat << SrcShift)
111 #define SrcImmFAddr (OpImmFAddr << SrcShift)
112 #define SrcMemFAddr (OpMemFAddr << SrcShift)
113 #define SrcAcc      (OpAcc << SrcShift)
114 #define SrcImmU16   (OpImmU16 << SrcShift)
115 #define SrcImm64    (OpImm64 << SrcShift)
116 #define SrcDX       (OpDX << SrcShift)
117 #define SrcMem8     (OpMem8 << SrcShift)
118 #define SrcAccHi    (OpAccHi << SrcShift)
119 #define SrcMask     (OpMask << SrcShift)
120 #define BitOp       (1<<11)
121 #define MemAbs      (1<<12)      /* Memory operand is absolute displacement */
122 #define String      (1<<13)     /* String instruction (rep capable) */
123 #define Stack       (1<<14)     /* Stack instruction (push/pop) */
124 #define GroupMask   (7<<15)     /* Opcode uses one of the group mechanisms */
125 #define Group       (1<<15)     /* Bits 3:5 of modrm byte extend opcode */
126 #define GroupDual   (2<<15)     /* Alternate decoding of mod == 3 */
127 #define Prefix      (3<<15)     /* Instruction varies with 66/f2/f3 prefix */
128 #define RMExt       (4<<15)     /* Opcode extension in ModRM r/m if mod == 3 */
129 #define Escape      (5<<15)     /* Escape to coprocessor instruction */
130 #define InstrDual   (6<<15)     /* Alternate instruction decoding of mod == 3 */
131 #define ModeDual    (7<<15)     /* Different instruction for 32/64 bit */
132 #define Sse         (1<<18)     /* SSE Vector instruction */
133 /* Generic ModRM decode. */
134 #define ModRM       (1<<19)
135 /* Destination is only written; never read. */
136 #define Mov         (1<<20)
137 /* Misc flags */
138 #define Prot        (1<<21) /* instruction generates #UD if not in prot-mode */
139 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */
140 #define NoAccess    (1<<23) /* Don't access memory (lea/invlpg/verr etc) */
141 #define Op3264      (1<<24) /* Operand is 64b in long mode, 32b otherwise */
142 #define Undefined   (1<<25) /* No Such Instruction */
143 #define Lock        (1<<26) /* lock prefix is allowed for the instruction */
144 #define Priv        (1<<27) /* instruction generates #GP if current CPL != 0 */
145 #define No64	    (1<<28)
146 #define PageTable   (1 << 29)   /* instruction used to write page table */
147 #define NotImpl     (1 << 30)   /* instruction is not implemented */
148 /* Source 2 operand type */
149 #define Src2Shift   (31)
150 #define Src2None    (OpNone << Src2Shift)
151 #define Src2Mem     (OpMem << Src2Shift)
152 #define Src2CL      (OpCL << Src2Shift)
153 #define Src2ImmByte (OpImmByte << Src2Shift)
154 #define Src2One     (OpOne << Src2Shift)
155 #define Src2Imm     (OpImm << Src2Shift)
156 #define Src2ES      (OpES << Src2Shift)
157 #define Src2CS      (OpCS << Src2Shift)
158 #define Src2SS      (OpSS << Src2Shift)
159 #define Src2DS      (OpDS << Src2Shift)
160 #define Src2FS      (OpFS << Src2Shift)
161 #define Src2GS      (OpGS << Src2Shift)
162 #define Src2Mask    (OpMask << Src2Shift)
163 #define Mmx         ((u64)1 << 40)  /* MMX Vector instruction */
164 #define AlignMask   ((u64)7 << 41)
165 #define Aligned     ((u64)1 << 41)  /* Explicitly aligned (e.g. MOVDQA) */
166 #define Unaligned   ((u64)2 << 41)  /* Explicitly unaligned (e.g. MOVDQU) */
167 #define Avx         ((u64)3 << 41)  /* Advanced Vector Extensions */
168 #define Aligned16   ((u64)4 << 41)  /* Aligned to 16 byte boundary (e.g. FXSAVE) */
169 #define Fastop      ((u64)1 << 44)  /* Use opcode::u.fastop */
170 #define NoWrite     ((u64)1 << 45)  /* No writeback */
171 #define SrcWrite    ((u64)1 << 46)  /* Write back src operand */
172 #define NoMod	    ((u64)1 << 47)  /* Mod field is ignored */
173 #define Intercept   ((u64)1 << 48)  /* Has valid intercept field */
174 #define CheckPerm   ((u64)1 << 49)  /* Has valid check_perm field */
175 #define PrivUD      ((u64)1 << 51)  /* #UD instead of #GP on CPL > 0 */
176 #define NearBranch  ((u64)1 << 52)  /* Near branches */
177 #define No16	    ((u64)1 << 53)  /* No 16 bit operand */
178 #define IncSP       ((u64)1 << 54)  /* SP is incremented before ModRM calc */
179 #define TwoMemOp    ((u64)1 << 55)  /* Instruction has two memory operand */
180 #define IsBranch    ((u64)1 << 56)  /* Instruction is considered a branch. */
181 
182 #define DstXacc     (DstAccLo | SrcAccHi | SrcWrite)
183 
184 #define X2(x...) x, x
185 #define X3(x...) X2(x), x
186 #define X4(x...) X2(x), X2(x)
187 #define X5(x...) X4(x), x
188 #define X6(x...) X4(x), X2(x)
189 #define X7(x...) X4(x), X3(x)
190 #define X8(x...) X4(x), X4(x)
191 #define X16(x...) X8(x), X8(x)
192 
193 struct opcode {
194 	u64 flags;
195 	u8 intercept;
196 	u8 pad[7];
197 	union {
198 		int (*execute)(struct x86_emulate_ctxt *ctxt);
199 		const struct opcode *group;
200 		const struct group_dual *gdual;
201 		const struct gprefix *gprefix;
202 		const struct escape *esc;
203 		const struct instr_dual *idual;
204 		const struct mode_dual *mdual;
205 		void (*fastop)(struct fastop *fake);
206 	} u;
207 	int (*check_perm)(struct x86_emulate_ctxt *ctxt);
208 };
209 
210 struct group_dual {
211 	struct opcode mod012[8];
212 	struct opcode mod3[8];
213 };
214 
215 struct gprefix {
216 	struct opcode pfx_no;
217 	struct opcode pfx_66;
218 	struct opcode pfx_f2;
219 	struct opcode pfx_f3;
220 };
221 
222 struct escape {
223 	struct opcode op[8];
224 	struct opcode high[64];
225 };
226 
227 struct instr_dual {
228 	struct opcode mod012;
229 	struct opcode mod3;
230 };
231 
232 struct mode_dual {
233 	struct opcode mode32;
234 	struct opcode mode64;
235 };
236 
237 #define EFLG_RESERVED_ZEROS_MASK 0xffc0802a
238 
239 enum x86_transfer_type {
240 	X86_TRANSFER_NONE,
241 	X86_TRANSFER_CALL_JMP,
242 	X86_TRANSFER_RET,
243 	X86_TRANSFER_TASK_SWITCH,
244 };
245 
246 static void writeback_registers(struct x86_emulate_ctxt *ctxt)
247 {
248 	unsigned long dirty = ctxt->regs_dirty;
249 	unsigned reg;
250 
251 	for_each_set_bit(reg, &dirty, NR_EMULATOR_GPRS)
252 		ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]);
253 }
254 
255 static void invalidate_registers(struct x86_emulate_ctxt *ctxt)
256 {
257 	ctxt->regs_dirty = 0;
258 	ctxt->regs_valid = 0;
259 }
260 
261 /*
262  * These EFLAGS bits are restored from saved value during emulation, and
263  * any changes are written back to the saved value after emulation.
264  */
265 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\
266 		     X86_EFLAGS_PF|X86_EFLAGS_CF)
267 
268 #ifdef CONFIG_X86_64
269 #define ON64(x) x
270 #else
271 #define ON64(x)
272 #endif
273 
274 /*
275  * fastop functions have a special calling convention:
276  *
277  * dst:    rax        (in/out)
278  * src:    rdx        (in/out)
279  * src2:   rcx        (in)
280  * flags:  rflags     (in/out)
281  * ex:     rsi        (in:fastop pointer, out:zero if exception)
282  *
283  * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for
284  * different operand sizes can be reached by calculation, rather than a jump
285  * table (which would be bigger than the code).
286  *
287  * The 16 byte alignment, considering 5 bytes for the RET thunk, 3 for ENDBR
288  * and 1 for the straight line speculation INT3, leaves 7 bytes for the
289  * body of the function.  Currently none is larger than 4.
290  */
291 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
292 
293 #define FASTOP_SIZE	16
294 
295 #define __FOP_FUNC(name) \
296 	".align " __stringify(FASTOP_SIZE) " \n\t" \
297 	".type " name ", @function \n\t" \
298 	name ":\n\t" \
299 	ASM_ENDBR \
300 	IBT_NOSEAL(name)
301 
302 #define FOP_FUNC(name) \
303 	__FOP_FUNC(#name)
304 
305 #define __FOP_RET(name) \
306 	"11: " ASM_RET \
307 	".size " name ", .-" name "\n\t"
308 
309 #define FOP_RET(name) \
310 	__FOP_RET(#name)
311 
312 #define __FOP_START(op, align) \
313 	extern void em_##op(struct fastop *fake); \
314 	asm(".pushsection .text, \"ax\" \n\t" \
315 	    ".global em_" #op " \n\t" \
316 	    ".align " __stringify(align) " \n\t" \
317 	    "em_" #op ":\n\t"
318 
319 #define FOP_START(op) __FOP_START(op, FASTOP_SIZE)
320 
321 #define FOP_END \
322 	    ".popsection")
323 
324 #define __FOPNOP(name) \
325 	__FOP_FUNC(name) \
326 	__FOP_RET(name)
327 
328 #define FOPNOP() \
329 	__FOPNOP(__stringify(__UNIQUE_ID(nop)))
330 
331 #define FOP1E(op,  dst) \
332 	__FOP_FUNC(#op "_" #dst) \
333 	"10: " #op " %" #dst " \n\t" \
334 	__FOP_RET(#op "_" #dst)
335 
336 #define FOP1EEX(op,  dst) \
337 	FOP1E(op, dst) _ASM_EXTABLE_TYPE_REG(10b, 11b, EX_TYPE_ZERO_REG, %%esi)
338 
339 #define FASTOP1(op) \
340 	FOP_START(op) \
341 	FOP1E(op##b, al) \
342 	FOP1E(op##w, ax) \
343 	FOP1E(op##l, eax) \
344 	ON64(FOP1E(op##q, rax))	\
345 	FOP_END
346 
347 /* 1-operand, using src2 (for MUL/DIV r/m) */
348 #define FASTOP1SRC2(op, name) \
349 	FOP_START(name) \
350 	FOP1E(op, cl) \
351 	FOP1E(op, cx) \
352 	FOP1E(op, ecx) \
353 	ON64(FOP1E(op, rcx)) \
354 	FOP_END
355 
356 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */
357 #define FASTOP1SRC2EX(op, name) \
358 	FOP_START(name) \
359 	FOP1EEX(op, cl) \
360 	FOP1EEX(op, cx) \
361 	FOP1EEX(op, ecx) \
362 	ON64(FOP1EEX(op, rcx)) \
363 	FOP_END
364 
365 #define FOP2E(op,  dst, src)	   \
366 	__FOP_FUNC(#op "_" #dst "_" #src) \
367 	#op " %" #src ", %" #dst " \n\t" \
368 	__FOP_RET(#op "_" #dst "_" #src)
369 
370 #define FASTOP2(op) \
371 	FOP_START(op) \
372 	FOP2E(op##b, al, dl) \
373 	FOP2E(op##w, ax, dx) \
374 	FOP2E(op##l, eax, edx) \
375 	ON64(FOP2E(op##q, rax, rdx)) \
376 	FOP_END
377 
378 /* 2 operand, word only */
379 #define FASTOP2W(op) \
380 	FOP_START(op) \
381 	FOPNOP() \
382 	FOP2E(op##w, ax, dx) \
383 	FOP2E(op##l, eax, edx) \
384 	ON64(FOP2E(op##q, rax, rdx)) \
385 	FOP_END
386 
387 /* 2 operand, src is CL */
388 #define FASTOP2CL(op) \
389 	FOP_START(op) \
390 	FOP2E(op##b, al, cl) \
391 	FOP2E(op##w, ax, cl) \
392 	FOP2E(op##l, eax, cl) \
393 	ON64(FOP2E(op##q, rax, cl)) \
394 	FOP_END
395 
396 /* 2 operand, src and dest are reversed */
397 #define FASTOP2R(op, name) \
398 	FOP_START(name) \
399 	FOP2E(op##b, dl, al) \
400 	FOP2E(op##w, dx, ax) \
401 	FOP2E(op##l, edx, eax) \
402 	ON64(FOP2E(op##q, rdx, rax)) \
403 	FOP_END
404 
405 #define FOP3E(op,  dst, src, src2) \
406 	__FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \
407 	#op " %" #src2 ", %" #src ", %" #dst " \n\t"\
408 	__FOP_RET(#op "_" #dst "_" #src "_" #src2)
409 
410 /* 3-operand, word-only, src2=cl */
411 #define FASTOP3WCL(op) \
412 	FOP_START(op) \
413 	FOPNOP() \
414 	FOP3E(op##w, ax, dx, cl) \
415 	FOP3E(op##l, eax, edx, cl) \
416 	ON64(FOP3E(op##q, rax, rdx, cl)) \
417 	FOP_END
418 
419 /* Special case for SETcc - 1 instruction per cc */
420 #define FOP_SETCC(op) \
421 	FOP_FUNC(op) \
422 	#op " %al \n\t" \
423 	FOP_RET(op)
424 
425 FOP_START(setcc)
426 FOP_SETCC(seto)
427 FOP_SETCC(setno)
428 FOP_SETCC(setc)
429 FOP_SETCC(setnc)
430 FOP_SETCC(setz)
431 FOP_SETCC(setnz)
432 FOP_SETCC(setbe)
433 FOP_SETCC(setnbe)
434 FOP_SETCC(sets)
435 FOP_SETCC(setns)
436 FOP_SETCC(setp)
437 FOP_SETCC(setnp)
438 FOP_SETCC(setl)
439 FOP_SETCC(setnl)
440 FOP_SETCC(setle)
441 FOP_SETCC(setnle)
442 FOP_END;
443 
444 FOP_START(salc)
445 FOP_FUNC(salc)
446 "pushf; sbb %al, %al; popf \n\t"
447 FOP_RET(salc)
448 FOP_END;
449 
450 /*
451  * XXX: inoutclob user must know where the argument is being expanded.
452  *      Using asm goto would allow us to remove _fault.
453  */
454 #define asm_safe(insn, inoutclob...) \
455 ({ \
456 	int _fault = 0; \
457  \
458 	asm volatile("1:" insn "\n" \
459 	             "2:\n" \
460 		     _ASM_EXTABLE_TYPE_REG(1b, 2b, EX_TYPE_ONE_REG, %[_fault]) \
461 	             : [_fault] "+r"(_fault) inoutclob ); \
462  \
463 	_fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \
464 })
465 
466 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt,
467 				    enum x86_intercept intercept,
468 				    enum x86_intercept_stage stage)
469 {
470 	struct x86_instruction_info info = {
471 		.intercept  = intercept,
472 		.rep_prefix = ctxt->rep_prefix,
473 		.modrm_mod  = ctxt->modrm_mod,
474 		.modrm_reg  = ctxt->modrm_reg,
475 		.modrm_rm   = ctxt->modrm_rm,
476 		.src_val    = ctxt->src.val64,
477 		.dst_val    = ctxt->dst.val64,
478 		.src_bytes  = ctxt->src.bytes,
479 		.dst_bytes  = ctxt->dst.bytes,
480 		.ad_bytes   = ctxt->ad_bytes,
481 		.next_rip   = ctxt->eip,
482 	};
483 
484 	return ctxt->ops->intercept(ctxt, &info, stage);
485 }
486 
487 static void assign_masked(ulong *dest, ulong src, ulong mask)
488 {
489 	*dest = (*dest & ~mask) | (src & mask);
490 }
491 
492 static void assign_register(unsigned long *reg, u64 val, int bytes)
493 {
494 	/* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */
495 	switch (bytes) {
496 	case 1:
497 		*(u8 *)reg = (u8)val;
498 		break;
499 	case 2:
500 		*(u16 *)reg = (u16)val;
501 		break;
502 	case 4:
503 		*reg = (u32)val;
504 		break;	/* 64b: zero-extend */
505 	case 8:
506 		*reg = val;
507 		break;
508 	}
509 }
510 
511 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt)
512 {
513 	return (1UL << (ctxt->ad_bytes << 3)) - 1;
514 }
515 
516 static ulong stack_mask(struct x86_emulate_ctxt *ctxt)
517 {
518 	u16 sel;
519 	struct desc_struct ss;
520 
521 	if (ctxt->mode == X86EMUL_MODE_PROT64)
522 		return ~0UL;
523 	ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS);
524 	return ~0U >> ((ss.d ^ 1) * 16);  /* d=0: 0xffff; d=1: 0xffffffff */
525 }
526 
527 static int stack_size(struct x86_emulate_ctxt *ctxt)
528 {
529 	return (__fls(stack_mask(ctxt)) + 1) >> 3;
530 }
531 
532 /* Access/update address held in a register, based on addressing mode. */
533 static inline unsigned long
534 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg)
535 {
536 	if (ctxt->ad_bytes == sizeof(unsigned long))
537 		return reg;
538 	else
539 		return reg & ad_mask(ctxt);
540 }
541 
542 static inline unsigned long
543 register_address(struct x86_emulate_ctxt *ctxt, int reg)
544 {
545 	return address_mask(ctxt, reg_read(ctxt, reg));
546 }
547 
548 static void masked_increment(ulong *reg, ulong mask, int inc)
549 {
550 	assign_masked(reg, *reg + inc, mask);
551 }
552 
553 static inline void
554 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc)
555 {
556 	ulong *preg = reg_rmw(ctxt, reg);
557 
558 	assign_register(preg, *preg + inc, ctxt->ad_bytes);
559 }
560 
561 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc)
562 {
563 	masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc);
564 }
565 
566 static u32 desc_limit_scaled(struct desc_struct *desc)
567 {
568 	u32 limit = get_desc_limit(desc);
569 
570 	return desc->g ? (limit << 12) | 0xfff : limit;
571 }
572 
573 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg)
574 {
575 	if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS)
576 		return 0;
577 
578 	return ctxt->ops->get_cached_segment_base(ctxt, seg);
579 }
580 
581 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec,
582 			     u32 error, bool valid)
583 {
584 	if (KVM_EMULATOR_BUG_ON(vec > 0x1f, ctxt))
585 		return X86EMUL_UNHANDLEABLE;
586 
587 	ctxt->exception.vector = vec;
588 	ctxt->exception.error_code = error;
589 	ctxt->exception.error_code_valid = valid;
590 	return X86EMUL_PROPAGATE_FAULT;
591 }
592 
593 static int emulate_db(struct x86_emulate_ctxt *ctxt)
594 {
595 	return emulate_exception(ctxt, DB_VECTOR, 0, false);
596 }
597 
598 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err)
599 {
600 	return emulate_exception(ctxt, GP_VECTOR, err, true);
601 }
602 
603 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err)
604 {
605 	return emulate_exception(ctxt, SS_VECTOR, err, true);
606 }
607 
608 static int emulate_ud(struct x86_emulate_ctxt *ctxt)
609 {
610 	return emulate_exception(ctxt, UD_VECTOR, 0, false);
611 }
612 
613 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err)
614 {
615 	return emulate_exception(ctxt, TS_VECTOR, err, true);
616 }
617 
618 static int emulate_de(struct x86_emulate_ctxt *ctxt)
619 {
620 	return emulate_exception(ctxt, DE_VECTOR, 0, false);
621 }
622 
623 static int emulate_nm(struct x86_emulate_ctxt *ctxt)
624 {
625 	return emulate_exception(ctxt, NM_VECTOR, 0, false);
626 }
627 
628 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg)
629 {
630 	u16 selector;
631 	struct desc_struct desc;
632 
633 	ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg);
634 	return selector;
635 }
636 
637 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector,
638 				 unsigned seg)
639 {
640 	u16 dummy;
641 	u32 base3;
642 	struct desc_struct desc;
643 
644 	ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg);
645 	ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg);
646 }
647 
648 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
649 {
650 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
651 }
652 
653 static inline bool emul_is_noncanonical_address(u64 la,
654 						struct x86_emulate_ctxt *ctxt)
655 {
656 	return !__is_canonical_address(la, ctxt_virt_addr_bits(ctxt));
657 }
658 
659 /*
660  * x86 defines three classes of vector instructions: explicitly
661  * aligned, explicitly unaligned, and the rest, which change behaviour
662  * depending on whether they're AVX encoded or not.
663  *
664  * Also included is CMPXCHG16B which is not a vector instruction, yet it is
665  * subject to the same check.  FXSAVE and FXRSTOR are checked here too as their
666  * 512 bytes of data must be aligned to a 16 byte boundary.
667  */
668 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size)
669 {
670 	u64 alignment = ctxt->d & AlignMask;
671 
672 	if (likely(size < 16))
673 		return 1;
674 
675 	switch (alignment) {
676 	case Unaligned:
677 	case Avx:
678 		return 1;
679 	case Aligned16:
680 		return 16;
681 	case Aligned:
682 	default:
683 		return size;
684 	}
685 }
686 
687 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt,
688 				       struct segmented_address addr,
689 				       unsigned *max_size, unsigned size,
690 				       enum x86emul_mode mode, ulong *linear,
691 				       unsigned int flags)
692 {
693 	struct desc_struct desc;
694 	bool usable;
695 	ulong la;
696 	u32 lim;
697 	u16 sel;
698 	u8  va_bits;
699 
700 	la = seg_base(ctxt, addr.seg) + addr.ea;
701 	*max_size = 0;
702 	switch (mode) {
703 	case X86EMUL_MODE_PROT64:
704 		*linear = la = ctxt->ops->get_untagged_addr(ctxt, la, flags);
705 		va_bits = ctxt_virt_addr_bits(ctxt);
706 		if (!__is_canonical_address(la, va_bits))
707 			goto bad;
708 
709 		*max_size = min_t(u64, ~0u, (1ull << va_bits) - la);
710 		if (size > *max_size)
711 			goto bad;
712 		break;
713 	default:
714 		*linear = la = (u32)la;
715 		usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL,
716 						addr.seg);
717 		if (!usable)
718 			goto bad;
719 		/* code segment in protected mode or read-only data segment */
720 		if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8)) || !(desc.type & 2)) &&
721 		    (flags & X86EMUL_F_WRITE))
722 			goto bad;
723 		/* unreadable code segment */
724 		if (!(flags & X86EMUL_F_FETCH) && (desc.type & 8) && !(desc.type & 2))
725 			goto bad;
726 		lim = desc_limit_scaled(&desc);
727 		if (!(desc.type & 8) && (desc.type & 4)) {
728 			/* expand-down segment */
729 			if (addr.ea <= lim)
730 				goto bad;
731 			lim = desc.d ? 0xffffffff : 0xffff;
732 		}
733 		if (addr.ea > lim)
734 			goto bad;
735 		if (lim == 0xffffffff)
736 			*max_size = ~0u;
737 		else {
738 			*max_size = (u64)lim + 1 - addr.ea;
739 			if (size > *max_size)
740 				goto bad;
741 		}
742 		break;
743 	}
744 	if (la & (insn_alignment(ctxt, size) - 1))
745 		return emulate_gp(ctxt, 0);
746 	return X86EMUL_CONTINUE;
747 bad:
748 	if (addr.seg == VCPU_SREG_SS)
749 		return emulate_ss(ctxt, 0);
750 	else
751 		return emulate_gp(ctxt, 0);
752 }
753 
754 static int linearize(struct x86_emulate_ctxt *ctxt,
755 		     struct segmented_address addr,
756 		     unsigned size, bool write,
757 		     ulong *linear)
758 {
759 	unsigned max_size;
760 	return __linearize(ctxt, addr, &max_size, size, ctxt->mode, linear,
761 			   write ? X86EMUL_F_WRITE : 0);
762 }
763 
764 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst)
765 {
766 	ulong linear;
767 	int rc;
768 	unsigned max_size;
769 	struct segmented_address addr = { .seg = VCPU_SREG_CS,
770 					   .ea = dst };
771 
772 	if (ctxt->op_bytes != sizeof(unsigned long))
773 		addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1);
774 	rc = __linearize(ctxt, addr, &max_size, 1, ctxt->mode, &linear,
775 			 X86EMUL_F_FETCH);
776 	if (rc == X86EMUL_CONTINUE)
777 		ctxt->_eip = addr.ea;
778 	return rc;
779 }
780 
781 static inline int emulator_recalc_and_set_mode(struct x86_emulate_ctxt *ctxt)
782 {
783 	u64 efer;
784 	struct desc_struct cs;
785 	u16 selector;
786 	u32 base3;
787 
788 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
789 
790 	if (!(ctxt->ops->get_cr(ctxt, 0) & X86_CR0_PE)) {
791 		/* Real mode. cpu must not have long mode active */
792 		if (efer & EFER_LMA)
793 			return X86EMUL_UNHANDLEABLE;
794 		ctxt->mode = X86EMUL_MODE_REAL;
795 		return X86EMUL_CONTINUE;
796 	}
797 
798 	if (ctxt->eflags & X86_EFLAGS_VM) {
799 		/* Protected/VM86 mode. cpu must not have long mode active */
800 		if (efer & EFER_LMA)
801 			return X86EMUL_UNHANDLEABLE;
802 		ctxt->mode = X86EMUL_MODE_VM86;
803 		return X86EMUL_CONTINUE;
804 	}
805 
806 	if (!ctxt->ops->get_segment(ctxt, &selector, &cs, &base3, VCPU_SREG_CS))
807 		return X86EMUL_UNHANDLEABLE;
808 
809 	if (efer & EFER_LMA) {
810 		if (cs.l) {
811 			/* Proper long mode */
812 			ctxt->mode = X86EMUL_MODE_PROT64;
813 		} else if (cs.d) {
814 			/* 32 bit compatibility mode*/
815 			ctxt->mode = X86EMUL_MODE_PROT32;
816 		} else {
817 			ctxt->mode = X86EMUL_MODE_PROT16;
818 		}
819 	} else {
820 		/* Legacy 32 bit / 16 bit mode */
821 		ctxt->mode = cs.d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
822 	}
823 
824 	return X86EMUL_CONTINUE;
825 }
826 
827 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst)
828 {
829 	return assign_eip(ctxt, dst);
830 }
831 
832 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst)
833 {
834 	int rc = emulator_recalc_and_set_mode(ctxt);
835 
836 	if (rc != X86EMUL_CONTINUE)
837 		return rc;
838 
839 	return assign_eip(ctxt, dst);
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, ctxt->mode, &linear,
912 			 X86EMUL_F_FETCH);
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 int reg;
1141 
1142 	if (ctxt->d & ModRM)
1143 		reg = ctxt->modrm_reg;
1144 	else
1145 		reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3);
1146 
1147 	if (ctxt->d & Sse) {
1148 		op->type = OP_XMM;
1149 		op->bytes = 16;
1150 		op->addr.xmm = reg;
1151 		kvm_read_sse_reg(reg, &op->vec_val);
1152 		return;
1153 	}
1154 	if (ctxt->d & Mmx) {
1155 		reg &= 7;
1156 		op->type = OP_MM;
1157 		op->bytes = 8;
1158 		op->addr.mm = reg;
1159 		return;
1160 	}
1161 
1162 	op->type = OP_REG;
1163 	op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1164 	op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp);
1165 
1166 	fetch_register_operand(op);
1167 	op->orig_val = op->val;
1168 }
1169 
1170 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg)
1171 {
1172 	if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP)
1173 		ctxt->modrm_seg = VCPU_SREG_SS;
1174 }
1175 
1176 static int decode_modrm(struct x86_emulate_ctxt *ctxt,
1177 			struct operand *op)
1178 {
1179 	u8 sib;
1180 	int index_reg, base_reg, scale;
1181 	int rc = X86EMUL_CONTINUE;
1182 	ulong modrm_ea = 0;
1183 
1184 	ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */
1185 	index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */
1186 	base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */
1187 
1188 	ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6;
1189 	ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3;
1190 	ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07);
1191 	ctxt->modrm_seg = VCPU_SREG_DS;
1192 
1193 	if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) {
1194 		op->type = OP_REG;
1195 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
1196 		op->addr.reg = decode_register(ctxt, ctxt->modrm_rm,
1197 				ctxt->d & ByteOp);
1198 		if (ctxt->d & Sse) {
1199 			op->type = OP_XMM;
1200 			op->bytes = 16;
1201 			op->addr.xmm = ctxt->modrm_rm;
1202 			kvm_read_sse_reg(ctxt->modrm_rm, &op->vec_val);
1203 			return rc;
1204 		}
1205 		if (ctxt->d & Mmx) {
1206 			op->type = OP_MM;
1207 			op->bytes = 8;
1208 			op->addr.mm = ctxt->modrm_rm & 7;
1209 			return rc;
1210 		}
1211 		fetch_register_operand(op);
1212 		return rc;
1213 	}
1214 
1215 	op->type = OP_MEM;
1216 
1217 	if (ctxt->ad_bytes == 2) {
1218 		unsigned bx = reg_read(ctxt, VCPU_REGS_RBX);
1219 		unsigned bp = reg_read(ctxt, VCPU_REGS_RBP);
1220 		unsigned si = reg_read(ctxt, VCPU_REGS_RSI);
1221 		unsigned di = reg_read(ctxt, VCPU_REGS_RDI);
1222 
1223 		/* 16-bit ModR/M decode. */
1224 		switch (ctxt->modrm_mod) {
1225 		case 0:
1226 			if (ctxt->modrm_rm == 6)
1227 				modrm_ea += insn_fetch(u16, ctxt);
1228 			break;
1229 		case 1:
1230 			modrm_ea += insn_fetch(s8, ctxt);
1231 			break;
1232 		case 2:
1233 			modrm_ea += insn_fetch(u16, ctxt);
1234 			break;
1235 		}
1236 		switch (ctxt->modrm_rm) {
1237 		case 0:
1238 			modrm_ea += bx + si;
1239 			break;
1240 		case 1:
1241 			modrm_ea += bx + di;
1242 			break;
1243 		case 2:
1244 			modrm_ea += bp + si;
1245 			break;
1246 		case 3:
1247 			modrm_ea += bp + di;
1248 			break;
1249 		case 4:
1250 			modrm_ea += si;
1251 			break;
1252 		case 5:
1253 			modrm_ea += di;
1254 			break;
1255 		case 6:
1256 			if (ctxt->modrm_mod != 0)
1257 				modrm_ea += bp;
1258 			break;
1259 		case 7:
1260 			modrm_ea += bx;
1261 			break;
1262 		}
1263 		if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 ||
1264 		    (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0))
1265 			ctxt->modrm_seg = VCPU_SREG_SS;
1266 		modrm_ea = (u16)modrm_ea;
1267 	} else {
1268 		/* 32/64-bit ModR/M decode. */
1269 		if ((ctxt->modrm_rm & 7) == 4) {
1270 			sib = insn_fetch(u8, ctxt);
1271 			index_reg |= (sib >> 3) & 7;
1272 			base_reg |= sib & 7;
1273 			scale = sib >> 6;
1274 
1275 			if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0)
1276 				modrm_ea += insn_fetch(s32, ctxt);
1277 			else {
1278 				modrm_ea += reg_read(ctxt, base_reg);
1279 				adjust_modrm_seg(ctxt, base_reg);
1280 				/* Increment ESP on POP [ESP] */
1281 				if ((ctxt->d & IncSP) &&
1282 				    base_reg == VCPU_REGS_RSP)
1283 					modrm_ea += ctxt->op_bytes;
1284 			}
1285 			if (index_reg != 4)
1286 				modrm_ea += reg_read(ctxt, index_reg) << scale;
1287 		} else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) {
1288 			modrm_ea += insn_fetch(s32, ctxt);
1289 			if (ctxt->mode == X86EMUL_MODE_PROT64)
1290 				ctxt->rip_relative = 1;
1291 		} else {
1292 			base_reg = ctxt->modrm_rm;
1293 			modrm_ea += reg_read(ctxt, base_reg);
1294 			adjust_modrm_seg(ctxt, base_reg);
1295 		}
1296 		switch (ctxt->modrm_mod) {
1297 		case 1:
1298 			modrm_ea += insn_fetch(s8, ctxt);
1299 			break;
1300 		case 2:
1301 			modrm_ea += insn_fetch(s32, ctxt);
1302 			break;
1303 		}
1304 	}
1305 	op->addr.mem.ea = modrm_ea;
1306 	if (ctxt->ad_bytes != 8)
1307 		ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea;
1308 
1309 done:
1310 	return rc;
1311 }
1312 
1313 static int decode_abs(struct x86_emulate_ctxt *ctxt,
1314 		      struct operand *op)
1315 {
1316 	int rc = X86EMUL_CONTINUE;
1317 
1318 	op->type = OP_MEM;
1319 	switch (ctxt->ad_bytes) {
1320 	case 2:
1321 		op->addr.mem.ea = insn_fetch(u16, ctxt);
1322 		break;
1323 	case 4:
1324 		op->addr.mem.ea = insn_fetch(u32, ctxt);
1325 		break;
1326 	case 8:
1327 		op->addr.mem.ea = insn_fetch(u64, ctxt);
1328 		break;
1329 	}
1330 done:
1331 	return rc;
1332 }
1333 
1334 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt)
1335 {
1336 	long sv = 0, mask;
1337 
1338 	if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) {
1339 		mask = ~((long)ctxt->dst.bytes * 8 - 1);
1340 
1341 		if (ctxt->src.bytes == 2)
1342 			sv = (s16)ctxt->src.val & (s16)mask;
1343 		else if (ctxt->src.bytes == 4)
1344 			sv = (s32)ctxt->src.val & (s32)mask;
1345 		else
1346 			sv = (s64)ctxt->src.val & (s64)mask;
1347 
1348 		ctxt->dst.addr.mem.ea = address_mask(ctxt,
1349 					   ctxt->dst.addr.mem.ea + (sv >> 3));
1350 	}
1351 
1352 	/* only subword offset */
1353 	ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
1354 }
1355 
1356 static int read_emulated(struct x86_emulate_ctxt *ctxt,
1357 			 unsigned long addr, void *dest, unsigned size)
1358 {
1359 	int rc;
1360 	struct read_cache *mc = &ctxt->mem_read;
1361 
1362 	if (mc->pos < mc->end)
1363 		goto read_cached;
1364 
1365 	if (KVM_EMULATOR_BUG_ON((mc->end + size) >= sizeof(mc->data), ctxt))
1366 		return X86EMUL_UNHANDLEABLE;
1367 
1368 	rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size,
1369 				      &ctxt->exception);
1370 	if (rc != X86EMUL_CONTINUE)
1371 		return rc;
1372 
1373 	mc->end += size;
1374 
1375 read_cached:
1376 	memcpy(dest, mc->data + mc->pos, size);
1377 	mc->pos += size;
1378 	return X86EMUL_CONTINUE;
1379 }
1380 
1381 static int segmented_read(struct x86_emulate_ctxt *ctxt,
1382 			  struct segmented_address addr,
1383 			  void *data,
1384 			  unsigned size)
1385 {
1386 	int rc;
1387 	ulong linear;
1388 
1389 	rc = linearize(ctxt, addr, size, false, &linear);
1390 	if (rc != X86EMUL_CONTINUE)
1391 		return rc;
1392 	return read_emulated(ctxt, linear, data, size);
1393 }
1394 
1395 static int segmented_write(struct x86_emulate_ctxt *ctxt,
1396 			   struct segmented_address addr,
1397 			   const void *data,
1398 			   unsigned size)
1399 {
1400 	int rc;
1401 	ulong linear;
1402 
1403 	rc = linearize(ctxt, addr, size, true, &linear);
1404 	if (rc != X86EMUL_CONTINUE)
1405 		return rc;
1406 	return ctxt->ops->write_emulated(ctxt, linear, data, size,
1407 					 &ctxt->exception);
1408 }
1409 
1410 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt,
1411 			     struct segmented_address addr,
1412 			     const void *orig_data, const void *data,
1413 			     unsigned size)
1414 {
1415 	int rc;
1416 	ulong linear;
1417 
1418 	rc = linearize(ctxt, addr, size, true, &linear);
1419 	if (rc != X86EMUL_CONTINUE)
1420 		return rc;
1421 	return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data,
1422 					   size, &ctxt->exception);
1423 }
1424 
1425 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt,
1426 			   unsigned int size, unsigned short port,
1427 			   void *dest)
1428 {
1429 	struct read_cache *rc = &ctxt->io_read;
1430 
1431 	if (rc->pos == rc->end) { /* refill pio read ahead */
1432 		unsigned int in_page, n;
1433 		unsigned int count = ctxt->rep_prefix ?
1434 			address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1;
1435 		in_page = (ctxt->eflags & X86_EFLAGS_DF) ?
1436 			offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) :
1437 			PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI));
1438 		n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count);
1439 		if (n == 0)
1440 			n = 1;
1441 		rc->pos = rc->end = 0;
1442 		if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n))
1443 			return 0;
1444 		rc->end = n * size;
1445 	}
1446 
1447 	if (ctxt->rep_prefix && (ctxt->d & String) &&
1448 	    !(ctxt->eflags & X86_EFLAGS_DF)) {
1449 		ctxt->dst.data = rc->data + rc->pos;
1450 		ctxt->dst.type = OP_MEM_STR;
1451 		ctxt->dst.count = (rc->end - rc->pos) / size;
1452 		rc->pos = rc->end;
1453 	} else {
1454 		memcpy(dest, rc->data + rc->pos, size);
1455 		rc->pos += size;
1456 	}
1457 	return 1;
1458 }
1459 
1460 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt,
1461 				     u16 index, struct desc_struct *desc)
1462 {
1463 	struct desc_ptr dt;
1464 	ulong addr;
1465 
1466 	ctxt->ops->get_idt(ctxt, &dt);
1467 
1468 	if (dt.size < index * 8 + 7)
1469 		return emulate_gp(ctxt, index << 3 | 0x2);
1470 
1471 	addr = dt.address + index * 8;
1472 	return linear_read_system(ctxt, addr, desc, sizeof(*desc));
1473 }
1474 
1475 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt,
1476 				     u16 selector, struct desc_ptr *dt)
1477 {
1478 	const struct x86_emulate_ops *ops = ctxt->ops;
1479 	u32 base3 = 0;
1480 
1481 	if (selector & 1 << 2) {
1482 		struct desc_struct desc;
1483 		u16 sel;
1484 
1485 		memset(dt, 0, sizeof(*dt));
1486 		if (!ops->get_segment(ctxt, &sel, &desc, &base3,
1487 				      VCPU_SREG_LDTR))
1488 			return;
1489 
1490 		dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */
1491 		dt->address = get_desc_base(&desc) | ((u64)base3 << 32);
1492 	} else
1493 		ops->get_gdt(ctxt, dt);
1494 }
1495 
1496 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt,
1497 			      u16 selector, ulong *desc_addr_p)
1498 {
1499 	struct desc_ptr dt;
1500 	u16 index = selector >> 3;
1501 	ulong addr;
1502 
1503 	get_descriptor_table_ptr(ctxt, selector, &dt);
1504 
1505 	if (dt.size < index * 8 + 7)
1506 		return emulate_gp(ctxt, selector & 0xfffc);
1507 
1508 	addr = dt.address + index * 8;
1509 
1510 #ifdef CONFIG_X86_64
1511 	if (addr >> 32 != 0) {
1512 		u64 efer = 0;
1513 
1514 		ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1515 		if (!(efer & EFER_LMA))
1516 			addr &= (u32)-1;
1517 	}
1518 #endif
1519 
1520 	*desc_addr_p = addr;
1521 	return X86EMUL_CONTINUE;
1522 }
1523 
1524 /* allowed just for 8 bytes segments */
1525 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1526 				   u16 selector, struct desc_struct *desc,
1527 				   ulong *desc_addr_p)
1528 {
1529 	int rc;
1530 
1531 	rc = get_descriptor_ptr(ctxt, selector, desc_addr_p);
1532 	if (rc != X86EMUL_CONTINUE)
1533 		return rc;
1534 
1535 	return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc));
1536 }
1537 
1538 /* allowed just for 8 bytes segments */
1539 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1540 				    u16 selector, struct desc_struct *desc)
1541 {
1542 	int rc;
1543 	ulong addr;
1544 
1545 	rc = get_descriptor_ptr(ctxt, selector, &addr);
1546 	if (rc != X86EMUL_CONTINUE)
1547 		return rc;
1548 
1549 	return linear_write_system(ctxt, addr, desc, sizeof(*desc));
1550 }
1551 
1552 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1553 				     u16 selector, int seg, u8 cpl,
1554 				     enum x86_transfer_type transfer,
1555 				     struct desc_struct *desc)
1556 {
1557 	struct desc_struct seg_desc, old_desc;
1558 	u8 dpl, rpl;
1559 	unsigned err_vec = GP_VECTOR;
1560 	u32 err_code = 0;
1561 	bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
1562 	ulong desc_addr;
1563 	int ret;
1564 	u16 dummy;
1565 	u32 base3 = 0;
1566 
1567 	memset(&seg_desc, 0, sizeof(seg_desc));
1568 
1569 	if (ctxt->mode == X86EMUL_MODE_REAL) {
1570 		/* set real mode segment descriptor (keep limit etc. for
1571 		 * unreal mode) */
1572 		ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg);
1573 		set_desc_base(&seg_desc, selector << 4);
1574 		goto load;
1575 	} else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) {
1576 		/* VM86 needs a clean new segment descriptor */
1577 		set_desc_base(&seg_desc, selector << 4);
1578 		set_desc_limit(&seg_desc, 0xffff);
1579 		seg_desc.type = 3;
1580 		seg_desc.p = 1;
1581 		seg_desc.s = 1;
1582 		seg_desc.dpl = 3;
1583 		goto load;
1584 	}
1585 
1586 	rpl = selector & 3;
1587 
1588 	/* TR should be in GDT only */
1589 	if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
1590 		goto exception;
1591 
1592 	/* NULL selector is not valid for TR, CS and (except for long mode) SS */
1593 	if (null_selector) {
1594 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR)
1595 			goto exception;
1596 
1597 		if (seg == VCPU_SREG_SS) {
1598 			if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl)
1599 				goto exception;
1600 
1601 			/*
1602 			 * ctxt->ops->set_segment expects the CPL to be in
1603 			 * SS.DPL, so fake an expand-up 32-bit data segment.
1604 			 */
1605 			seg_desc.type = 3;
1606 			seg_desc.p = 1;
1607 			seg_desc.s = 1;
1608 			seg_desc.dpl = cpl;
1609 			seg_desc.d = 1;
1610 			seg_desc.g = 1;
1611 		}
1612 
1613 		/* Skip all following checks */
1614 		goto load;
1615 	}
1616 
1617 	ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr);
1618 	if (ret != X86EMUL_CONTINUE)
1619 		return ret;
1620 
1621 	err_code = selector & 0xfffc;
1622 	err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR :
1623 							   GP_VECTOR;
1624 
1625 	/* can't load system descriptor into segment selector */
1626 	if (seg <= VCPU_SREG_GS && !seg_desc.s) {
1627 		if (transfer == X86_TRANSFER_CALL_JMP)
1628 			return X86EMUL_UNHANDLEABLE;
1629 		goto exception;
1630 	}
1631 
1632 	dpl = seg_desc.dpl;
1633 
1634 	switch (seg) {
1635 	case VCPU_SREG_SS:
1636 		/*
1637 		 * segment is not a writable data segment or segment
1638 		 * selector's RPL != CPL or DPL != CPL
1639 		 */
1640 		if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl)
1641 			goto exception;
1642 		break;
1643 	case VCPU_SREG_CS:
1644 		/*
1645 		 * KVM uses "none" when loading CS as part of emulating Real
1646 		 * Mode exceptions and IRET (handled above).  In all other
1647 		 * cases, loading CS without a control transfer is a KVM bug.
1648 		 */
1649 		if (WARN_ON_ONCE(transfer == X86_TRANSFER_NONE))
1650 			goto exception;
1651 
1652 		if (!(seg_desc.type & 8))
1653 			goto exception;
1654 
1655 		if (transfer == X86_TRANSFER_RET) {
1656 			/* RET can never return to an inner privilege level. */
1657 			if (rpl < cpl)
1658 				goto exception;
1659 			/* Outer-privilege level return is not implemented */
1660 			if (rpl > cpl)
1661 				return X86EMUL_UNHANDLEABLE;
1662 		}
1663 		if (transfer == X86_TRANSFER_RET || transfer == X86_TRANSFER_TASK_SWITCH) {
1664 			if (seg_desc.type & 4) {
1665 				/* conforming */
1666 				if (dpl > rpl)
1667 					goto exception;
1668 			} else {
1669 				/* nonconforming */
1670 				if (dpl != rpl)
1671 					goto exception;
1672 			}
1673 		} else { /* X86_TRANSFER_CALL_JMP */
1674 			if (seg_desc.type & 4) {
1675 				/* conforming */
1676 				if (dpl > cpl)
1677 					goto exception;
1678 			} else {
1679 				/* nonconforming */
1680 				if (rpl > cpl || dpl != cpl)
1681 					goto exception;
1682 			}
1683 		}
1684 		/* in long-mode d/b must be clear if l is set */
1685 		if (seg_desc.d && seg_desc.l) {
1686 			u64 efer = 0;
1687 
1688 			ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
1689 			if (efer & EFER_LMA)
1690 				goto exception;
1691 		}
1692 
1693 		/* CS(RPL) <- CPL */
1694 		selector = (selector & 0xfffc) | cpl;
1695 		break;
1696 	case VCPU_SREG_TR:
1697 		if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9))
1698 			goto exception;
1699 		break;
1700 	case VCPU_SREG_LDTR:
1701 		if (seg_desc.s || seg_desc.type != 2)
1702 			goto exception;
1703 		break;
1704 	default: /*  DS, ES, FS, or GS */
1705 		/*
1706 		 * segment is not a data or readable code segment or
1707 		 * ((segment is a data or nonconforming code segment)
1708 		 * and ((RPL > DPL) or (CPL > DPL)))
1709 		 */
1710 		if ((seg_desc.type & 0xa) == 0x8 ||
1711 		    (((seg_desc.type & 0xc) != 0xc) &&
1712 		     (rpl > dpl || cpl > dpl)))
1713 			goto exception;
1714 		break;
1715 	}
1716 
1717 	if (!seg_desc.p) {
1718 		err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
1719 		goto exception;
1720 	}
1721 
1722 	if (seg_desc.s) {
1723 		/* mark segment as accessed */
1724 		if (!(seg_desc.type & 1)) {
1725 			seg_desc.type |= 1;
1726 			ret = write_segment_descriptor(ctxt, selector,
1727 						       &seg_desc);
1728 			if (ret != X86EMUL_CONTINUE)
1729 				return ret;
1730 		}
1731 	} else if (ctxt->mode == X86EMUL_MODE_PROT64) {
1732 		ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3));
1733 		if (ret != X86EMUL_CONTINUE)
1734 			return ret;
1735 		if (emul_is_noncanonical_address(get_desc_base(&seg_desc) |
1736 						 ((u64)base3 << 32), ctxt))
1737 			return emulate_gp(ctxt, err_code);
1738 	}
1739 
1740 	if (seg == VCPU_SREG_TR) {
1741 		old_desc = seg_desc;
1742 		seg_desc.type |= 2; /* busy */
1743 		ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc,
1744 						  sizeof(seg_desc), &ctxt->exception);
1745 		if (ret != X86EMUL_CONTINUE)
1746 			return ret;
1747 	}
1748 load:
1749 	ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg);
1750 	if (desc)
1751 		*desc = seg_desc;
1752 	return X86EMUL_CONTINUE;
1753 exception:
1754 	return emulate_exception(ctxt, err_vec, err_code, true);
1755 }
1756 
1757 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
1758 				   u16 selector, int seg)
1759 {
1760 	u8 cpl = ctxt->ops->cpl(ctxt);
1761 
1762 	/*
1763 	 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but
1764 	 * they can load it at CPL<3 (Intel's manual says only LSS can,
1765 	 * but it's wrong).
1766 	 *
1767 	 * However, the Intel manual says that putting IST=1/DPL=3 in
1768 	 * an interrupt gate will result in SS=3 (the AMD manual instead
1769 	 * says it doesn't), so allow SS=3 in __load_segment_descriptor
1770 	 * and only forbid it here.
1771 	 */
1772 	if (seg == VCPU_SREG_SS && selector == 3 &&
1773 	    ctxt->mode == X86EMUL_MODE_PROT64)
1774 		return emulate_exception(ctxt, GP_VECTOR, 0, true);
1775 
1776 	return __load_segment_descriptor(ctxt, selector, seg, cpl,
1777 					 X86_TRANSFER_NONE, NULL);
1778 }
1779 
1780 static void write_register_operand(struct operand *op)
1781 {
1782 	return assign_register(op->addr.reg, op->val, op->bytes);
1783 }
1784 
1785 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op)
1786 {
1787 	switch (op->type) {
1788 	case OP_REG:
1789 		write_register_operand(op);
1790 		break;
1791 	case OP_MEM:
1792 		if (ctxt->lock_prefix)
1793 			return segmented_cmpxchg(ctxt,
1794 						 op->addr.mem,
1795 						 &op->orig_val,
1796 						 &op->val,
1797 						 op->bytes);
1798 		else
1799 			return segmented_write(ctxt,
1800 					       op->addr.mem,
1801 					       &op->val,
1802 					       op->bytes);
1803 	case OP_MEM_STR:
1804 		return segmented_write(ctxt,
1805 				       op->addr.mem,
1806 				       op->data,
1807 				       op->bytes * op->count);
1808 	case OP_XMM:
1809 		kvm_write_sse_reg(op->addr.xmm, &op->vec_val);
1810 		break;
1811 	case OP_MM:
1812 		kvm_write_mmx_reg(op->addr.mm, &op->mm_val);
1813 		break;
1814 	case OP_NONE:
1815 		/* no writeback */
1816 		break;
1817 	default:
1818 		break;
1819 	}
1820 	return X86EMUL_CONTINUE;
1821 }
1822 
1823 static int emulate_push(struct x86_emulate_ctxt *ctxt, const void *data, int len)
1824 {
1825 	struct segmented_address addr;
1826 
1827 	rsp_increment(ctxt, -len);
1828 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1829 	addr.seg = VCPU_SREG_SS;
1830 
1831 	return segmented_write(ctxt, addr, data, len);
1832 }
1833 
1834 static int em_push(struct x86_emulate_ctxt *ctxt)
1835 {
1836 	/* Disable writeback. */
1837 	ctxt->dst.type = OP_NONE;
1838 	return emulate_push(ctxt, &ctxt->src.val, ctxt->op_bytes);
1839 }
1840 
1841 static int emulate_pop(struct x86_emulate_ctxt *ctxt,
1842 		       void *dest, int len)
1843 {
1844 	int rc;
1845 	struct segmented_address addr;
1846 
1847 	addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt);
1848 	addr.seg = VCPU_SREG_SS;
1849 	rc = segmented_read(ctxt, addr, dest, len);
1850 	if (rc != X86EMUL_CONTINUE)
1851 		return rc;
1852 
1853 	rsp_increment(ctxt, len);
1854 	return rc;
1855 }
1856 
1857 static int em_pop(struct x86_emulate_ctxt *ctxt)
1858 {
1859 	return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1860 }
1861 
1862 static int emulate_popf(struct x86_emulate_ctxt *ctxt,
1863 			void *dest, int len)
1864 {
1865 	int rc;
1866 	unsigned long val = 0;
1867 	unsigned long change_mask;
1868 	int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
1869 	int cpl = ctxt->ops->cpl(ctxt);
1870 
1871 	rc = emulate_pop(ctxt, &val, len);
1872 	if (rc != X86EMUL_CONTINUE)
1873 		return rc;
1874 
1875 	change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
1876 		      X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF |
1877 		      X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT |
1878 		      X86_EFLAGS_AC | X86_EFLAGS_ID;
1879 
1880 	switch(ctxt->mode) {
1881 	case X86EMUL_MODE_PROT64:
1882 	case X86EMUL_MODE_PROT32:
1883 	case X86EMUL_MODE_PROT16:
1884 		if (cpl == 0)
1885 			change_mask |= X86_EFLAGS_IOPL;
1886 		if (cpl <= iopl)
1887 			change_mask |= X86_EFLAGS_IF;
1888 		break;
1889 	case X86EMUL_MODE_VM86:
1890 		if (iopl < 3)
1891 			return emulate_gp(ctxt, 0);
1892 		change_mask |= X86_EFLAGS_IF;
1893 		break;
1894 	default: /* real mode */
1895 		change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF);
1896 		break;
1897 	}
1898 
1899 	*(unsigned long *)dest =
1900 		(ctxt->eflags & ~change_mask) | (val & change_mask);
1901 
1902 	return rc;
1903 }
1904 
1905 static int em_popf(struct x86_emulate_ctxt *ctxt)
1906 {
1907 	ctxt->dst.type = OP_REG;
1908 	ctxt->dst.addr.reg = &ctxt->eflags;
1909 	ctxt->dst.bytes = ctxt->op_bytes;
1910 	return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes);
1911 }
1912 
1913 static int em_enter(struct x86_emulate_ctxt *ctxt)
1914 {
1915 	int rc;
1916 	unsigned frame_size = ctxt->src.val;
1917 	unsigned nesting_level = ctxt->src2.val & 31;
1918 	ulong rbp;
1919 
1920 	if (nesting_level)
1921 		return X86EMUL_UNHANDLEABLE;
1922 
1923 	rbp = reg_read(ctxt, VCPU_REGS_RBP);
1924 	rc = emulate_push(ctxt, &rbp, stack_size(ctxt));
1925 	if (rc != X86EMUL_CONTINUE)
1926 		return rc;
1927 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP),
1928 		      stack_mask(ctxt));
1929 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP),
1930 		      reg_read(ctxt, VCPU_REGS_RSP) - frame_size,
1931 		      stack_mask(ctxt));
1932 	return X86EMUL_CONTINUE;
1933 }
1934 
1935 static int em_leave(struct x86_emulate_ctxt *ctxt)
1936 {
1937 	assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP),
1938 		      stack_mask(ctxt));
1939 	return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes);
1940 }
1941 
1942 static int em_push_sreg(struct x86_emulate_ctxt *ctxt)
1943 {
1944 	int seg = ctxt->src2.val;
1945 
1946 	ctxt->src.val = get_segment_selector(ctxt, seg);
1947 	if (ctxt->op_bytes == 4) {
1948 		rsp_increment(ctxt, -2);
1949 		ctxt->op_bytes = 2;
1950 	}
1951 
1952 	return em_push(ctxt);
1953 }
1954 
1955 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt)
1956 {
1957 	int seg = ctxt->src2.val;
1958 	unsigned long selector = 0;
1959 	int rc;
1960 
1961 	rc = emulate_pop(ctxt, &selector, 2);
1962 	if (rc != X86EMUL_CONTINUE)
1963 		return rc;
1964 
1965 	if (seg == VCPU_SREG_SS)
1966 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
1967 	if (ctxt->op_bytes > 2)
1968 		rsp_increment(ctxt, ctxt->op_bytes - 2);
1969 
1970 	rc = load_segment_descriptor(ctxt, (u16)selector, seg);
1971 	return rc;
1972 }
1973 
1974 static int em_pusha(struct x86_emulate_ctxt *ctxt)
1975 {
1976 	unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP);
1977 	int rc = X86EMUL_CONTINUE;
1978 	int reg = VCPU_REGS_RAX;
1979 
1980 	while (reg <= VCPU_REGS_RDI) {
1981 		(reg == VCPU_REGS_RSP) ?
1982 		(ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg));
1983 
1984 		rc = em_push(ctxt);
1985 		if (rc != X86EMUL_CONTINUE)
1986 			return rc;
1987 
1988 		++reg;
1989 	}
1990 
1991 	return rc;
1992 }
1993 
1994 static int em_pushf(struct x86_emulate_ctxt *ctxt)
1995 {
1996 	ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM;
1997 	return em_push(ctxt);
1998 }
1999 
2000 static int em_popa(struct x86_emulate_ctxt *ctxt)
2001 {
2002 	int rc = X86EMUL_CONTINUE;
2003 	int reg = VCPU_REGS_RDI;
2004 	u32 val = 0;
2005 
2006 	while (reg >= VCPU_REGS_RAX) {
2007 		if (reg == VCPU_REGS_RSP) {
2008 			rsp_increment(ctxt, ctxt->op_bytes);
2009 			--reg;
2010 		}
2011 
2012 		rc = emulate_pop(ctxt, &val, ctxt->op_bytes);
2013 		if (rc != X86EMUL_CONTINUE)
2014 			break;
2015 		assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes);
2016 		--reg;
2017 	}
2018 	return rc;
2019 }
2020 
2021 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2022 {
2023 	const struct x86_emulate_ops *ops = ctxt->ops;
2024 	int rc;
2025 	struct desc_ptr dt;
2026 	gva_t cs_addr;
2027 	gva_t eip_addr;
2028 	u16 cs, eip;
2029 
2030 	/* TODO: Add limit checks */
2031 	ctxt->src.val = ctxt->eflags;
2032 	rc = em_push(ctxt);
2033 	if (rc != X86EMUL_CONTINUE)
2034 		return rc;
2035 
2036 	ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC);
2037 
2038 	ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS);
2039 	rc = em_push(ctxt);
2040 	if (rc != X86EMUL_CONTINUE)
2041 		return rc;
2042 
2043 	ctxt->src.val = ctxt->_eip;
2044 	rc = em_push(ctxt);
2045 	if (rc != X86EMUL_CONTINUE)
2046 		return rc;
2047 
2048 	ops->get_idt(ctxt, &dt);
2049 
2050 	eip_addr = dt.address + (irq << 2);
2051 	cs_addr = dt.address + (irq << 2) + 2;
2052 
2053 	rc = linear_read_system(ctxt, cs_addr, &cs, 2);
2054 	if (rc != X86EMUL_CONTINUE)
2055 		return rc;
2056 
2057 	rc = linear_read_system(ctxt, eip_addr, &eip, 2);
2058 	if (rc != X86EMUL_CONTINUE)
2059 		return rc;
2060 
2061 	rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS);
2062 	if (rc != X86EMUL_CONTINUE)
2063 		return rc;
2064 
2065 	ctxt->_eip = eip;
2066 
2067 	return rc;
2068 }
2069 
2070 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq)
2071 {
2072 	int rc;
2073 
2074 	invalidate_registers(ctxt);
2075 	rc = __emulate_int_real(ctxt, irq);
2076 	if (rc == X86EMUL_CONTINUE)
2077 		writeback_registers(ctxt);
2078 	return rc;
2079 }
2080 
2081 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq)
2082 {
2083 	switch(ctxt->mode) {
2084 	case X86EMUL_MODE_REAL:
2085 		return __emulate_int_real(ctxt, irq);
2086 	case X86EMUL_MODE_VM86:
2087 	case X86EMUL_MODE_PROT16:
2088 	case X86EMUL_MODE_PROT32:
2089 	case X86EMUL_MODE_PROT64:
2090 	default:
2091 		/* Protected mode interrupts unimplemented yet */
2092 		return X86EMUL_UNHANDLEABLE;
2093 	}
2094 }
2095 
2096 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
2097 {
2098 	int rc = X86EMUL_CONTINUE;
2099 	unsigned long temp_eip = 0;
2100 	unsigned long temp_eflags = 0;
2101 	unsigned long cs = 0;
2102 	unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
2103 			     X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF |
2104 			     X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF |
2105 			     X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF |
2106 			     X86_EFLAGS_AC | X86_EFLAGS_ID |
2107 			     X86_EFLAGS_FIXED;
2108 	unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
2109 				  X86_EFLAGS_VIP;
2110 
2111 	/* TODO: Add stack limit check */
2112 
2113 	rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes);
2114 
2115 	if (rc != X86EMUL_CONTINUE)
2116 		return rc;
2117 
2118 	if (temp_eip & ~0xffff)
2119 		return emulate_gp(ctxt, 0);
2120 
2121 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2122 
2123 	if (rc != X86EMUL_CONTINUE)
2124 		return rc;
2125 
2126 	rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes);
2127 
2128 	if (rc != X86EMUL_CONTINUE)
2129 		return rc;
2130 
2131 	rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
2132 
2133 	if (rc != X86EMUL_CONTINUE)
2134 		return rc;
2135 
2136 	ctxt->_eip = temp_eip;
2137 
2138 	if (ctxt->op_bytes == 4)
2139 		ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask));
2140 	else if (ctxt->op_bytes == 2) {
2141 		ctxt->eflags &= ~0xffff;
2142 		ctxt->eflags |= temp_eflags;
2143 	}
2144 
2145 	ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */
2146 	ctxt->eflags |= X86_EFLAGS_FIXED;
2147 	ctxt->ops->set_nmi_mask(ctxt, false);
2148 
2149 	return rc;
2150 }
2151 
2152 static int em_iret(struct x86_emulate_ctxt *ctxt)
2153 {
2154 	switch(ctxt->mode) {
2155 	case X86EMUL_MODE_REAL:
2156 		return emulate_iret_real(ctxt);
2157 	case X86EMUL_MODE_VM86:
2158 	case X86EMUL_MODE_PROT16:
2159 	case X86EMUL_MODE_PROT32:
2160 	case X86EMUL_MODE_PROT64:
2161 	default:
2162 		/* iret from protected mode unimplemented yet */
2163 		return X86EMUL_UNHANDLEABLE;
2164 	}
2165 }
2166 
2167 static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
2168 {
2169 	int rc;
2170 	unsigned short sel;
2171 	struct desc_struct new_desc;
2172 	u8 cpl = ctxt->ops->cpl(ctxt);
2173 
2174 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2175 
2176 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
2177 				       X86_TRANSFER_CALL_JMP,
2178 				       &new_desc);
2179 	if (rc != X86EMUL_CONTINUE)
2180 		return rc;
2181 
2182 	rc = assign_eip_far(ctxt, ctxt->src.val);
2183 	/* Error handling is not implemented. */
2184 	if (rc != X86EMUL_CONTINUE)
2185 		return X86EMUL_UNHANDLEABLE;
2186 
2187 	return rc;
2188 }
2189 
2190 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt)
2191 {
2192 	return assign_eip_near(ctxt, ctxt->src.val);
2193 }
2194 
2195 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt)
2196 {
2197 	int rc;
2198 	long int old_eip;
2199 
2200 	old_eip = ctxt->_eip;
2201 	rc = assign_eip_near(ctxt, ctxt->src.val);
2202 	if (rc != X86EMUL_CONTINUE)
2203 		return rc;
2204 	ctxt->src.val = old_eip;
2205 	rc = em_push(ctxt);
2206 	return rc;
2207 }
2208 
2209 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
2210 {
2211 	u64 old = ctxt->dst.orig_val64;
2212 
2213 	if (ctxt->dst.bytes == 16)
2214 		return X86EMUL_UNHANDLEABLE;
2215 
2216 	if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) ||
2217 	    ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) {
2218 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0);
2219 		*reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32);
2220 		ctxt->eflags &= ~X86_EFLAGS_ZF;
2221 	} else {
2222 		ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) |
2223 			(u32) reg_read(ctxt, VCPU_REGS_RBX);
2224 
2225 		ctxt->eflags |= X86_EFLAGS_ZF;
2226 	}
2227 	return X86EMUL_CONTINUE;
2228 }
2229 
2230 static int em_ret(struct x86_emulate_ctxt *ctxt)
2231 {
2232 	int rc;
2233 	unsigned long eip = 0;
2234 
2235 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2236 	if (rc != X86EMUL_CONTINUE)
2237 		return rc;
2238 
2239 	return assign_eip_near(ctxt, eip);
2240 }
2241 
2242 static int em_ret_far(struct x86_emulate_ctxt *ctxt)
2243 {
2244 	int rc;
2245 	unsigned long eip = 0;
2246 	unsigned long cs = 0;
2247 	int cpl = ctxt->ops->cpl(ctxt);
2248 	struct desc_struct new_desc;
2249 
2250 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
2251 	if (rc != X86EMUL_CONTINUE)
2252 		return rc;
2253 	rc = emulate_pop(ctxt, &cs, ctxt->op_bytes);
2254 	if (rc != X86EMUL_CONTINUE)
2255 		return rc;
2256 	rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
2257 				       X86_TRANSFER_RET,
2258 				       &new_desc);
2259 	if (rc != X86EMUL_CONTINUE)
2260 		return rc;
2261 	rc = assign_eip_far(ctxt, eip);
2262 	/* Error handling is not implemented. */
2263 	if (rc != X86EMUL_CONTINUE)
2264 		return X86EMUL_UNHANDLEABLE;
2265 
2266 	return rc;
2267 }
2268 
2269 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt)
2270 {
2271         int rc;
2272 
2273         rc = em_ret_far(ctxt);
2274         if (rc != X86EMUL_CONTINUE)
2275                 return rc;
2276         rsp_increment(ctxt, ctxt->src.val);
2277         return X86EMUL_CONTINUE;
2278 }
2279 
2280 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
2281 {
2282 	/* Save real source value, then compare EAX against destination. */
2283 	ctxt->dst.orig_val = ctxt->dst.val;
2284 	ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX);
2285 	ctxt->src.orig_val = ctxt->src.val;
2286 	ctxt->src.val = ctxt->dst.orig_val;
2287 	fastop(ctxt, em_cmp);
2288 
2289 	if (ctxt->eflags & X86_EFLAGS_ZF) {
2290 		/* Success: write back to memory; no update of EAX */
2291 		ctxt->src.type = OP_NONE;
2292 		ctxt->dst.val = ctxt->src.orig_val;
2293 	} else {
2294 		/* Failure: write the value we saw to EAX. */
2295 		ctxt->src.type = OP_REG;
2296 		ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
2297 		ctxt->src.val = ctxt->dst.orig_val;
2298 		/* Create write-cycle to dest by writing the same value */
2299 		ctxt->dst.val = ctxt->dst.orig_val;
2300 	}
2301 	return X86EMUL_CONTINUE;
2302 }
2303 
2304 static int em_lseg(struct x86_emulate_ctxt *ctxt)
2305 {
2306 	int seg = ctxt->src2.val;
2307 	unsigned short sel;
2308 	int rc;
2309 
2310 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
2311 
2312 	rc = load_segment_descriptor(ctxt, sel, seg);
2313 	if (rc != X86EMUL_CONTINUE)
2314 		return rc;
2315 
2316 	ctxt->dst.val = ctxt->src.val;
2317 	return rc;
2318 }
2319 
2320 static int em_rsm(struct x86_emulate_ctxt *ctxt)
2321 {
2322 	if (!ctxt->ops->is_smm(ctxt))
2323 		return emulate_ud(ctxt);
2324 
2325 	if (ctxt->ops->leave_smm(ctxt))
2326 		ctxt->ops->triple_fault(ctxt);
2327 
2328 	return emulator_recalc_and_set_mode(ctxt);
2329 }
2330 
2331 static void
2332 setup_syscalls_segments(struct desc_struct *cs, struct desc_struct *ss)
2333 {
2334 	cs->l = 0;		/* will be adjusted later */
2335 	set_desc_base(cs, 0);	/* flat segment */
2336 	cs->g = 1;		/* 4kb granularity */
2337 	set_desc_limit(cs, 0xfffff);	/* 4GB limit */
2338 	cs->type = 0x0b;	/* Read, Execute, Accessed */
2339 	cs->s = 1;
2340 	cs->dpl = 0;		/* will be adjusted later */
2341 	cs->p = 1;
2342 	cs->d = 1;
2343 	cs->avl = 0;
2344 
2345 	set_desc_base(ss, 0);	/* flat segment */
2346 	set_desc_limit(ss, 0xfffff);	/* 4GB limit */
2347 	ss->g = 1;		/* 4kb granularity */
2348 	ss->s = 1;
2349 	ss->type = 0x03;	/* Read/Write, Accessed */
2350 	ss->d = 1;		/* 32bit stack segment */
2351 	ss->dpl = 0;
2352 	ss->p = 1;
2353 	ss->l = 0;
2354 	ss->avl = 0;
2355 }
2356 
2357 static bool vendor_intel(struct x86_emulate_ctxt *ctxt)
2358 {
2359 	u32 eax, ebx, ecx, edx;
2360 
2361 	eax = ecx = 0;
2362 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2363 	return is_guest_vendor_intel(ebx, ecx, edx);
2364 }
2365 
2366 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt)
2367 {
2368 	const struct x86_emulate_ops *ops = ctxt->ops;
2369 	u32 eax, ebx, ecx, edx;
2370 
2371 	/*
2372 	 * syscall should always be enabled in longmode - so only become
2373 	 * vendor specific (cpuid) if other modes are active...
2374 	 */
2375 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2376 		return true;
2377 
2378 	eax = 0x00000000;
2379 	ecx = 0x00000000;
2380 	ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true);
2381 	/*
2382 	 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a
2383 	 * 64bit guest with a 32bit compat-app running will #UD !! While this
2384 	 * behaviour can be fixed (by emulating) into AMD response - CPUs of
2385 	 * AMD can't behave like Intel.
2386 	 */
2387 	if (is_guest_vendor_intel(ebx, ecx, edx))
2388 		return false;
2389 
2390 	if (is_guest_vendor_amd(ebx, ecx, edx) ||
2391 	    is_guest_vendor_hygon(ebx, ecx, edx))
2392 		return true;
2393 
2394 	/*
2395 	 * default: (not Intel, not AMD, not Hygon), apply Intel's
2396 	 * stricter rules...
2397 	 */
2398 	return false;
2399 }
2400 
2401 static int em_syscall(struct x86_emulate_ctxt *ctxt)
2402 {
2403 	const struct x86_emulate_ops *ops = ctxt->ops;
2404 	struct desc_struct cs, ss;
2405 	u64 msr_data;
2406 	u16 cs_sel, ss_sel;
2407 	u64 efer = 0;
2408 
2409 	/* syscall is not available in real mode */
2410 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2411 	    ctxt->mode == X86EMUL_MODE_VM86)
2412 		return emulate_ud(ctxt);
2413 
2414 	if (!(em_syscall_is_enabled(ctxt)))
2415 		return emulate_ud(ctxt);
2416 
2417 	ops->get_msr(ctxt, MSR_EFER, &efer);
2418 	if (!(efer & EFER_SCE))
2419 		return emulate_ud(ctxt);
2420 
2421 	setup_syscalls_segments(&cs, &ss);
2422 	ops->get_msr(ctxt, MSR_STAR, &msr_data);
2423 	msr_data >>= 32;
2424 	cs_sel = (u16)(msr_data & 0xfffc);
2425 	ss_sel = (u16)(msr_data + 8);
2426 
2427 	if (efer & EFER_LMA) {
2428 		cs.d = 0;
2429 		cs.l = 1;
2430 	}
2431 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2432 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2433 
2434 	*reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip;
2435 	if (efer & EFER_LMA) {
2436 #ifdef CONFIG_X86_64
2437 		*reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags;
2438 
2439 		ops->get_msr(ctxt,
2440 			     ctxt->mode == X86EMUL_MODE_PROT64 ?
2441 			     MSR_LSTAR : MSR_CSTAR, &msr_data);
2442 		ctxt->_eip = msr_data;
2443 
2444 		ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data);
2445 		ctxt->eflags &= ~msr_data;
2446 		ctxt->eflags |= X86_EFLAGS_FIXED;
2447 #endif
2448 	} else {
2449 		/* legacy mode */
2450 		ops->get_msr(ctxt, MSR_STAR, &msr_data);
2451 		ctxt->_eip = (u32)msr_data;
2452 
2453 		ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2454 	}
2455 
2456 	ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
2457 	return X86EMUL_CONTINUE;
2458 }
2459 
2460 static int em_sysenter(struct x86_emulate_ctxt *ctxt)
2461 {
2462 	const struct x86_emulate_ops *ops = ctxt->ops;
2463 	struct desc_struct cs, ss;
2464 	u64 msr_data;
2465 	u16 cs_sel, ss_sel;
2466 	u64 efer = 0;
2467 
2468 	ops->get_msr(ctxt, MSR_EFER, &efer);
2469 	/* inject #GP if in real mode */
2470 	if (ctxt->mode == X86EMUL_MODE_REAL)
2471 		return emulate_gp(ctxt, 0);
2472 
2473 	/*
2474 	 * Not recognized on AMD in compat mode (but is recognized in legacy
2475 	 * mode).
2476 	 */
2477 	if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA)
2478 	    && !vendor_intel(ctxt))
2479 		return emulate_ud(ctxt);
2480 
2481 	/* sysenter/sysexit have not been tested in 64bit mode. */
2482 	if (ctxt->mode == X86EMUL_MODE_PROT64)
2483 		return X86EMUL_UNHANDLEABLE;
2484 
2485 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2486 	if ((msr_data & 0xfffc) == 0x0)
2487 		return emulate_gp(ctxt, 0);
2488 
2489 	setup_syscalls_segments(&cs, &ss);
2490 	ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
2491 	cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK;
2492 	ss_sel = cs_sel + 8;
2493 	if (efer & EFER_LMA) {
2494 		cs.d = 0;
2495 		cs.l = 1;
2496 	}
2497 
2498 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2499 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2500 
2501 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data);
2502 	ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data;
2503 
2504 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data);
2505 	*reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data :
2506 							      (u32)msr_data;
2507 	if (efer & EFER_LMA)
2508 		ctxt->mode = X86EMUL_MODE_PROT64;
2509 
2510 	return X86EMUL_CONTINUE;
2511 }
2512 
2513 static int em_sysexit(struct x86_emulate_ctxt *ctxt)
2514 {
2515 	const struct x86_emulate_ops *ops = ctxt->ops;
2516 	struct desc_struct cs, ss;
2517 	u64 msr_data, rcx, rdx;
2518 	int usermode;
2519 	u16 cs_sel = 0, ss_sel = 0;
2520 
2521 	/* inject #GP if in real mode or Virtual 8086 mode */
2522 	if (ctxt->mode == X86EMUL_MODE_REAL ||
2523 	    ctxt->mode == X86EMUL_MODE_VM86)
2524 		return emulate_gp(ctxt, 0);
2525 
2526 	setup_syscalls_segments(&cs, &ss);
2527 
2528 	if ((ctxt->rex_prefix & 0x8) != 0x0)
2529 		usermode = X86EMUL_MODE_PROT64;
2530 	else
2531 		usermode = X86EMUL_MODE_PROT32;
2532 
2533 	rcx = reg_read(ctxt, VCPU_REGS_RCX);
2534 	rdx = reg_read(ctxt, VCPU_REGS_RDX);
2535 
2536 	cs.dpl = 3;
2537 	ss.dpl = 3;
2538 	ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data);
2539 	switch (usermode) {
2540 	case X86EMUL_MODE_PROT32:
2541 		cs_sel = (u16)(msr_data + 16);
2542 		if ((msr_data & 0xfffc) == 0x0)
2543 			return emulate_gp(ctxt, 0);
2544 		ss_sel = (u16)(msr_data + 24);
2545 		rcx = (u32)rcx;
2546 		rdx = (u32)rdx;
2547 		break;
2548 	case X86EMUL_MODE_PROT64:
2549 		cs_sel = (u16)(msr_data + 32);
2550 		if (msr_data == 0x0)
2551 			return emulate_gp(ctxt, 0);
2552 		ss_sel = cs_sel + 8;
2553 		cs.d = 0;
2554 		cs.l = 1;
2555 		if (emul_is_noncanonical_address(rcx, ctxt) ||
2556 		    emul_is_noncanonical_address(rdx, ctxt))
2557 			return emulate_gp(ctxt, 0);
2558 		break;
2559 	}
2560 	cs_sel |= SEGMENT_RPL_MASK;
2561 	ss_sel |= SEGMENT_RPL_MASK;
2562 
2563 	ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS);
2564 	ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS);
2565 
2566 	ctxt->_eip = rdx;
2567 	ctxt->mode = usermode;
2568 	*reg_write(ctxt, VCPU_REGS_RSP) = rcx;
2569 
2570 	return X86EMUL_CONTINUE;
2571 }
2572 
2573 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt)
2574 {
2575 	int iopl;
2576 	if (ctxt->mode == X86EMUL_MODE_REAL)
2577 		return false;
2578 	if (ctxt->mode == X86EMUL_MODE_VM86)
2579 		return true;
2580 	iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT;
2581 	return ctxt->ops->cpl(ctxt) > iopl;
2582 }
2583 
2584 #define VMWARE_PORT_VMPORT	(0x5658)
2585 #define VMWARE_PORT_VMRPC	(0x5659)
2586 
2587 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt,
2588 					    u16 port, u16 len)
2589 {
2590 	const struct x86_emulate_ops *ops = ctxt->ops;
2591 	struct desc_struct tr_seg;
2592 	u32 base3;
2593 	int r;
2594 	u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7;
2595 	unsigned mask = (1 << len) - 1;
2596 	unsigned long base;
2597 
2598 	/*
2599 	 * VMware allows access to these ports even if denied
2600 	 * by TSS I/O permission bitmap. Mimic behavior.
2601 	 */
2602 	if (enable_vmware_backdoor &&
2603 	    ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC)))
2604 		return true;
2605 
2606 	ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR);
2607 	if (!tr_seg.p)
2608 		return false;
2609 	if (desc_limit_scaled(&tr_seg) < 103)
2610 		return false;
2611 	base = get_desc_base(&tr_seg);
2612 #ifdef CONFIG_X86_64
2613 	base |= ((u64)base3) << 32;
2614 #endif
2615 	r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true);
2616 	if (r != X86EMUL_CONTINUE)
2617 		return false;
2618 	if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg))
2619 		return false;
2620 	r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true);
2621 	if (r != X86EMUL_CONTINUE)
2622 		return false;
2623 	if ((perm >> bit_idx) & mask)
2624 		return false;
2625 	return true;
2626 }
2627 
2628 static bool emulator_io_permitted(struct x86_emulate_ctxt *ctxt,
2629 				  u16 port, u16 len)
2630 {
2631 	if (ctxt->perm_ok)
2632 		return true;
2633 
2634 	if (emulator_bad_iopl(ctxt))
2635 		if (!emulator_io_port_access_allowed(ctxt, port, len))
2636 			return false;
2637 
2638 	ctxt->perm_ok = true;
2639 
2640 	return true;
2641 }
2642 
2643 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt)
2644 {
2645 	/*
2646 	 * Intel CPUs mask the counter and pointers in quite strange
2647 	 * manner when ECX is zero due to REP-string optimizations.
2648 	 */
2649 #ifdef CONFIG_X86_64
2650 	if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt))
2651 		return;
2652 
2653 	*reg_write(ctxt, VCPU_REGS_RCX) = 0;
2654 
2655 	switch (ctxt->b) {
2656 	case 0xa4:	/* movsb */
2657 	case 0xa5:	/* movsd/w */
2658 		*reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1;
2659 		fallthrough;
2660 	case 0xaa:	/* stosb */
2661 	case 0xab:	/* stosd/w */
2662 		*reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1;
2663 	}
2664 #endif
2665 }
2666 
2667 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt,
2668 				struct tss_segment_16 *tss)
2669 {
2670 	tss->ip = ctxt->_eip;
2671 	tss->flag = ctxt->eflags;
2672 	tss->ax = reg_read(ctxt, VCPU_REGS_RAX);
2673 	tss->cx = reg_read(ctxt, VCPU_REGS_RCX);
2674 	tss->dx = reg_read(ctxt, VCPU_REGS_RDX);
2675 	tss->bx = reg_read(ctxt, VCPU_REGS_RBX);
2676 	tss->sp = reg_read(ctxt, VCPU_REGS_RSP);
2677 	tss->bp = reg_read(ctxt, VCPU_REGS_RBP);
2678 	tss->si = reg_read(ctxt, VCPU_REGS_RSI);
2679 	tss->di = reg_read(ctxt, VCPU_REGS_RDI);
2680 
2681 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2682 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2683 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2684 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2685 	tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR);
2686 }
2687 
2688 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt,
2689 				 struct tss_segment_16 *tss)
2690 {
2691 	int ret;
2692 	u8 cpl;
2693 
2694 	ctxt->_eip = tss->ip;
2695 	ctxt->eflags = tss->flag | 2;
2696 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->ax;
2697 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->cx;
2698 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->dx;
2699 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->bx;
2700 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->sp;
2701 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->bp;
2702 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->si;
2703 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->di;
2704 
2705 	/*
2706 	 * SDM says that segment selectors are loaded before segment
2707 	 * descriptors
2708 	 */
2709 	set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR);
2710 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2711 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2712 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2713 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2714 
2715 	cpl = tss->cs & 3;
2716 
2717 	/*
2718 	 * Now load segment descriptors. If fault happens at this stage
2719 	 * it is handled in a context of new task
2720 	 */
2721 	ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl,
2722 					X86_TRANSFER_TASK_SWITCH, NULL);
2723 	if (ret != X86EMUL_CONTINUE)
2724 		return ret;
2725 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
2726 					X86_TRANSFER_TASK_SWITCH, NULL);
2727 	if (ret != X86EMUL_CONTINUE)
2728 		return ret;
2729 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
2730 					X86_TRANSFER_TASK_SWITCH, NULL);
2731 	if (ret != X86EMUL_CONTINUE)
2732 		return ret;
2733 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
2734 					X86_TRANSFER_TASK_SWITCH, NULL);
2735 	if (ret != X86EMUL_CONTINUE)
2736 		return ret;
2737 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
2738 					X86_TRANSFER_TASK_SWITCH, NULL);
2739 	if (ret != X86EMUL_CONTINUE)
2740 		return ret;
2741 
2742 	return X86EMUL_CONTINUE;
2743 }
2744 
2745 static int task_switch_16(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
2746 			  ulong old_tss_base, struct desc_struct *new_desc)
2747 {
2748 	struct tss_segment_16 tss_seg;
2749 	int ret;
2750 	u32 new_tss_base = get_desc_base(new_desc);
2751 
2752 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2753 	if (ret != X86EMUL_CONTINUE)
2754 		return ret;
2755 
2756 	save_state_to_tss16(ctxt, &tss_seg);
2757 
2758 	ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2759 	if (ret != X86EMUL_CONTINUE)
2760 		return ret;
2761 
2762 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
2763 	if (ret != X86EMUL_CONTINUE)
2764 		return ret;
2765 
2766 	if (old_tss_sel != 0xffff) {
2767 		tss_seg.prev_task_link = old_tss_sel;
2768 
2769 		ret = linear_write_system(ctxt, new_tss_base,
2770 					  &tss_seg.prev_task_link,
2771 					  sizeof(tss_seg.prev_task_link));
2772 		if (ret != X86EMUL_CONTINUE)
2773 			return ret;
2774 	}
2775 
2776 	return load_state_from_tss16(ctxt, &tss_seg);
2777 }
2778 
2779 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt,
2780 				struct tss_segment_32 *tss)
2781 {
2782 	/* CR3 and ldt selector are not saved intentionally */
2783 	tss->eip = ctxt->_eip;
2784 	tss->eflags = ctxt->eflags;
2785 	tss->eax = reg_read(ctxt, VCPU_REGS_RAX);
2786 	tss->ecx = reg_read(ctxt, VCPU_REGS_RCX);
2787 	tss->edx = reg_read(ctxt, VCPU_REGS_RDX);
2788 	tss->ebx = reg_read(ctxt, VCPU_REGS_RBX);
2789 	tss->esp = reg_read(ctxt, VCPU_REGS_RSP);
2790 	tss->ebp = reg_read(ctxt, VCPU_REGS_RBP);
2791 	tss->esi = reg_read(ctxt, VCPU_REGS_RSI);
2792 	tss->edi = reg_read(ctxt, VCPU_REGS_RDI);
2793 
2794 	tss->es = get_segment_selector(ctxt, VCPU_SREG_ES);
2795 	tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS);
2796 	tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS);
2797 	tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS);
2798 	tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS);
2799 	tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS);
2800 }
2801 
2802 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt,
2803 				 struct tss_segment_32 *tss)
2804 {
2805 	int ret;
2806 	u8 cpl;
2807 
2808 	if (ctxt->ops->set_cr(ctxt, 3, tss->cr3))
2809 		return emulate_gp(ctxt, 0);
2810 	ctxt->_eip = tss->eip;
2811 	ctxt->eflags = tss->eflags | 2;
2812 
2813 	/* General purpose registers */
2814 	*reg_write(ctxt, VCPU_REGS_RAX) = tss->eax;
2815 	*reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx;
2816 	*reg_write(ctxt, VCPU_REGS_RDX) = tss->edx;
2817 	*reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx;
2818 	*reg_write(ctxt, VCPU_REGS_RSP) = tss->esp;
2819 	*reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp;
2820 	*reg_write(ctxt, VCPU_REGS_RSI) = tss->esi;
2821 	*reg_write(ctxt, VCPU_REGS_RDI) = tss->edi;
2822 
2823 	/*
2824 	 * SDM says that segment selectors are loaded before segment
2825 	 * descriptors.  This is important because CPL checks will
2826 	 * use CS.RPL.
2827 	 */
2828 	set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR);
2829 	set_segment_selector(ctxt, tss->es, VCPU_SREG_ES);
2830 	set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS);
2831 	set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS);
2832 	set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS);
2833 	set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS);
2834 	set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS);
2835 
2836 	/*
2837 	 * If we're switching between Protected Mode and VM86, we need to make
2838 	 * sure to update the mode before loading the segment descriptors so
2839 	 * that the selectors are interpreted correctly.
2840 	 */
2841 	if (ctxt->eflags & X86_EFLAGS_VM) {
2842 		ctxt->mode = X86EMUL_MODE_VM86;
2843 		cpl = 3;
2844 	} else {
2845 		ctxt->mode = X86EMUL_MODE_PROT32;
2846 		cpl = tss->cs & 3;
2847 	}
2848 
2849 	/*
2850 	 * Now load segment descriptors. If fault happens at this stage
2851 	 * it is handled in a context of new task
2852 	 */
2853 	ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR,
2854 					cpl, X86_TRANSFER_TASK_SWITCH, NULL);
2855 	if (ret != X86EMUL_CONTINUE)
2856 		return ret;
2857 	ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl,
2858 					X86_TRANSFER_TASK_SWITCH, NULL);
2859 	if (ret != X86EMUL_CONTINUE)
2860 		return ret;
2861 	ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl,
2862 					X86_TRANSFER_TASK_SWITCH, NULL);
2863 	if (ret != X86EMUL_CONTINUE)
2864 		return ret;
2865 	ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl,
2866 					X86_TRANSFER_TASK_SWITCH, NULL);
2867 	if (ret != X86EMUL_CONTINUE)
2868 		return ret;
2869 	ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl,
2870 					X86_TRANSFER_TASK_SWITCH, NULL);
2871 	if (ret != X86EMUL_CONTINUE)
2872 		return ret;
2873 	ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl,
2874 					X86_TRANSFER_TASK_SWITCH, NULL);
2875 	if (ret != X86EMUL_CONTINUE)
2876 		return ret;
2877 	ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl,
2878 					X86_TRANSFER_TASK_SWITCH, NULL);
2879 
2880 	return ret;
2881 }
2882 
2883 static int task_switch_32(struct x86_emulate_ctxt *ctxt, u16 old_tss_sel,
2884 			  ulong old_tss_base, struct desc_struct *new_desc)
2885 {
2886 	struct tss_segment_32 tss_seg;
2887 	int ret;
2888 	u32 new_tss_base = get_desc_base(new_desc);
2889 	u32 eip_offset = offsetof(struct tss_segment_32, eip);
2890 	u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector);
2891 
2892 	ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg));
2893 	if (ret != X86EMUL_CONTINUE)
2894 		return ret;
2895 
2896 	save_state_to_tss32(ctxt, &tss_seg);
2897 
2898 	/* Only GP registers and segment selectors are saved */
2899 	ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip,
2900 				  ldt_sel_offset - eip_offset);
2901 	if (ret != X86EMUL_CONTINUE)
2902 		return ret;
2903 
2904 	ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg));
2905 	if (ret != X86EMUL_CONTINUE)
2906 		return ret;
2907 
2908 	if (old_tss_sel != 0xffff) {
2909 		tss_seg.prev_task_link = old_tss_sel;
2910 
2911 		ret = linear_write_system(ctxt, new_tss_base,
2912 					  &tss_seg.prev_task_link,
2913 					  sizeof(tss_seg.prev_task_link));
2914 		if (ret != X86EMUL_CONTINUE)
2915 			return ret;
2916 	}
2917 
2918 	return load_state_from_tss32(ctxt, &tss_seg);
2919 }
2920 
2921 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt,
2922 				   u16 tss_selector, int idt_index, int reason,
2923 				   bool has_error_code, u32 error_code)
2924 {
2925 	const struct x86_emulate_ops *ops = ctxt->ops;
2926 	struct desc_struct curr_tss_desc, next_tss_desc;
2927 	int ret;
2928 	u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR);
2929 	ulong old_tss_base =
2930 		ops->get_cached_segment_base(ctxt, VCPU_SREG_TR);
2931 	u32 desc_limit;
2932 	ulong desc_addr, dr7;
2933 
2934 	/* FIXME: old_tss_base == ~0 ? */
2935 
2936 	ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr);
2937 	if (ret != X86EMUL_CONTINUE)
2938 		return ret;
2939 	ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr);
2940 	if (ret != X86EMUL_CONTINUE)
2941 		return ret;
2942 
2943 	/* FIXME: check that next_tss_desc is tss */
2944 
2945 	/*
2946 	 * Check privileges. The three cases are task switch caused by...
2947 	 *
2948 	 * 1. jmp/call/int to task gate: Check against DPL of the task gate
2949 	 * 2. Exception/IRQ/iret: No check is performed
2950 	 * 3. jmp/call to TSS/task-gate: No check is performed since the
2951 	 *    hardware checks it before exiting.
2952 	 */
2953 	if (reason == TASK_SWITCH_GATE) {
2954 		if (idt_index != -1) {
2955 			/* Software interrupts */
2956 			struct desc_struct task_gate_desc;
2957 			int dpl;
2958 
2959 			ret = read_interrupt_descriptor(ctxt, idt_index,
2960 							&task_gate_desc);
2961 			if (ret != X86EMUL_CONTINUE)
2962 				return ret;
2963 
2964 			dpl = task_gate_desc.dpl;
2965 			if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl)
2966 				return emulate_gp(ctxt, (idt_index << 3) | 0x2);
2967 		}
2968 	}
2969 
2970 	desc_limit = desc_limit_scaled(&next_tss_desc);
2971 	if (!next_tss_desc.p ||
2972 	    ((desc_limit < 0x67 && (next_tss_desc.type & 8)) ||
2973 	     desc_limit < 0x2b)) {
2974 		return emulate_ts(ctxt, tss_selector & 0xfffc);
2975 	}
2976 
2977 	if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
2978 		curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */
2979 		write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc);
2980 	}
2981 
2982 	if (reason == TASK_SWITCH_IRET)
2983 		ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT;
2984 
2985 	/* set back link to prev task only if NT bit is set in eflags
2986 	   note that old_tss_sel is not used after this point */
2987 	if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
2988 		old_tss_sel = 0xffff;
2989 
2990 	if (next_tss_desc.type & 8)
2991 		ret = task_switch_32(ctxt, old_tss_sel, old_tss_base, &next_tss_desc);
2992 	else
2993 		ret = task_switch_16(ctxt, old_tss_sel,
2994 				     old_tss_base, &next_tss_desc);
2995 	if (ret != X86EMUL_CONTINUE)
2996 		return ret;
2997 
2998 	if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE)
2999 		ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT;
3000 
3001 	if (reason != TASK_SWITCH_IRET) {
3002 		next_tss_desc.type |= (1 << 1); /* set busy flag */
3003 		write_segment_descriptor(ctxt, tss_selector, &next_tss_desc);
3004 	}
3005 
3006 	ops->set_cr(ctxt, 0,  ops->get_cr(ctxt, 0) | X86_CR0_TS);
3007 	ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR);
3008 
3009 	if (has_error_code) {
3010 		ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2;
3011 		ctxt->lock_prefix = 0;
3012 		ctxt->src.val = (unsigned long) error_code;
3013 		ret = em_push(ctxt);
3014 	}
3015 
3016 	dr7 = ops->get_dr(ctxt, 7);
3017 	ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN));
3018 
3019 	return ret;
3020 }
3021 
3022 int emulator_task_switch(struct x86_emulate_ctxt *ctxt,
3023 			 u16 tss_selector, int idt_index, int reason,
3024 			 bool has_error_code, u32 error_code)
3025 {
3026 	int rc;
3027 
3028 	invalidate_registers(ctxt);
3029 	ctxt->_eip = ctxt->eip;
3030 	ctxt->dst.type = OP_NONE;
3031 
3032 	rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason,
3033 				     has_error_code, error_code);
3034 
3035 	if (rc == X86EMUL_CONTINUE) {
3036 		ctxt->eip = ctxt->_eip;
3037 		writeback_registers(ctxt);
3038 	}
3039 
3040 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
3041 }
3042 
3043 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg,
3044 		struct operand *op)
3045 {
3046 	int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count;
3047 
3048 	register_address_increment(ctxt, reg, df * op->bytes);
3049 	op->addr.mem.ea = register_address(ctxt, reg);
3050 }
3051 
3052 static int em_das(struct x86_emulate_ctxt *ctxt)
3053 {
3054 	u8 al, old_al;
3055 	bool af, cf, old_cf;
3056 
3057 	cf = ctxt->eflags & X86_EFLAGS_CF;
3058 	al = ctxt->dst.val;
3059 
3060 	old_al = al;
3061 	old_cf = cf;
3062 	cf = false;
3063 	af = ctxt->eflags & X86_EFLAGS_AF;
3064 	if ((al & 0x0f) > 9 || af) {
3065 		al -= 6;
3066 		cf = old_cf | (al >= 250);
3067 		af = true;
3068 	} else {
3069 		af = false;
3070 	}
3071 	if (old_al > 0x99 || old_cf) {
3072 		al -= 0x60;
3073 		cf = true;
3074 	}
3075 
3076 	ctxt->dst.val = al;
3077 	/* Set PF, ZF, SF */
3078 	ctxt->src.type = OP_IMM;
3079 	ctxt->src.val = 0;
3080 	ctxt->src.bytes = 1;
3081 	fastop(ctxt, em_or);
3082 	ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF);
3083 	if (cf)
3084 		ctxt->eflags |= X86_EFLAGS_CF;
3085 	if (af)
3086 		ctxt->eflags |= X86_EFLAGS_AF;
3087 	return X86EMUL_CONTINUE;
3088 }
3089 
3090 static int em_aam(struct x86_emulate_ctxt *ctxt)
3091 {
3092 	u8 al, ah;
3093 
3094 	if (ctxt->src.val == 0)
3095 		return emulate_de(ctxt);
3096 
3097 	al = ctxt->dst.val & 0xff;
3098 	ah = al / ctxt->src.val;
3099 	al %= ctxt->src.val;
3100 
3101 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8);
3102 
3103 	/* Set PF, ZF, SF */
3104 	ctxt->src.type = OP_IMM;
3105 	ctxt->src.val = 0;
3106 	ctxt->src.bytes = 1;
3107 	fastop(ctxt, em_or);
3108 
3109 	return X86EMUL_CONTINUE;
3110 }
3111 
3112 static int em_aad(struct x86_emulate_ctxt *ctxt)
3113 {
3114 	u8 al = ctxt->dst.val & 0xff;
3115 	u8 ah = (ctxt->dst.val >> 8) & 0xff;
3116 
3117 	al = (al + (ah * ctxt->src.val)) & 0xff;
3118 
3119 	ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al;
3120 
3121 	/* Set PF, ZF, SF */
3122 	ctxt->src.type = OP_IMM;
3123 	ctxt->src.val = 0;
3124 	ctxt->src.bytes = 1;
3125 	fastop(ctxt, em_or);
3126 
3127 	return X86EMUL_CONTINUE;
3128 }
3129 
3130 static int em_call(struct x86_emulate_ctxt *ctxt)
3131 {
3132 	int rc;
3133 	long rel = ctxt->src.val;
3134 
3135 	ctxt->src.val = (unsigned long)ctxt->_eip;
3136 	rc = jmp_rel(ctxt, rel);
3137 	if (rc != X86EMUL_CONTINUE)
3138 		return rc;
3139 	return em_push(ctxt);
3140 }
3141 
3142 static int em_call_far(struct x86_emulate_ctxt *ctxt)
3143 {
3144 	u16 sel, old_cs;
3145 	ulong old_eip;
3146 	int rc;
3147 	struct desc_struct old_desc, new_desc;
3148 	const struct x86_emulate_ops *ops = ctxt->ops;
3149 	int cpl = ctxt->ops->cpl(ctxt);
3150 	enum x86emul_mode prev_mode = ctxt->mode;
3151 
3152 	old_eip = ctxt->_eip;
3153 	ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS);
3154 
3155 	memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2);
3156 	rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl,
3157 				       X86_TRANSFER_CALL_JMP, &new_desc);
3158 	if (rc != X86EMUL_CONTINUE)
3159 		return rc;
3160 
3161 	rc = assign_eip_far(ctxt, ctxt->src.val);
3162 	if (rc != X86EMUL_CONTINUE)
3163 		goto fail;
3164 
3165 	ctxt->src.val = old_cs;
3166 	rc = em_push(ctxt);
3167 	if (rc != X86EMUL_CONTINUE)
3168 		goto fail;
3169 
3170 	ctxt->src.val = old_eip;
3171 	rc = em_push(ctxt);
3172 	/* If we failed, we tainted the memory, but the very least we should
3173 	   restore cs */
3174 	if (rc != X86EMUL_CONTINUE) {
3175 		pr_warn_once("faulting far call emulation tainted memory\n");
3176 		goto fail;
3177 	}
3178 	return rc;
3179 fail:
3180 	ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS);
3181 	ctxt->mode = prev_mode;
3182 	return rc;
3183 
3184 }
3185 
3186 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt)
3187 {
3188 	int rc;
3189 	unsigned long eip = 0;
3190 
3191 	rc = emulate_pop(ctxt, &eip, ctxt->op_bytes);
3192 	if (rc != X86EMUL_CONTINUE)
3193 		return rc;
3194 	rc = assign_eip_near(ctxt, eip);
3195 	if (rc != X86EMUL_CONTINUE)
3196 		return rc;
3197 	rsp_increment(ctxt, ctxt->src.val);
3198 	return X86EMUL_CONTINUE;
3199 }
3200 
3201 static int em_xchg(struct x86_emulate_ctxt *ctxt)
3202 {
3203 	/* Write back the register source. */
3204 	ctxt->src.val = ctxt->dst.val;
3205 	write_register_operand(&ctxt->src);
3206 
3207 	/* Write back the memory destination with implicit LOCK prefix. */
3208 	ctxt->dst.val = ctxt->src.orig_val;
3209 	ctxt->lock_prefix = 1;
3210 	return X86EMUL_CONTINUE;
3211 }
3212 
3213 static int em_imul_3op(struct x86_emulate_ctxt *ctxt)
3214 {
3215 	ctxt->dst.val = ctxt->src2.val;
3216 	return fastop(ctxt, em_imul);
3217 }
3218 
3219 static int em_cwd(struct x86_emulate_ctxt *ctxt)
3220 {
3221 	ctxt->dst.type = OP_REG;
3222 	ctxt->dst.bytes = ctxt->src.bytes;
3223 	ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
3224 	ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1);
3225 
3226 	return X86EMUL_CONTINUE;
3227 }
3228 
3229 static int em_rdpid(struct x86_emulate_ctxt *ctxt)
3230 {
3231 	u64 tsc_aux = 0;
3232 
3233 	if (!ctxt->ops->guest_has_rdpid(ctxt))
3234 		return emulate_ud(ctxt);
3235 
3236 	ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux);
3237 	ctxt->dst.val = tsc_aux;
3238 	return X86EMUL_CONTINUE;
3239 }
3240 
3241 static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
3242 {
3243 	u64 tsc = 0;
3244 
3245 	ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc);
3246 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc;
3247 	*reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32;
3248 	return X86EMUL_CONTINUE;
3249 }
3250 
3251 static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
3252 {
3253 	u64 pmc;
3254 
3255 	if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc))
3256 		return emulate_gp(ctxt, 0);
3257 	*reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc;
3258 	*reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32;
3259 	return X86EMUL_CONTINUE;
3260 }
3261 
3262 static int em_mov(struct x86_emulate_ctxt *ctxt)
3263 {
3264 	memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr));
3265 	return X86EMUL_CONTINUE;
3266 }
3267 
3268 static int em_movbe(struct x86_emulate_ctxt *ctxt)
3269 {
3270 	u16 tmp;
3271 
3272 	if (!ctxt->ops->guest_has_movbe(ctxt))
3273 		return emulate_ud(ctxt);
3274 
3275 	switch (ctxt->op_bytes) {
3276 	case 2:
3277 		/*
3278 		 * From MOVBE definition: "...When the operand size is 16 bits,
3279 		 * the upper word of the destination register remains unchanged
3280 		 * ..."
3281 		 *
3282 		 * Both casting ->valptr and ->val to u16 breaks strict aliasing
3283 		 * rules so we have to do the operation almost per hand.
3284 		 */
3285 		tmp = (u16)ctxt->src.val;
3286 		ctxt->dst.val &= ~0xffffUL;
3287 		ctxt->dst.val |= (unsigned long)swab16(tmp);
3288 		break;
3289 	case 4:
3290 		ctxt->dst.val = swab32((u32)ctxt->src.val);
3291 		break;
3292 	case 8:
3293 		ctxt->dst.val = swab64(ctxt->src.val);
3294 		break;
3295 	default:
3296 		BUG();
3297 	}
3298 	return X86EMUL_CONTINUE;
3299 }
3300 
3301 static int em_cr_write(struct x86_emulate_ctxt *ctxt)
3302 {
3303 	int cr_num = ctxt->modrm_reg;
3304 	int r;
3305 
3306 	if (ctxt->ops->set_cr(ctxt, cr_num, ctxt->src.val))
3307 		return emulate_gp(ctxt, 0);
3308 
3309 	/* Disable writeback. */
3310 	ctxt->dst.type = OP_NONE;
3311 
3312 	if (cr_num == 0) {
3313 		/*
3314 		 * CR0 write might have updated CR0.PE and/or CR0.PG
3315 		 * which can affect the cpu's execution mode.
3316 		 */
3317 		r = emulator_recalc_and_set_mode(ctxt);
3318 		if (r != X86EMUL_CONTINUE)
3319 			return r;
3320 	}
3321 
3322 	return X86EMUL_CONTINUE;
3323 }
3324 
3325 static int em_dr_write(struct x86_emulate_ctxt *ctxt)
3326 {
3327 	unsigned long val;
3328 
3329 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3330 		val = ctxt->src.val & ~0ULL;
3331 	else
3332 		val = ctxt->src.val & ~0U;
3333 
3334 	/* #UD condition is already handled. */
3335 	if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
3336 		return emulate_gp(ctxt, 0);
3337 
3338 	/* Disable writeback. */
3339 	ctxt->dst.type = OP_NONE;
3340 	return X86EMUL_CONTINUE;
3341 }
3342 
3343 static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
3344 {
3345 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3346 	u64 msr_data;
3347 	int r;
3348 
3349 	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
3350 		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
3351 	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);
3352 
3353 	if (r == X86EMUL_PROPAGATE_FAULT)
3354 		return emulate_gp(ctxt, 0);
3355 
3356 	return r;
3357 }
3358 
3359 static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
3360 {
3361 	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
3362 	u64 msr_data;
3363 	int r;
3364 
3365 	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);
3366 
3367 	if (r == X86EMUL_PROPAGATE_FAULT)
3368 		return emulate_gp(ctxt, 0);
3369 
3370 	if (r == X86EMUL_CONTINUE) {
3371 		*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
3372 		*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
3373 	}
3374 	return r;
3375 }
3376 
3377 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment)
3378 {
3379 	if (segment > VCPU_SREG_GS &&
3380 	    (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3381 	    ctxt->ops->cpl(ctxt) > 0)
3382 		return emulate_gp(ctxt, 0);
3383 
3384 	ctxt->dst.val = get_segment_selector(ctxt, segment);
3385 	if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM)
3386 		ctxt->dst.bytes = 2;
3387 	return X86EMUL_CONTINUE;
3388 }
3389 
3390 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
3391 {
3392 	if (ctxt->modrm_reg > VCPU_SREG_GS)
3393 		return emulate_ud(ctxt);
3394 
3395 	return em_store_sreg(ctxt, ctxt->modrm_reg);
3396 }
3397 
3398 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt)
3399 {
3400 	u16 sel = ctxt->src.val;
3401 
3402 	if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS)
3403 		return emulate_ud(ctxt);
3404 
3405 	if (ctxt->modrm_reg == VCPU_SREG_SS)
3406 		ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS;
3407 
3408 	/* Disable writeback. */
3409 	ctxt->dst.type = OP_NONE;
3410 	return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg);
3411 }
3412 
3413 static int em_sldt(struct x86_emulate_ctxt *ctxt)
3414 {
3415 	return em_store_sreg(ctxt, VCPU_SREG_LDTR);
3416 }
3417 
3418 static int em_lldt(struct x86_emulate_ctxt *ctxt)
3419 {
3420 	u16 sel = ctxt->src.val;
3421 
3422 	/* Disable writeback. */
3423 	ctxt->dst.type = OP_NONE;
3424 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR);
3425 }
3426 
3427 static int em_str(struct x86_emulate_ctxt *ctxt)
3428 {
3429 	return em_store_sreg(ctxt, VCPU_SREG_TR);
3430 }
3431 
3432 static int em_ltr(struct x86_emulate_ctxt *ctxt)
3433 {
3434 	u16 sel = ctxt->src.val;
3435 
3436 	/* Disable writeback. */
3437 	ctxt->dst.type = OP_NONE;
3438 	return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR);
3439 }
3440 
3441 static int em_invlpg(struct x86_emulate_ctxt *ctxt)
3442 {
3443 	int rc;
3444 	ulong linear;
3445 	unsigned int max_size;
3446 
3447 	rc = __linearize(ctxt, ctxt->src.addr.mem, &max_size, 1, ctxt->mode,
3448 			 &linear, X86EMUL_F_INVLPG);
3449 	if (rc == X86EMUL_CONTINUE)
3450 		ctxt->ops->invlpg(ctxt, linear);
3451 	/* Disable writeback. */
3452 	ctxt->dst.type = OP_NONE;
3453 	return X86EMUL_CONTINUE;
3454 }
3455 
3456 static int em_clts(struct x86_emulate_ctxt *ctxt)
3457 {
3458 	ulong cr0;
3459 
3460 	cr0 = ctxt->ops->get_cr(ctxt, 0);
3461 	cr0 &= ~X86_CR0_TS;
3462 	ctxt->ops->set_cr(ctxt, 0, cr0);
3463 	return X86EMUL_CONTINUE;
3464 }
3465 
3466 static int em_hypercall(struct x86_emulate_ctxt *ctxt)
3467 {
3468 	int rc = ctxt->ops->fix_hypercall(ctxt);
3469 
3470 	if (rc != X86EMUL_CONTINUE)
3471 		return rc;
3472 
3473 	/* Let the processor re-execute the fixed hypercall */
3474 	ctxt->_eip = ctxt->eip;
3475 	/* Disable writeback. */
3476 	ctxt->dst.type = OP_NONE;
3477 	return X86EMUL_CONTINUE;
3478 }
3479 
3480 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt,
3481 				  void (*get)(struct x86_emulate_ctxt *ctxt,
3482 					      struct desc_ptr *ptr))
3483 {
3484 	struct desc_ptr desc_ptr;
3485 
3486 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3487 	    ctxt->ops->cpl(ctxt) > 0)
3488 		return emulate_gp(ctxt, 0);
3489 
3490 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3491 		ctxt->op_bytes = 8;
3492 	get(ctxt, &desc_ptr);
3493 	if (ctxt->op_bytes == 2) {
3494 		ctxt->op_bytes = 4;
3495 		desc_ptr.address &= 0x00ffffff;
3496 	}
3497 	/* Disable writeback. */
3498 	ctxt->dst.type = OP_NONE;
3499 	return segmented_write_std(ctxt, ctxt->dst.addr.mem,
3500 				   &desc_ptr, 2 + ctxt->op_bytes);
3501 }
3502 
3503 static int em_sgdt(struct x86_emulate_ctxt *ctxt)
3504 {
3505 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt);
3506 }
3507 
3508 static int em_sidt(struct x86_emulate_ctxt *ctxt)
3509 {
3510 	return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt);
3511 }
3512 
3513 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt)
3514 {
3515 	struct desc_ptr desc_ptr;
3516 	int rc;
3517 
3518 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3519 		ctxt->op_bytes = 8;
3520 	rc = read_descriptor(ctxt, ctxt->src.addr.mem,
3521 			     &desc_ptr.size, &desc_ptr.address,
3522 			     ctxt->op_bytes);
3523 	if (rc != X86EMUL_CONTINUE)
3524 		return rc;
3525 	if (ctxt->mode == X86EMUL_MODE_PROT64 &&
3526 	    emul_is_noncanonical_address(desc_ptr.address, ctxt))
3527 		return emulate_gp(ctxt, 0);
3528 	if (lgdt)
3529 		ctxt->ops->set_gdt(ctxt, &desc_ptr);
3530 	else
3531 		ctxt->ops->set_idt(ctxt, &desc_ptr);
3532 	/* Disable writeback. */
3533 	ctxt->dst.type = OP_NONE;
3534 	return X86EMUL_CONTINUE;
3535 }
3536 
3537 static int em_lgdt(struct x86_emulate_ctxt *ctxt)
3538 {
3539 	return em_lgdt_lidt(ctxt, true);
3540 }
3541 
3542 static int em_lidt(struct x86_emulate_ctxt *ctxt)
3543 {
3544 	return em_lgdt_lidt(ctxt, false);
3545 }
3546 
3547 static int em_smsw(struct x86_emulate_ctxt *ctxt)
3548 {
3549 	if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) &&
3550 	    ctxt->ops->cpl(ctxt) > 0)
3551 		return emulate_gp(ctxt, 0);
3552 
3553 	if (ctxt->dst.type == OP_MEM)
3554 		ctxt->dst.bytes = 2;
3555 	ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0);
3556 	return X86EMUL_CONTINUE;
3557 }
3558 
3559 static int em_lmsw(struct x86_emulate_ctxt *ctxt)
3560 {
3561 	ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul)
3562 			  | (ctxt->src.val & 0x0f));
3563 	ctxt->dst.type = OP_NONE;
3564 	return X86EMUL_CONTINUE;
3565 }
3566 
3567 static int em_loop(struct x86_emulate_ctxt *ctxt)
3568 {
3569 	int rc = X86EMUL_CONTINUE;
3570 
3571 	register_address_increment(ctxt, VCPU_REGS_RCX, -1);
3572 	if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) &&
3573 	    (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags)))
3574 		rc = jmp_rel(ctxt, ctxt->src.val);
3575 
3576 	return rc;
3577 }
3578 
3579 static int em_jcxz(struct x86_emulate_ctxt *ctxt)
3580 {
3581 	int rc = X86EMUL_CONTINUE;
3582 
3583 	if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0)
3584 		rc = jmp_rel(ctxt, ctxt->src.val);
3585 
3586 	return rc;
3587 }
3588 
3589 static int em_in(struct x86_emulate_ctxt *ctxt)
3590 {
3591 	if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3592 			     &ctxt->dst.val))
3593 		return X86EMUL_IO_NEEDED;
3594 
3595 	return X86EMUL_CONTINUE;
3596 }
3597 
3598 static int em_out(struct x86_emulate_ctxt *ctxt)
3599 {
3600 	ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3601 				    &ctxt->src.val, 1);
3602 	/* Disable writeback. */
3603 	ctxt->dst.type = OP_NONE;
3604 	return X86EMUL_CONTINUE;
3605 }
3606 
3607 static int em_cli(struct x86_emulate_ctxt *ctxt)
3608 {
3609 	if (emulator_bad_iopl(ctxt))
3610 		return emulate_gp(ctxt, 0);
3611 
3612 	ctxt->eflags &= ~X86_EFLAGS_IF;
3613 	return X86EMUL_CONTINUE;
3614 }
3615 
3616 static int em_sti(struct x86_emulate_ctxt *ctxt)
3617 {
3618 	if (emulator_bad_iopl(ctxt))
3619 		return emulate_gp(ctxt, 0);
3620 
3621 	ctxt->interruptibility = KVM_X86_SHADOW_INT_STI;
3622 	ctxt->eflags |= X86_EFLAGS_IF;
3623 	return X86EMUL_CONTINUE;
3624 }
3625 
3626 static int em_cpuid(struct x86_emulate_ctxt *ctxt)
3627 {
3628 	u32 eax, ebx, ecx, edx;
3629 	u64 msr = 0;
3630 
3631 	ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr);
3632 	if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3633 	    ctxt->ops->cpl(ctxt)) {
3634 		return emulate_gp(ctxt, 0);
3635 	}
3636 
3637 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3638 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3639 	ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false);
3640 	*reg_write(ctxt, VCPU_REGS_RAX) = eax;
3641 	*reg_write(ctxt, VCPU_REGS_RBX) = ebx;
3642 	*reg_write(ctxt, VCPU_REGS_RCX) = ecx;
3643 	*reg_write(ctxt, VCPU_REGS_RDX) = edx;
3644 	return X86EMUL_CONTINUE;
3645 }
3646 
3647 static int em_sahf(struct x86_emulate_ctxt *ctxt)
3648 {
3649 	u32 flags;
3650 
3651 	flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
3652 		X86_EFLAGS_SF;
3653 	flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8;
3654 
3655 	ctxt->eflags &= ~0xffUL;
3656 	ctxt->eflags |= flags | X86_EFLAGS_FIXED;
3657 	return X86EMUL_CONTINUE;
3658 }
3659 
3660 static int em_lahf(struct x86_emulate_ctxt *ctxt)
3661 {
3662 	*reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL;
3663 	*reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8;
3664 	return X86EMUL_CONTINUE;
3665 }
3666 
3667 static int em_bswap(struct x86_emulate_ctxt *ctxt)
3668 {
3669 	switch (ctxt->op_bytes) {
3670 #ifdef CONFIG_X86_64
3671 	case 8:
3672 		asm("bswap %0" : "+r"(ctxt->dst.val));
3673 		break;
3674 #endif
3675 	default:
3676 		asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val));
3677 		break;
3678 	}
3679 	return X86EMUL_CONTINUE;
3680 }
3681 
3682 static int em_clflush(struct x86_emulate_ctxt *ctxt)
3683 {
3684 	/* emulating clflush regardless of cpuid */
3685 	return X86EMUL_CONTINUE;
3686 }
3687 
3688 static int em_clflushopt(struct x86_emulate_ctxt *ctxt)
3689 {
3690 	/* emulating clflushopt regardless of cpuid */
3691 	return X86EMUL_CONTINUE;
3692 }
3693 
3694 static int em_movsxd(struct x86_emulate_ctxt *ctxt)
3695 {
3696 	ctxt->dst.val = (s32) ctxt->src.val;
3697 	return X86EMUL_CONTINUE;
3698 }
3699 
3700 static int check_fxsr(struct x86_emulate_ctxt *ctxt)
3701 {
3702 	if (!ctxt->ops->guest_has_fxsr(ctxt))
3703 		return emulate_ud(ctxt);
3704 
3705 	if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM))
3706 		return emulate_nm(ctxt);
3707 
3708 	/*
3709 	 * Don't emulate a case that should never be hit, instead of working
3710 	 * around a lack of fxsave64/fxrstor64 on old compilers.
3711 	 */
3712 	if (ctxt->mode >= X86EMUL_MODE_PROT64)
3713 		return X86EMUL_UNHANDLEABLE;
3714 
3715 	return X86EMUL_CONTINUE;
3716 }
3717 
3718 /*
3719  * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save
3720  * and restore MXCSR.
3721  */
3722 static size_t __fxstate_size(int nregs)
3723 {
3724 	return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16;
3725 }
3726 
3727 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt)
3728 {
3729 	bool cr4_osfxsr;
3730 	if (ctxt->mode == X86EMUL_MODE_PROT64)
3731 		return __fxstate_size(16);
3732 
3733 	cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR;
3734 	return __fxstate_size(cr4_osfxsr ? 8 : 0);
3735 }
3736 
3737 /*
3738  * FXSAVE and FXRSTOR have 4 different formats depending on execution mode,
3739  *  1) 16 bit mode
3740  *  2) 32 bit mode
3741  *     - like (1), but FIP and FDP (foo) are only 16 bit.  At least Intel CPUs
3742  *       preserve whole 32 bit values, though, so (1) and (2) are the same wrt.
3743  *       save and restore
3744  *  3) 64-bit mode with REX.W prefix
3745  *     - like (2), but XMM 8-15 are being saved and restored
3746  *  4) 64-bit mode without REX.W prefix
3747  *     - like (3), but FIP and FDP are 64 bit
3748  *
3749  * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the
3750  * desired result.  (4) is not emulated.
3751  *
3752  * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS
3753  * and FPU DS) should match.
3754  */
3755 static int em_fxsave(struct x86_emulate_ctxt *ctxt)
3756 {
3757 	struct fxregs_state fx_state;
3758 	int rc;
3759 
3760 	rc = check_fxsr(ctxt);
3761 	if (rc != X86EMUL_CONTINUE)
3762 		return rc;
3763 
3764 	kvm_fpu_get();
3765 
3766 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state));
3767 
3768 	kvm_fpu_put();
3769 
3770 	if (rc != X86EMUL_CONTINUE)
3771 		return rc;
3772 
3773 	return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state,
3774 		                   fxstate_size(ctxt));
3775 }
3776 
3777 /*
3778  * FXRSTOR might restore XMM registers not provided by the guest. Fill
3779  * in the host registers (via FXSAVE) instead, so they won't be modified.
3780  * (preemption has to stay disabled until FXRSTOR).
3781  *
3782  * Use noinline to keep the stack for other functions called by callers small.
3783  */
3784 static noinline int fxregs_fixup(struct fxregs_state *fx_state,
3785 				 const size_t used_size)
3786 {
3787 	struct fxregs_state fx_tmp;
3788 	int rc;
3789 
3790 	rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp));
3791 	memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size,
3792 	       __fxstate_size(16) - used_size);
3793 
3794 	return rc;
3795 }
3796 
3797 static int em_fxrstor(struct x86_emulate_ctxt *ctxt)
3798 {
3799 	struct fxregs_state fx_state;
3800 	int rc;
3801 	size_t size;
3802 
3803 	rc = check_fxsr(ctxt);
3804 	if (rc != X86EMUL_CONTINUE)
3805 		return rc;
3806 
3807 	size = fxstate_size(ctxt);
3808 	rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size);
3809 	if (rc != X86EMUL_CONTINUE)
3810 		return rc;
3811 
3812 	kvm_fpu_get();
3813 
3814 	if (size < __fxstate_size(16)) {
3815 		rc = fxregs_fixup(&fx_state, size);
3816 		if (rc != X86EMUL_CONTINUE)
3817 			goto out;
3818 	}
3819 
3820 	if (fx_state.mxcsr >> 16) {
3821 		rc = emulate_gp(ctxt, 0);
3822 		goto out;
3823 	}
3824 
3825 	if (rc == X86EMUL_CONTINUE)
3826 		rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state));
3827 
3828 out:
3829 	kvm_fpu_put();
3830 
3831 	return rc;
3832 }
3833 
3834 static int em_xsetbv(struct x86_emulate_ctxt *ctxt)
3835 {
3836 	u32 eax, ecx, edx;
3837 
3838 	if (!(ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSXSAVE))
3839 		return emulate_ud(ctxt);
3840 
3841 	eax = reg_read(ctxt, VCPU_REGS_RAX);
3842 	edx = reg_read(ctxt, VCPU_REGS_RDX);
3843 	ecx = reg_read(ctxt, VCPU_REGS_RCX);
3844 
3845 	if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax))
3846 		return emulate_gp(ctxt, 0);
3847 
3848 	return X86EMUL_CONTINUE;
3849 }
3850 
3851 static bool valid_cr(int nr)
3852 {
3853 	switch (nr) {
3854 	case 0:
3855 	case 2 ... 4:
3856 	case 8:
3857 		return true;
3858 	default:
3859 		return false;
3860 	}
3861 }
3862 
3863 static int check_cr_access(struct x86_emulate_ctxt *ctxt)
3864 {
3865 	if (!valid_cr(ctxt->modrm_reg))
3866 		return emulate_ud(ctxt);
3867 
3868 	return X86EMUL_CONTINUE;
3869 }
3870 
3871 static int check_dr_read(struct x86_emulate_ctxt *ctxt)
3872 {
3873 	int dr = ctxt->modrm_reg;
3874 	u64 cr4;
3875 
3876 	if (dr > 7)
3877 		return emulate_ud(ctxt);
3878 
3879 	cr4 = ctxt->ops->get_cr(ctxt, 4);
3880 	if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5))
3881 		return emulate_ud(ctxt);
3882 
3883 	if (ctxt->ops->get_dr(ctxt, 7) & DR7_GD) {
3884 		ulong dr6;
3885 
3886 		dr6 = ctxt->ops->get_dr(ctxt, 6);
3887 		dr6 &= ~DR_TRAP_BITS;
3888 		dr6 |= DR6_BD | DR6_ACTIVE_LOW;
3889 		ctxt->ops->set_dr(ctxt, 6, dr6);
3890 		return emulate_db(ctxt);
3891 	}
3892 
3893 	return X86EMUL_CONTINUE;
3894 }
3895 
3896 static int check_dr_write(struct x86_emulate_ctxt *ctxt)
3897 {
3898 	u64 new_val = ctxt->src.val64;
3899 	int dr = ctxt->modrm_reg;
3900 
3901 	if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL))
3902 		return emulate_gp(ctxt, 0);
3903 
3904 	return check_dr_read(ctxt);
3905 }
3906 
3907 static int check_svme(struct x86_emulate_ctxt *ctxt)
3908 {
3909 	u64 efer = 0;
3910 
3911 	ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
3912 
3913 	if (!(efer & EFER_SVME))
3914 		return emulate_ud(ctxt);
3915 
3916 	return X86EMUL_CONTINUE;
3917 }
3918 
3919 static int check_svme_pa(struct x86_emulate_ctxt *ctxt)
3920 {
3921 	u64 rax = reg_read(ctxt, VCPU_REGS_RAX);
3922 
3923 	/* Valid physical address? */
3924 	if (rax & 0xffff000000000000ULL)
3925 		return emulate_gp(ctxt, 0);
3926 
3927 	return check_svme(ctxt);
3928 }
3929 
3930 static int check_rdtsc(struct x86_emulate_ctxt *ctxt)
3931 {
3932 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
3933 
3934 	if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt))
3935 		return emulate_gp(ctxt, 0);
3936 
3937 	return X86EMUL_CONTINUE;
3938 }
3939 
3940 static int check_rdpmc(struct x86_emulate_ctxt *ctxt)
3941 {
3942 	u64 cr4 = ctxt->ops->get_cr(ctxt, 4);
3943 	u64 rcx = reg_read(ctxt, VCPU_REGS_RCX);
3944 
3945 	/*
3946 	 * VMware allows access to these Pseduo-PMCs even when read via RDPMC
3947 	 * in Ring3 when CR4.PCE=0.
3948 	 */
3949 	if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx))
3950 		return X86EMUL_CONTINUE;
3951 
3952 	/*
3953 	 * If CR4.PCE is set, the SDM requires CPL=0 or CR0.PE=0.  The CR0.PE
3954 	 * check however is unnecessary because CPL is always 0 outside
3955 	 * protected mode.
3956 	 */
3957 	if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) ||
3958 	    ctxt->ops->check_rdpmc_early(ctxt, rcx))
3959 		return emulate_gp(ctxt, 0);
3960 
3961 	return X86EMUL_CONTINUE;
3962 }
3963 
3964 static int check_perm_in(struct x86_emulate_ctxt *ctxt)
3965 {
3966 	ctxt->dst.bytes = min(ctxt->dst.bytes, 4u);
3967 	if (!emulator_io_permitted(ctxt, ctxt->src.val, ctxt->dst.bytes))
3968 		return emulate_gp(ctxt, 0);
3969 
3970 	return X86EMUL_CONTINUE;
3971 }
3972 
3973 static int check_perm_out(struct x86_emulate_ctxt *ctxt)
3974 {
3975 	ctxt->src.bytes = min(ctxt->src.bytes, 4u);
3976 	if (!emulator_io_permitted(ctxt, ctxt->dst.val, ctxt->src.bytes))
3977 		return emulate_gp(ctxt, 0);
3978 
3979 	return X86EMUL_CONTINUE;
3980 }
3981 
3982 #define D(_y) { .flags = (_y) }
3983 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i }
3984 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \
3985 		      .intercept = x86_intercept_##_i, .check_perm = (_p) }
3986 #define N    D(NotImpl)
3987 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) }
3988 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) }
3989 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) }
3990 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) }
3991 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) }
3992 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) }
3993 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) }
3994 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) }
3995 #define II(_f, _e, _i) \
3996 	{ .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i }
3997 #define IIP(_f, _e, _i, _p) \
3998 	{ .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \
3999 	  .intercept = x86_intercept_##_i, .check_perm = (_p) }
4000 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) }
4001 
4002 #define D2bv(_f)      D((_f) | ByteOp), D(_f)
4003 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
4004 #define I2bv(_f, _e)  I((_f) | ByteOp, _e), I(_f, _e)
4005 #define F2bv(_f, _e)  F((_f) | ByteOp, _e), F(_f, _e)
4006 #define I2bvIP(_f, _e, _i, _p) \
4007 	IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
4008 
4009 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e),		\
4010 		F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e),	\
4011 		F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e)
4012 
4013 static const struct opcode group7_rm0[] = {
4014 	N,
4015 	I(SrcNone | Priv | EmulateOnUD,	em_hypercall),
4016 	N, N, N, N, N, N,
4017 };
4018 
4019 static const struct opcode group7_rm1[] = {
4020 	DI(SrcNone | Priv, monitor),
4021 	DI(SrcNone | Priv, mwait),
4022 	N, N, N, N, N, N,
4023 };
4024 
4025 static const struct opcode group7_rm2[] = {
4026 	N,
4027 	II(ImplicitOps | Priv,			em_xsetbv,	xsetbv),
4028 	N, N, N, N, N, N,
4029 };
4030 
4031 static const struct opcode group7_rm3[] = {
4032 	DIP(SrcNone | Prot | Priv,		vmrun,		check_svme_pa),
4033 	II(SrcNone  | Prot | EmulateOnUD,	em_hypercall,	vmmcall),
4034 	DIP(SrcNone | Prot | Priv,		vmload,		check_svme_pa),
4035 	DIP(SrcNone | Prot | Priv,		vmsave,		check_svme_pa),
4036 	DIP(SrcNone | Prot | Priv,		stgi,		check_svme),
4037 	DIP(SrcNone | Prot | Priv,		clgi,		check_svme),
4038 	DIP(SrcNone | Prot | Priv,		skinit,		check_svme),
4039 	DIP(SrcNone | Prot | Priv,		invlpga,	check_svme),
4040 };
4041 
4042 static const struct opcode group7_rm7[] = {
4043 	N,
4044 	DIP(SrcNone, rdtscp, check_rdtsc),
4045 	N, N, N, N, N, N,
4046 };
4047 
4048 static const struct opcode group1[] = {
4049 	F(Lock, em_add),
4050 	F(Lock | PageTable, em_or),
4051 	F(Lock, em_adc),
4052 	F(Lock, em_sbb),
4053 	F(Lock | PageTable, em_and),
4054 	F(Lock, em_sub),
4055 	F(Lock, em_xor),
4056 	F(NoWrite, em_cmp),
4057 };
4058 
4059 static const struct opcode group1A[] = {
4060 	I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N,
4061 };
4062 
4063 static const struct opcode group2[] = {
4064 	F(DstMem | ModRM, em_rol),
4065 	F(DstMem | ModRM, em_ror),
4066 	F(DstMem | ModRM, em_rcl),
4067 	F(DstMem | ModRM, em_rcr),
4068 	F(DstMem | ModRM, em_shl),
4069 	F(DstMem | ModRM, em_shr),
4070 	F(DstMem | ModRM, em_shl),
4071 	F(DstMem | ModRM, em_sar),
4072 };
4073 
4074 static const struct opcode group3[] = {
4075 	F(DstMem | SrcImm | NoWrite, em_test),
4076 	F(DstMem | SrcImm | NoWrite, em_test),
4077 	F(DstMem | SrcNone | Lock, em_not),
4078 	F(DstMem | SrcNone | Lock, em_neg),
4079 	F(DstXacc | Src2Mem, em_mul_ex),
4080 	F(DstXacc | Src2Mem, em_imul_ex),
4081 	F(DstXacc | Src2Mem, em_div_ex),
4082 	F(DstXacc | Src2Mem, em_idiv_ex),
4083 };
4084 
4085 static const struct opcode group4[] = {
4086 	F(ByteOp | DstMem | SrcNone | Lock, em_inc),
4087 	F(ByteOp | DstMem | SrcNone | Lock, em_dec),
4088 	N, N, N, N, N, N,
4089 };
4090 
4091 static const struct opcode group5[] = {
4092 	F(DstMem | SrcNone | Lock,		em_inc),
4093 	F(DstMem | SrcNone | Lock,		em_dec),
4094 	I(SrcMem | NearBranch | IsBranch,       em_call_near_abs),
4095 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_call_far),
4096 	I(SrcMem | NearBranch | IsBranch,       em_jmp_abs),
4097 	I(SrcMemFAddr | ImplicitOps | IsBranch, em_jmp_far),
4098 	I(SrcMem | Stack | TwoMemOp,		em_push), D(Undefined),
4099 };
4100 
4101 static const struct opcode group6[] = {
4102 	II(Prot | DstMem,	   em_sldt, sldt),
4103 	II(Prot | DstMem,	   em_str, str),
4104 	II(Prot | Priv | SrcMem16, em_lldt, lldt),
4105 	II(Prot | Priv | SrcMem16, em_ltr, ltr),
4106 	N, N, N, N,
4107 };
4108 
4109 static const struct group_dual group7 = { {
4110 	II(Mov | DstMem,			em_sgdt, sgdt),
4111 	II(Mov | DstMem,			em_sidt, sidt),
4112 	II(SrcMem | Priv,			em_lgdt, lgdt),
4113 	II(SrcMem | Priv,			em_lidt, lidt),
4114 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4115 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4116 	II(SrcMem | ByteOp | Priv | NoAccess,	em_invlpg, invlpg),
4117 }, {
4118 	EXT(0, group7_rm0),
4119 	EXT(0, group7_rm1),
4120 	EXT(0, group7_rm2),
4121 	EXT(0, group7_rm3),
4122 	II(SrcNone | DstMem | Mov,		em_smsw, smsw), N,
4123 	II(SrcMem16 | Mov | Priv,		em_lmsw, lmsw),
4124 	EXT(0, group7_rm7),
4125 } };
4126 
4127 static const struct opcode group8[] = {
4128 	N, N, N, N,
4129 	F(DstMem | SrcImmByte | NoWrite,		em_bt),
4130 	F(DstMem | SrcImmByte | Lock | PageTable,	em_bts),
4131 	F(DstMem | SrcImmByte | Lock,			em_btr),
4132 	F(DstMem | SrcImmByte | Lock | PageTable,	em_btc),
4133 };
4134 
4135 /*
4136  * The "memory" destination is actually always a register, since we come
4137  * from the register case of group9.
4138  */
4139 static const struct gprefix pfx_0f_c7_7 = {
4140 	N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdpid),
4141 };
4142 
4143 
4144 static const struct group_dual group9 = { {
4145 	N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
4146 }, {
4147 	N, N, N, N, N, N, N,
4148 	GP(0, &pfx_0f_c7_7),
4149 } };
4150 
4151 static const struct opcode group11[] = {
4152 	I(DstMem | SrcImm | Mov | PageTable, em_mov),
4153 	X7(D(Undefined)),
4154 };
4155 
4156 static const struct gprefix pfx_0f_ae_7 = {
4157 	I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N,
4158 };
4159 
4160 static const struct group_dual group15 = { {
4161 	I(ModRM | Aligned16, em_fxsave),
4162 	I(ModRM | Aligned16, em_fxrstor),
4163 	N, N, N, N, N, GP(0, &pfx_0f_ae_7),
4164 }, {
4165 	N, N, N, N, N, N, N, N,
4166 } };
4167 
4168 static const struct gprefix pfx_0f_6f_0f_7f = {
4169 	I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov),
4170 };
4171 
4172 static const struct instr_dual instr_dual_0f_2b = {
4173 	I(0, em_mov), N
4174 };
4175 
4176 static const struct gprefix pfx_0f_2b = {
4177 	ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N,
4178 };
4179 
4180 static const struct gprefix pfx_0f_10_0f_11 = {
4181 	I(Unaligned, em_mov), I(Unaligned, em_mov), N, N,
4182 };
4183 
4184 static const struct gprefix pfx_0f_28_0f_29 = {
4185 	I(Aligned, em_mov), I(Aligned, em_mov), N, N,
4186 };
4187 
4188 static const struct gprefix pfx_0f_e7 = {
4189 	N, I(Sse, em_mov), N, N,
4190 };
4191 
4192 static const struct escape escape_d9 = { {
4193 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw),
4194 }, {
4195 	/* 0xC0 - 0xC7 */
4196 	N, N, N, N, N, N, N, N,
4197 	/* 0xC8 - 0xCF */
4198 	N, N, N, N, N, N, N, N,
4199 	/* 0xD0 - 0xC7 */
4200 	N, N, N, N, N, N, N, N,
4201 	/* 0xD8 - 0xDF */
4202 	N, N, N, N, N, N, N, N,
4203 	/* 0xE0 - 0xE7 */
4204 	N, N, N, N, N, N, N, N,
4205 	/* 0xE8 - 0xEF */
4206 	N, N, N, N, N, N, N, N,
4207 	/* 0xF0 - 0xF7 */
4208 	N, N, N, N, N, N, N, N,
4209 	/* 0xF8 - 0xFF */
4210 	N, N, N, N, N, N, N, N,
4211 } };
4212 
4213 static const struct escape escape_db = { {
4214 	N, N, N, N, N, N, N, N,
4215 }, {
4216 	/* 0xC0 - 0xC7 */
4217 	N, N, N, N, N, N, N, N,
4218 	/* 0xC8 - 0xCF */
4219 	N, N, N, N, N, N, N, N,
4220 	/* 0xD0 - 0xC7 */
4221 	N, N, N, N, N, N, N, N,
4222 	/* 0xD8 - 0xDF */
4223 	N, N, N, N, N, N, N, N,
4224 	/* 0xE0 - 0xE7 */
4225 	N, N, N, I(ImplicitOps, em_fninit), N, N, N, N,
4226 	/* 0xE8 - 0xEF */
4227 	N, N, N, N, N, N, N, N,
4228 	/* 0xF0 - 0xF7 */
4229 	N, N, N, N, N, N, N, N,
4230 	/* 0xF8 - 0xFF */
4231 	N, N, N, N, N, N, N, N,
4232 } };
4233 
4234 static const struct escape escape_dd = { {
4235 	N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw),
4236 }, {
4237 	/* 0xC0 - 0xC7 */
4238 	N, N, N, N, N, N, N, N,
4239 	/* 0xC8 - 0xCF */
4240 	N, N, N, N, N, N, N, N,
4241 	/* 0xD0 - 0xC7 */
4242 	N, N, N, N, N, N, N, N,
4243 	/* 0xD8 - 0xDF */
4244 	N, N, N, N, N, N, N, N,
4245 	/* 0xE0 - 0xE7 */
4246 	N, N, N, N, N, N, N, N,
4247 	/* 0xE8 - 0xEF */
4248 	N, N, N, N, N, N, N, N,
4249 	/* 0xF0 - 0xF7 */
4250 	N, N, N, N, N, N, N, N,
4251 	/* 0xF8 - 0xFF */
4252 	N, N, N, N, N, N, N, N,
4253 } };
4254 
4255 static const struct instr_dual instr_dual_0f_c3 = {
4256 	I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N
4257 };
4258 
4259 static const struct mode_dual mode_dual_63 = {
4260 	N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd)
4261 };
4262 
4263 static const struct instr_dual instr_dual_8d = {
4264 	D(DstReg | SrcMem | ModRM | NoAccess), N
4265 };
4266 
4267 static const struct opcode opcode_table[256] = {
4268 	/* 0x00 - 0x07 */
4269 	F6ALU(Lock, em_add),
4270 	I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
4271 	I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
4272 	/* 0x08 - 0x0F */
4273 	F6ALU(Lock | PageTable, em_or),
4274 	I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
4275 	N,
4276 	/* 0x10 - 0x17 */
4277 	F6ALU(Lock, em_adc),
4278 	I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg),
4279 	I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg),
4280 	/* 0x18 - 0x1F */
4281 	F6ALU(Lock, em_sbb),
4282 	I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
4283 	I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
4284 	/* 0x20 - 0x27 */
4285 	F6ALU(Lock | PageTable, em_and), N, N,
4286 	/* 0x28 - 0x2F */
4287 	F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
4288 	/* 0x30 - 0x37 */
4289 	F6ALU(Lock, em_xor), N, N,
4290 	/* 0x38 - 0x3F */
4291 	F6ALU(NoWrite, em_cmp), N, N,
4292 	/* 0x40 - 0x4F */
4293 	X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)),
4294 	/* 0x50 - 0x57 */
4295 	X8(I(SrcReg | Stack, em_push)),
4296 	/* 0x58 - 0x5F */
4297 	X8(I(DstReg | Stack, em_pop)),
4298 	/* 0x60 - 0x67 */
4299 	I(ImplicitOps | Stack | No64, em_pusha),
4300 	I(ImplicitOps | Stack | No64, em_popa),
4301 	N, MD(ModRM, &mode_dual_63),
4302 	N, N, N, N,
4303 	/* 0x68 - 0x6F */
4304 	I(SrcImm | Mov | Stack, em_push),
4305 	I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
4306 	I(SrcImmByte | Mov | Stack, em_push),
4307 	I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
4308 	I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */
4309 	I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
4310 	/* 0x70 - 0x7F */
4311 	X16(D(SrcImmByte | NearBranch | IsBranch)),
4312 	/* 0x80 - 0x87 */
4313 	G(ByteOp | DstMem | SrcImm, group1),
4314 	G(DstMem | SrcImm, group1),
4315 	G(ByteOp | DstMem | SrcImm | No64, group1),
4316 	G(DstMem | SrcImmByte, group1),
4317 	F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test),
4318 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
4319 	/* 0x88 - 0x8F */
4320 	I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
4321 	I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
4322 	I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
4323 	ID(0, &instr_dual_8d),
4324 	I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
4325 	G(0, group1A),
4326 	/* 0x90 - 0x97 */
4327 	DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)),
4328 	/* 0x98 - 0x9F */
4329 	D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd),
4330 	I(SrcImmFAddr | No64 | IsBranch, em_call_far), N,
4331 	II(ImplicitOps | Stack, em_pushf, pushf),
4332 	II(ImplicitOps | Stack, em_popf, popf),
4333 	I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf),
4334 	/* 0xA0 - 0xA7 */
4335 	I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
4336 	I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
4337 	I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov),
4338 	F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r),
4339 	/* 0xA8 - 0xAF */
4340 	F2bv(DstAcc | SrcImm | NoWrite, em_test),
4341 	I2bv(SrcAcc | DstDI | Mov | String, em_mov),
4342 	I2bv(SrcSI | DstAcc | Mov | String, em_mov),
4343 	F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r),
4344 	/* 0xB0 - 0xB7 */
4345 	X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)),
4346 	/* 0xB8 - 0xBF */
4347 	X8(I(DstReg | SrcImm64 | Mov, em_mov)),
4348 	/* 0xC0 - 0xC7 */
4349 	G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2),
4350 	I(ImplicitOps | NearBranch | SrcImmU16 | IsBranch, em_ret_near_imm),
4351 	I(ImplicitOps | NearBranch | IsBranch, em_ret),
4352 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg),
4353 	I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg),
4354 	G(ByteOp, group11), G(0, group11),
4355 	/* 0xC8 - 0xCF */
4356 	I(Stack | SrcImmU16 | Src2ImmByte | IsBranch, em_enter),
4357 	I(Stack | IsBranch, em_leave),
4358 	I(ImplicitOps | SrcImmU16 | IsBranch, em_ret_far_imm),
4359 	I(ImplicitOps | IsBranch, em_ret_far),
4360 	D(ImplicitOps | IsBranch), DI(SrcImmByte | IsBranch, intn),
4361 	D(ImplicitOps | No64 | IsBranch),
4362 	II(ImplicitOps | IsBranch, em_iret, iret),
4363 	/* 0xD0 - 0xD7 */
4364 	G(Src2One | ByteOp, group2), G(Src2One, group2),
4365 	G(Src2CL | ByteOp, group2), G(Src2CL, group2),
4366 	I(DstAcc | SrcImmUByte | No64, em_aam),
4367 	I(DstAcc | SrcImmUByte | No64, em_aad),
4368 	F(DstAcc | ByteOp | No64, em_salc),
4369 	I(DstAcc | SrcXLat | ByteOp, em_mov),
4370 	/* 0xD8 - 0xDF */
4371 	N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N,
4372 	/* 0xE0 - 0xE7 */
4373 	X3(I(SrcImmByte | NearBranch | IsBranch, em_loop)),
4374 	I(SrcImmByte | NearBranch | IsBranch, em_jcxz),
4375 	I2bvIP(SrcImmUByte | DstAcc, em_in,  in,  check_perm_in),
4376 	I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
4377 	/* 0xE8 - 0xEF */
4378 	I(SrcImm | NearBranch | IsBranch, em_call),
4379 	D(SrcImm | ImplicitOps | NearBranch | IsBranch),
4380 	I(SrcImmFAddr | No64 | IsBranch, em_jmp_far),
4381 	D(SrcImmByte | ImplicitOps | NearBranch | IsBranch),
4382 	I2bvIP(SrcDX | DstAcc, em_in,  in,  check_perm_in),
4383 	I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
4384 	/* 0xF0 - 0xF7 */
4385 	N, DI(ImplicitOps, icebp), N, N,
4386 	DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
4387 	G(ByteOp, group3), G(0, group3),
4388 	/* 0xF8 - 0xFF */
4389 	D(ImplicitOps), D(ImplicitOps),
4390 	I(ImplicitOps, em_cli), I(ImplicitOps, em_sti),
4391 	D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5),
4392 };
4393 
4394 static const struct opcode twobyte_table[256] = {
4395 	/* 0x00 - 0x0F */
4396 	G(0, group6), GD(0, &group7), N, N,
4397 	N, I(ImplicitOps | EmulateOnUD | IsBranch, em_syscall),
4398 	II(ImplicitOps | Priv, em_clts, clts), N,
4399 	DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N,
4400 	N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4401 	/* 0x10 - 0x1F */
4402 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11),
4403 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11),
4404 	N, N, N, N, N, N,
4405 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */
4406 	D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N,
4407 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4408 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4409 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */
4410 	D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */
4411 	/* 0x20 - 0x2F */
4412 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_access),
4413 	DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read),
4414 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write,
4415 						check_cr_access),
4416 	IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write,
4417 						check_dr_write),
4418 	N, N, N, N,
4419 	GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29),
4420 	GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29),
4421 	N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b),
4422 	N, N, N, N,
4423 	/* 0x30 - 0x3F */
4424 	II(ImplicitOps | Priv, em_wrmsr, wrmsr),
4425 	IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
4426 	II(ImplicitOps | Priv, em_rdmsr, rdmsr),
4427 	IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
4428 	I(ImplicitOps | EmulateOnUD | IsBranch, em_sysenter),
4429 	I(ImplicitOps | Priv | EmulateOnUD | IsBranch, em_sysexit),
4430 	N, N,
4431 	N, N, N, N, N, N, N, N,
4432 	/* 0x40 - 0x4F */
4433 	X16(D(DstReg | SrcMem | ModRM)),
4434 	/* 0x50 - 0x5F */
4435 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4436 	/* 0x60 - 0x6F */
4437 	N, N, N, N,
4438 	N, N, N, N,
4439 	N, N, N, N,
4440 	N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f),
4441 	/* 0x70 - 0x7F */
4442 	N, N, N, N,
4443 	N, N, N, N,
4444 	N, N, N, N,
4445 	N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f),
4446 	/* 0x80 - 0x8F */
4447 	X16(D(SrcImm | NearBranch | IsBranch)),
4448 	/* 0x90 - 0x9F */
4449 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
4450 	/* 0xA0 - 0xA7 */
4451 	I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
4452 	II(ImplicitOps, em_cpuid, cpuid),
4453 	F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
4454 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
4455 	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
4456 	/* 0xA8 - 0xAF */
4457 	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
4458 	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
4459 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
4460 	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
4461 	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
4462 	GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
4463 	/* 0xB0 - 0xB7 */
4464 	I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
4465 	I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
4466 	F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
4467 	I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
4468 	I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
4469 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4470 	/* 0xB8 - 0xBF */
4471 	N, N,
4472 	G(BitOp, group8),
4473 	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
4474 	I(DstReg | SrcMem | ModRM, em_bsf_c),
4475 	I(DstReg | SrcMem | ModRM, em_bsr_c),
4476 	D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
4477 	/* 0xC0 - 0xC7 */
4478 	F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd),
4479 	N, ID(0, &instr_dual_0f_c3),
4480 	N, N, N, GD(0, &group9),
4481 	/* 0xC8 - 0xCF */
4482 	X8(I(DstReg, em_bswap)),
4483 	/* 0xD0 - 0xDF */
4484 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N,
4485 	/* 0xE0 - 0xEF */
4486 	N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7),
4487 	N, N, N, N, N, N, N, N,
4488 	/* 0xF0 - 0xFF */
4489 	N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N
4490 };
4491 
4492 static const struct instr_dual instr_dual_0f_38_f0 = {
4493 	I(DstReg | SrcMem | Mov, em_movbe), N
4494 };
4495 
4496 static const struct instr_dual instr_dual_0f_38_f1 = {
4497 	I(DstMem | SrcReg | Mov, em_movbe), N
4498 };
4499 
4500 static const struct gprefix three_byte_0f_38_f0 = {
4501 	ID(0, &instr_dual_0f_38_f0), ID(0, &instr_dual_0f_38_f0), N, N
4502 };
4503 
4504 static const struct gprefix three_byte_0f_38_f1 = {
4505 	ID(0, &instr_dual_0f_38_f1), ID(0, &instr_dual_0f_38_f1), N, N
4506 };
4507 
4508 /*
4509  * Insns below are selected by the prefix which indexed by the third opcode
4510  * byte.
4511  */
4512 static const struct opcode opcode_map_0f_38[256] = {
4513 	/* 0x00 - 0x7f */
4514 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4515 	/* 0x80 - 0xef */
4516 	X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N),
4517 	/* 0xf0 - 0xf1 */
4518 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0),
4519 	GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1),
4520 	/* 0xf2 - 0xff */
4521 	N, N, X4(N), X8(N)
4522 };
4523 
4524 #undef D
4525 #undef N
4526 #undef G
4527 #undef GD
4528 #undef I
4529 #undef GP
4530 #undef EXT
4531 #undef MD
4532 #undef ID
4533 
4534 #undef D2bv
4535 #undef D2bvIP
4536 #undef I2bv
4537 #undef I2bvIP
4538 #undef I6ALU
4539 
4540 static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
4541 {
4542 	unsigned size;
4543 
4544 	size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4545 	if (size == 8)
4546 		size = 4;
4547 	return size;
4548 }
4549 
4550 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op,
4551 		      unsigned size, bool sign_extension)
4552 {
4553 	int rc = X86EMUL_CONTINUE;
4554 
4555 	op->type = OP_IMM;
4556 	op->bytes = size;
4557 	op->addr.mem.ea = ctxt->_eip;
4558 	/* NB. Immediates are sign-extended as necessary. */
4559 	switch (op->bytes) {
4560 	case 1:
4561 		op->val = insn_fetch(s8, ctxt);
4562 		break;
4563 	case 2:
4564 		op->val = insn_fetch(s16, ctxt);
4565 		break;
4566 	case 4:
4567 		op->val = insn_fetch(s32, ctxt);
4568 		break;
4569 	case 8:
4570 		op->val = insn_fetch(s64, ctxt);
4571 		break;
4572 	}
4573 	if (!sign_extension) {
4574 		switch (op->bytes) {
4575 		case 1:
4576 			op->val &= 0xff;
4577 			break;
4578 		case 2:
4579 			op->val &= 0xffff;
4580 			break;
4581 		case 4:
4582 			op->val &= 0xffffffff;
4583 			break;
4584 		}
4585 	}
4586 done:
4587 	return rc;
4588 }
4589 
4590 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op,
4591 			  unsigned d)
4592 {
4593 	int rc = X86EMUL_CONTINUE;
4594 
4595 	switch (d) {
4596 	case OpReg:
4597 		decode_register_operand(ctxt, op);
4598 		break;
4599 	case OpImmUByte:
4600 		rc = decode_imm(ctxt, op, 1, false);
4601 		break;
4602 	case OpMem:
4603 		ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4604 	mem_common:
4605 		*op = ctxt->memop;
4606 		ctxt->memopp = op;
4607 		if (ctxt->d & BitOp)
4608 			fetch_bit_operand(ctxt);
4609 		op->orig_val = op->val;
4610 		break;
4611 	case OpMem64:
4612 		ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8;
4613 		goto mem_common;
4614 	case OpAcc:
4615 		op->type = OP_REG;
4616 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4617 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4618 		fetch_register_operand(op);
4619 		op->orig_val = op->val;
4620 		break;
4621 	case OpAccLo:
4622 		op->type = OP_REG;
4623 		op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes;
4624 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
4625 		fetch_register_operand(op);
4626 		op->orig_val = op->val;
4627 		break;
4628 	case OpAccHi:
4629 		if (ctxt->d & ByteOp) {
4630 			op->type = OP_NONE;
4631 			break;
4632 		}
4633 		op->type = OP_REG;
4634 		op->bytes = ctxt->op_bytes;
4635 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4636 		fetch_register_operand(op);
4637 		op->orig_val = op->val;
4638 		break;
4639 	case OpDI:
4640 		op->type = OP_MEM;
4641 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4642 		op->addr.mem.ea =
4643 			register_address(ctxt, VCPU_REGS_RDI);
4644 		op->addr.mem.seg = VCPU_SREG_ES;
4645 		op->val = 0;
4646 		op->count = 1;
4647 		break;
4648 	case OpDX:
4649 		op->type = OP_REG;
4650 		op->bytes = 2;
4651 		op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX);
4652 		fetch_register_operand(op);
4653 		break;
4654 	case OpCL:
4655 		op->type = OP_IMM;
4656 		op->bytes = 1;
4657 		op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff;
4658 		break;
4659 	case OpImmByte:
4660 		rc = decode_imm(ctxt, op, 1, true);
4661 		break;
4662 	case OpOne:
4663 		op->type = OP_IMM;
4664 		op->bytes = 1;
4665 		op->val = 1;
4666 		break;
4667 	case OpImm:
4668 		rc = decode_imm(ctxt, op, imm_size(ctxt), true);
4669 		break;
4670 	case OpImm64:
4671 		rc = decode_imm(ctxt, op, ctxt->op_bytes, true);
4672 		break;
4673 	case OpMem8:
4674 		ctxt->memop.bytes = 1;
4675 		if (ctxt->memop.type == OP_REG) {
4676 			ctxt->memop.addr.reg = decode_register(ctxt,
4677 					ctxt->modrm_rm, true);
4678 			fetch_register_operand(&ctxt->memop);
4679 		}
4680 		goto mem_common;
4681 	case OpMem16:
4682 		ctxt->memop.bytes = 2;
4683 		goto mem_common;
4684 	case OpMem32:
4685 		ctxt->memop.bytes = 4;
4686 		goto mem_common;
4687 	case OpImmU16:
4688 		rc = decode_imm(ctxt, op, 2, false);
4689 		break;
4690 	case OpImmU:
4691 		rc = decode_imm(ctxt, op, imm_size(ctxt), false);
4692 		break;
4693 	case OpSI:
4694 		op->type = OP_MEM;
4695 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4696 		op->addr.mem.ea =
4697 			register_address(ctxt, VCPU_REGS_RSI);
4698 		op->addr.mem.seg = ctxt->seg_override;
4699 		op->val = 0;
4700 		op->count = 1;
4701 		break;
4702 	case OpXLat:
4703 		op->type = OP_MEM;
4704 		op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes;
4705 		op->addr.mem.ea =
4706 			address_mask(ctxt,
4707 				reg_read(ctxt, VCPU_REGS_RBX) +
4708 				(reg_read(ctxt, VCPU_REGS_RAX) & 0xff));
4709 		op->addr.mem.seg = ctxt->seg_override;
4710 		op->val = 0;
4711 		break;
4712 	case OpImmFAddr:
4713 		op->type = OP_IMM;
4714 		op->addr.mem.ea = ctxt->_eip;
4715 		op->bytes = ctxt->op_bytes + 2;
4716 		insn_fetch_arr(op->valptr, op->bytes, ctxt);
4717 		break;
4718 	case OpMemFAddr:
4719 		ctxt->memop.bytes = ctxt->op_bytes + 2;
4720 		goto mem_common;
4721 	case OpES:
4722 		op->type = OP_IMM;
4723 		op->val = VCPU_SREG_ES;
4724 		break;
4725 	case OpCS:
4726 		op->type = OP_IMM;
4727 		op->val = VCPU_SREG_CS;
4728 		break;
4729 	case OpSS:
4730 		op->type = OP_IMM;
4731 		op->val = VCPU_SREG_SS;
4732 		break;
4733 	case OpDS:
4734 		op->type = OP_IMM;
4735 		op->val = VCPU_SREG_DS;
4736 		break;
4737 	case OpFS:
4738 		op->type = OP_IMM;
4739 		op->val = VCPU_SREG_FS;
4740 		break;
4741 	case OpGS:
4742 		op->type = OP_IMM;
4743 		op->val = VCPU_SREG_GS;
4744 		break;
4745 	case OpImplicit:
4746 		/* Special instructions do their own operand decoding. */
4747 	default:
4748 		op->type = OP_NONE; /* Disable writeback. */
4749 		break;
4750 	}
4751 
4752 done:
4753 	return rc;
4754 }
4755 
4756 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int emulation_type)
4757 {
4758 	int rc = X86EMUL_CONTINUE;
4759 	int mode = ctxt->mode;
4760 	int def_op_bytes, def_ad_bytes, goffset, simd_prefix;
4761 	bool op_prefix = false;
4762 	bool has_seg_override = false;
4763 	struct opcode opcode;
4764 	u16 dummy;
4765 	struct desc_struct desc;
4766 
4767 	ctxt->memop.type = OP_NONE;
4768 	ctxt->memopp = NULL;
4769 	ctxt->_eip = ctxt->eip;
4770 	ctxt->fetch.ptr = ctxt->fetch.data;
4771 	ctxt->fetch.end = ctxt->fetch.data + insn_len;
4772 	ctxt->opcode_len = 1;
4773 	ctxt->intercept = x86_intercept_none;
4774 	if (insn_len > 0)
4775 		memcpy(ctxt->fetch.data, insn, insn_len);
4776 	else {
4777 		rc = __do_insn_fetch_bytes(ctxt, 1);
4778 		if (rc != X86EMUL_CONTINUE)
4779 			goto done;
4780 	}
4781 
4782 	switch (mode) {
4783 	case X86EMUL_MODE_REAL:
4784 	case X86EMUL_MODE_VM86:
4785 		def_op_bytes = def_ad_bytes = 2;
4786 		ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS);
4787 		if (desc.d)
4788 			def_op_bytes = def_ad_bytes = 4;
4789 		break;
4790 	case X86EMUL_MODE_PROT16:
4791 		def_op_bytes = def_ad_bytes = 2;
4792 		break;
4793 	case X86EMUL_MODE_PROT32:
4794 		def_op_bytes = def_ad_bytes = 4;
4795 		break;
4796 #ifdef CONFIG_X86_64
4797 	case X86EMUL_MODE_PROT64:
4798 		def_op_bytes = 4;
4799 		def_ad_bytes = 8;
4800 		break;
4801 #endif
4802 	default:
4803 		return EMULATION_FAILED;
4804 	}
4805 
4806 	ctxt->op_bytes = def_op_bytes;
4807 	ctxt->ad_bytes = def_ad_bytes;
4808 
4809 	/* Legacy prefixes. */
4810 	for (;;) {
4811 		switch (ctxt->b = insn_fetch(u8, ctxt)) {
4812 		case 0x66:	/* operand-size override */
4813 			op_prefix = true;
4814 			/* switch between 2/4 bytes */
4815 			ctxt->op_bytes = def_op_bytes ^ 6;
4816 			break;
4817 		case 0x67:	/* address-size override */
4818 			if (mode == X86EMUL_MODE_PROT64)
4819 				/* switch between 4/8 bytes */
4820 				ctxt->ad_bytes = def_ad_bytes ^ 12;
4821 			else
4822 				/* switch between 2/4 bytes */
4823 				ctxt->ad_bytes = def_ad_bytes ^ 6;
4824 			break;
4825 		case 0x26:	/* ES override */
4826 			has_seg_override = true;
4827 			ctxt->seg_override = VCPU_SREG_ES;
4828 			break;
4829 		case 0x2e:	/* CS override */
4830 			has_seg_override = true;
4831 			ctxt->seg_override = VCPU_SREG_CS;
4832 			break;
4833 		case 0x36:	/* SS override */
4834 			has_seg_override = true;
4835 			ctxt->seg_override = VCPU_SREG_SS;
4836 			break;
4837 		case 0x3e:	/* DS override */
4838 			has_seg_override = true;
4839 			ctxt->seg_override = VCPU_SREG_DS;
4840 			break;
4841 		case 0x64:	/* FS override */
4842 			has_seg_override = true;
4843 			ctxt->seg_override = VCPU_SREG_FS;
4844 			break;
4845 		case 0x65:	/* GS override */
4846 			has_seg_override = true;
4847 			ctxt->seg_override = VCPU_SREG_GS;
4848 			break;
4849 		case 0x40 ... 0x4f: /* REX */
4850 			if (mode != X86EMUL_MODE_PROT64)
4851 				goto done_prefixes;
4852 			ctxt->rex_prefix = ctxt->b;
4853 			continue;
4854 		case 0xf0:	/* LOCK */
4855 			ctxt->lock_prefix = 1;
4856 			break;
4857 		case 0xf2:	/* REPNE/REPNZ */
4858 		case 0xf3:	/* REP/REPE/REPZ */
4859 			ctxt->rep_prefix = ctxt->b;
4860 			break;
4861 		default:
4862 			goto done_prefixes;
4863 		}
4864 
4865 		/* Any legacy prefix after a REX prefix nullifies its effect. */
4866 
4867 		ctxt->rex_prefix = 0;
4868 	}
4869 
4870 done_prefixes:
4871 
4872 	/* REX prefix. */
4873 	if (ctxt->rex_prefix & 8)
4874 		ctxt->op_bytes = 8;	/* REX.W */
4875 
4876 	/* Opcode byte(s). */
4877 	opcode = opcode_table[ctxt->b];
4878 	/* Two-byte opcode? */
4879 	if (ctxt->b == 0x0f) {
4880 		ctxt->opcode_len = 2;
4881 		ctxt->b = insn_fetch(u8, ctxt);
4882 		opcode = twobyte_table[ctxt->b];
4883 
4884 		/* 0F_38 opcode map */
4885 		if (ctxt->b == 0x38) {
4886 			ctxt->opcode_len = 3;
4887 			ctxt->b = insn_fetch(u8, ctxt);
4888 			opcode = opcode_map_0f_38[ctxt->b];
4889 		}
4890 	}
4891 	ctxt->d = opcode.flags;
4892 
4893 	if (ctxt->d & ModRM)
4894 		ctxt->modrm = insn_fetch(u8, ctxt);
4895 
4896 	/* vex-prefix instructions are not implemented */
4897 	if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) &&
4898 	    (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) {
4899 		ctxt->d = NotImpl;
4900 	}
4901 
4902 	while (ctxt->d & GroupMask) {
4903 		switch (ctxt->d & GroupMask) {
4904 		case Group:
4905 			goffset = (ctxt->modrm >> 3) & 7;
4906 			opcode = opcode.u.group[goffset];
4907 			break;
4908 		case GroupDual:
4909 			goffset = (ctxt->modrm >> 3) & 7;
4910 			if ((ctxt->modrm >> 6) == 3)
4911 				opcode = opcode.u.gdual->mod3[goffset];
4912 			else
4913 				opcode = opcode.u.gdual->mod012[goffset];
4914 			break;
4915 		case RMExt:
4916 			goffset = ctxt->modrm & 7;
4917 			opcode = opcode.u.group[goffset];
4918 			break;
4919 		case Prefix:
4920 			if (ctxt->rep_prefix && op_prefix)
4921 				return EMULATION_FAILED;
4922 			simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix;
4923 			switch (simd_prefix) {
4924 			case 0x00: opcode = opcode.u.gprefix->pfx_no; break;
4925 			case 0x66: opcode = opcode.u.gprefix->pfx_66; break;
4926 			case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break;
4927 			case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break;
4928 			}
4929 			break;
4930 		case Escape:
4931 			if (ctxt->modrm > 0xbf) {
4932 				size_t size = ARRAY_SIZE(opcode.u.esc->high);
4933 				u32 index = array_index_nospec(
4934 					ctxt->modrm - 0xc0, size);
4935 
4936 				opcode = opcode.u.esc->high[index];
4937 			} else {
4938 				opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7];
4939 			}
4940 			break;
4941 		case InstrDual:
4942 			if ((ctxt->modrm >> 6) == 3)
4943 				opcode = opcode.u.idual->mod3;
4944 			else
4945 				opcode = opcode.u.idual->mod012;
4946 			break;
4947 		case ModeDual:
4948 			if (ctxt->mode == X86EMUL_MODE_PROT64)
4949 				opcode = opcode.u.mdual->mode64;
4950 			else
4951 				opcode = opcode.u.mdual->mode32;
4952 			break;
4953 		default:
4954 			return EMULATION_FAILED;
4955 		}
4956 
4957 		ctxt->d &= ~(u64)GroupMask;
4958 		ctxt->d |= opcode.flags;
4959 	}
4960 
4961 	ctxt->is_branch = opcode.flags & IsBranch;
4962 
4963 	/* Unrecognised? */
4964 	if (ctxt->d == 0)
4965 		return EMULATION_FAILED;
4966 
4967 	ctxt->execute = opcode.u.execute;
4968 
4969 	if (unlikely(emulation_type & EMULTYPE_TRAP_UD) &&
4970 	    likely(!(ctxt->d & EmulateOnUD)))
4971 		return EMULATION_FAILED;
4972 
4973 	if (unlikely(ctxt->d &
4974 	    (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch|
4975 	     No16))) {
4976 		/*
4977 		 * These are copied unconditionally here, and checked unconditionally
4978 		 * in x86_emulate_insn.
4979 		 */
4980 		ctxt->check_perm = opcode.check_perm;
4981 		ctxt->intercept = opcode.intercept;
4982 
4983 		if (ctxt->d & NotImpl)
4984 			return EMULATION_FAILED;
4985 
4986 		if (mode == X86EMUL_MODE_PROT64) {
4987 			if (ctxt->op_bytes == 4 && (ctxt->d & Stack))
4988 				ctxt->op_bytes = 8;
4989 			else if (ctxt->d & NearBranch)
4990 				ctxt->op_bytes = 8;
4991 		}
4992 
4993 		if (ctxt->d & Op3264) {
4994 			if (mode == X86EMUL_MODE_PROT64)
4995 				ctxt->op_bytes = 8;
4996 			else
4997 				ctxt->op_bytes = 4;
4998 		}
4999 
5000 		if ((ctxt->d & No16) && ctxt->op_bytes == 2)
5001 			ctxt->op_bytes = 4;
5002 
5003 		if (ctxt->d & Sse)
5004 			ctxt->op_bytes = 16;
5005 		else if (ctxt->d & Mmx)
5006 			ctxt->op_bytes = 8;
5007 	}
5008 
5009 	/* ModRM and SIB bytes. */
5010 	if (ctxt->d & ModRM) {
5011 		rc = decode_modrm(ctxt, &ctxt->memop);
5012 		if (!has_seg_override) {
5013 			has_seg_override = true;
5014 			ctxt->seg_override = ctxt->modrm_seg;
5015 		}
5016 	} else if (ctxt->d & MemAbs)
5017 		rc = decode_abs(ctxt, &ctxt->memop);
5018 	if (rc != X86EMUL_CONTINUE)
5019 		goto done;
5020 
5021 	if (!has_seg_override)
5022 		ctxt->seg_override = VCPU_SREG_DS;
5023 
5024 	ctxt->memop.addr.mem.seg = ctxt->seg_override;
5025 
5026 	/*
5027 	 * Decode and fetch the source operand: register, memory
5028 	 * or immediate.
5029 	 */
5030 	rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask);
5031 	if (rc != X86EMUL_CONTINUE)
5032 		goto done;
5033 
5034 	/*
5035 	 * Decode and fetch the second source operand: register, memory
5036 	 * or immediate.
5037 	 */
5038 	rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask);
5039 	if (rc != X86EMUL_CONTINUE)
5040 		goto done;
5041 
5042 	/* Decode and fetch the destination operand: register or memory. */
5043 	rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask);
5044 
5045 	if (ctxt->rip_relative && likely(ctxt->memopp))
5046 		ctxt->memopp->addr.mem.ea = address_mask(ctxt,
5047 					ctxt->memopp->addr.mem.ea + ctxt->_eip);
5048 
5049 done:
5050 	if (rc == X86EMUL_PROPAGATE_FAULT)
5051 		ctxt->have_exception = true;
5052 	return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
5053 }
5054 
5055 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
5056 {
5057 	return ctxt->d & PageTable;
5058 }
5059 
5060 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
5061 {
5062 	/* The second termination condition only applies for REPE
5063 	 * and REPNE. Test if the repeat string operation prefix is
5064 	 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the
5065 	 * corresponding termination condition according to:
5066 	 * 	- if REPE/REPZ and ZF = 0 then done
5067 	 * 	- if REPNE/REPNZ and ZF = 1 then done
5068 	 */
5069 	if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) ||
5070 	     (ctxt->b == 0xae) || (ctxt->b == 0xaf))
5071 	    && (((ctxt->rep_prefix == REPE_PREFIX) &&
5072 		 ((ctxt->eflags & X86_EFLAGS_ZF) == 0))
5073 		|| ((ctxt->rep_prefix == REPNE_PREFIX) &&
5074 		    ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF))))
5075 		return true;
5076 
5077 	return false;
5078 }
5079 
5080 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
5081 {
5082 	int rc;
5083 
5084 	kvm_fpu_get();
5085 	rc = asm_safe("fwait");
5086 	kvm_fpu_put();
5087 
5088 	if (unlikely(rc != X86EMUL_CONTINUE))
5089 		return emulate_exception(ctxt, MF_VECTOR, 0, false);
5090 
5091 	return X86EMUL_CONTINUE;
5092 }
5093 
5094 static void fetch_possible_mmx_operand(struct operand *op)
5095 {
5096 	if (op->type == OP_MM)
5097 		kvm_read_mmx_reg(op->addr.mm, &op->mm_val);
5098 }
5099 
5100 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop)
5101 {
5102 	ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
5103 
5104 	if (!(ctxt->d & ByteOp))
5105 		fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
5106 
5107 	asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n"
5108 	    : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
5109 	      [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT
5110 	    : "c"(ctxt->src2.val));
5111 
5112 	ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
5113 	if (!fop) /* exception is returned in fop variable */
5114 		return emulate_de(ctxt);
5115 	return X86EMUL_CONTINUE;
5116 }
5117 
5118 void init_decode_cache(struct x86_emulate_ctxt *ctxt)
5119 {
5120 	/* Clear fields that are set conditionally but read without a guard. */
5121 	ctxt->rip_relative = false;
5122 	ctxt->rex_prefix = 0;
5123 	ctxt->lock_prefix = 0;
5124 	ctxt->rep_prefix = 0;
5125 	ctxt->regs_valid = 0;
5126 	ctxt->regs_dirty = 0;
5127 
5128 	ctxt->io_read.pos = 0;
5129 	ctxt->io_read.end = 0;
5130 	ctxt->mem_read.end = 0;
5131 }
5132 
5133 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
5134 {
5135 	const struct x86_emulate_ops *ops = ctxt->ops;
5136 	int rc = X86EMUL_CONTINUE;
5137 	int saved_dst_type = ctxt->dst.type;
5138 	bool is_guest_mode = ctxt->ops->is_guest_mode(ctxt);
5139 
5140 	ctxt->mem_read.pos = 0;
5141 
5142 	/* LOCK prefix is allowed only with some instructions */
5143 	if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) {
5144 		rc = emulate_ud(ctxt);
5145 		goto done;
5146 	}
5147 
5148 	if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) {
5149 		rc = emulate_ud(ctxt);
5150 		goto done;
5151 	}
5152 
5153 	if (unlikely(ctxt->d &
5154 		     (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) {
5155 		if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) ||
5156 				(ctxt->d & Undefined)) {
5157 			rc = emulate_ud(ctxt);
5158 			goto done;
5159 		}
5160 
5161 		if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM)))
5162 		    || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) {
5163 			rc = emulate_ud(ctxt);
5164 			goto done;
5165 		}
5166 
5167 		if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) {
5168 			rc = emulate_nm(ctxt);
5169 			goto done;
5170 		}
5171 
5172 		if (ctxt->d & Mmx) {
5173 			rc = flush_pending_x87_faults(ctxt);
5174 			if (rc != X86EMUL_CONTINUE)
5175 				goto done;
5176 			/*
5177 			 * Now that we know the fpu is exception safe, we can fetch
5178 			 * operands from it.
5179 			 */
5180 			fetch_possible_mmx_operand(&ctxt->src);
5181 			fetch_possible_mmx_operand(&ctxt->src2);
5182 			if (!(ctxt->d & Mov))
5183 				fetch_possible_mmx_operand(&ctxt->dst);
5184 		}
5185 
5186 		if (unlikely(is_guest_mode) && ctxt->intercept) {
5187 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5188 						      X86_ICPT_PRE_EXCEPT);
5189 			if (rc != X86EMUL_CONTINUE)
5190 				goto done;
5191 		}
5192 
5193 		/* Instruction can only be executed in protected mode */
5194 		if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) {
5195 			rc = emulate_ud(ctxt);
5196 			goto done;
5197 		}
5198 
5199 		/* Privileged instruction can be executed only in CPL=0 */
5200 		if ((ctxt->d & Priv) && ops->cpl(ctxt)) {
5201 			if (ctxt->d & PrivUD)
5202 				rc = emulate_ud(ctxt);
5203 			else
5204 				rc = emulate_gp(ctxt, 0);
5205 			goto done;
5206 		}
5207 
5208 		/* Do instruction specific permission checks */
5209 		if (ctxt->d & CheckPerm) {
5210 			rc = ctxt->check_perm(ctxt);
5211 			if (rc != X86EMUL_CONTINUE)
5212 				goto done;
5213 		}
5214 
5215 		if (unlikely(is_guest_mode) && (ctxt->d & Intercept)) {
5216 			rc = emulator_check_intercept(ctxt, ctxt->intercept,
5217 						      X86_ICPT_POST_EXCEPT);
5218 			if (rc != X86EMUL_CONTINUE)
5219 				goto done;
5220 		}
5221 
5222 		if (ctxt->rep_prefix && (ctxt->d & String)) {
5223 			/* All REP prefixes have the same first termination condition */
5224 			if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) {
5225 				string_registers_quirk(ctxt);
5226 				ctxt->eip = ctxt->_eip;
5227 				ctxt->eflags &= ~X86_EFLAGS_RF;
5228 				goto done;
5229 			}
5230 		}
5231 	}
5232 
5233 	if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) {
5234 		rc = segmented_read(ctxt, ctxt->src.addr.mem,
5235 				    ctxt->src.valptr, ctxt->src.bytes);
5236 		if (rc != X86EMUL_CONTINUE)
5237 			goto done;
5238 		ctxt->src.orig_val64 = ctxt->src.val64;
5239 	}
5240 
5241 	if (ctxt->src2.type == OP_MEM) {
5242 		rc = segmented_read(ctxt, ctxt->src2.addr.mem,
5243 				    &ctxt->src2.val, ctxt->src2.bytes);
5244 		if (rc != X86EMUL_CONTINUE)
5245 			goto done;
5246 	}
5247 
5248 	if ((ctxt->d & DstMask) == ImplicitOps)
5249 		goto special_insn;
5250 
5251 
5252 	if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) {
5253 		/* optimisation - avoid slow emulated read if Mov */
5254 		rc = segmented_read(ctxt, ctxt->dst.addr.mem,
5255 				   &ctxt->dst.val, ctxt->dst.bytes);
5256 		if (rc != X86EMUL_CONTINUE) {
5257 			if (!(ctxt->d & NoWrite) &&
5258 			    rc == X86EMUL_PROPAGATE_FAULT &&
5259 			    ctxt->exception.vector == PF_VECTOR)
5260 				ctxt->exception.error_code |= PFERR_WRITE_MASK;
5261 			goto done;
5262 		}
5263 	}
5264 	/* Copy full 64-bit value for CMPXCHG8B.  */
5265 	ctxt->dst.orig_val64 = ctxt->dst.val64;
5266 
5267 special_insn:
5268 
5269 	if (unlikely(is_guest_mode) && (ctxt->d & Intercept)) {
5270 		rc = emulator_check_intercept(ctxt, ctxt->intercept,
5271 					      X86_ICPT_POST_MEMACCESS);
5272 		if (rc != X86EMUL_CONTINUE)
5273 			goto done;
5274 	}
5275 
5276 	if (ctxt->rep_prefix && (ctxt->d & String))
5277 		ctxt->eflags |= X86_EFLAGS_RF;
5278 	else
5279 		ctxt->eflags &= ~X86_EFLAGS_RF;
5280 
5281 	if (ctxt->execute) {
5282 		if (ctxt->d & Fastop)
5283 			rc = fastop(ctxt, ctxt->fop);
5284 		else
5285 			rc = ctxt->execute(ctxt);
5286 		if (rc != X86EMUL_CONTINUE)
5287 			goto done;
5288 		goto writeback;
5289 	}
5290 
5291 	if (ctxt->opcode_len == 2)
5292 		goto twobyte_insn;
5293 	else if (ctxt->opcode_len == 3)
5294 		goto threebyte_insn;
5295 
5296 	switch (ctxt->b) {
5297 	case 0x70 ... 0x7f: /* jcc (short) */
5298 		if (test_cc(ctxt->b, ctxt->eflags))
5299 			rc = jmp_rel(ctxt, ctxt->src.val);
5300 		break;
5301 	case 0x8d: /* lea r16/r32, m */
5302 		ctxt->dst.val = ctxt->src.addr.mem.ea;
5303 		break;
5304 	case 0x90 ... 0x97: /* nop / xchg reg, rax */
5305 		if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX))
5306 			ctxt->dst.type = OP_NONE;
5307 		else
5308 			rc = em_xchg(ctxt);
5309 		break;
5310 	case 0x98: /* cbw/cwde/cdqe */
5311 		switch (ctxt->op_bytes) {
5312 		case 2: ctxt->dst.val = (s8)ctxt->dst.val; break;
5313 		case 4: ctxt->dst.val = (s16)ctxt->dst.val; break;
5314 		case 8: ctxt->dst.val = (s32)ctxt->dst.val; break;
5315 		}
5316 		break;
5317 	case 0xcc:		/* int3 */
5318 		rc = emulate_int(ctxt, 3);
5319 		break;
5320 	case 0xcd:		/* int n */
5321 		rc = emulate_int(ctxt, ctxt->src.val);
5322 		break;
5323 	case 0xce:		/* into */
5324 		if (ctxt->eflags & X86_EFLAGS_OF)
5325 			rc = emulate_int(ctxt, 4);
5326 		break;
5327 	case 0xe9: /* jmp rel */
5328 	case 0xeb: /* jmp rel short */
5329 		rc = jmp_rel(ctxt, ctxt->src.val);
5330 		ctxt->dst.type = OP_NONE; /* Disable writeback. */
5331 		break;
5332 	case 0xf4:              /* hlt */
5333 		ctxt->ops->halt(ctxt);
5334 		break;
5335 	case 0xf5:	/* cmc */
5336 		/* complement carry flag from eflags reg */
5337 		ctxt->eflags ^= X86_EFLAGS_CF;
5338 		break;
5339 	case 0xf8: /* clc */
5340 		ctxt->eflags &= ~X86_EFLAGS_CF;
5341 		break;
5342 	case 0xf9: /* stc */
5343 		ctxt->eflags |= X86_EFLAGS_CF;
5344 		break;
5345 	case 0xfc: /* cld */
5346 		ctxt->eflags &= ~X86_EFLAGS_DF;
5347 		break;
5348 	case 0xfd: /* std */
5349 		ctxt->eflags |= X86_EFLAGS_DF;
5350 		break;
5351 	default:
5352 		goto cannot_emulate;
5353 	}
5354 
5355 	if (rc != X86EMUL_CONTINUE)
5356 		goto done;
5357 
5358 writeback:
5359 	if (ctxt->d & SrcWrite) {
5360 		BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR);
5361 		rc = writeback(ctxt, &ctxt->src);
5362 		if (rc != X86EMUL_CONTINUE)
5363 			goto done;
5364 	}
5365 	if (!(ctxt->d & NoWrite)) {
5366 		rc = writeback(ctxt, &ctxt->dst);
5367 		if (rc != X86EMUL_CONTINUE)
5368 			goto done;
5369 	}
5370 
5371 	/*
5372 	 * restore dst type in case the decoding will be reused
5373 	 * (happens for string instruction )
5374 	 */
5375 	ctxt->dst.type = saved_dst_type;
5376 
5377 	if ((ctxt->d & SrcMask) == SrcSI)
5378 		string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src);
5379 
5380 	if ((ctxt->d & DstMask) == DstDI)
5381 		string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst);
5382 
5383 	if (ctxt->rep_prefix && (ctxt->d & String)) {
5384 		unsigned int count;
5385 		struct read_cache *r = &ctxt->io_read;
5386 		if ((ctxt->d & SrcMask) == SrcSI)
5387 			count = ctxt->src.count;
5388 		else
5389 			count = ctxt->dst.count;
5390 		register_address_increment(ctxt, VCPU_REGS_RCX, -count);
5391 
5392 		if (!string_insn_completed(ctxt)) {
5393 			/*
5394 			 * Re-enter guest when pio read ahead buffer is empty
5395 			 * or, if it is not used, after each 1024 iteration.
5396 			 */
5397 			if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) &&
5398 			    (r->end == 0 || r->end != r->pos)) {
5399 				/*
5400 				 * Reset read cache. Usually happens before
5401 				 * decode, but since instruction is restarted
5402 				 * we have to do it here.
5403 				 */
5404 				ctxt->mem_read.end = 0;
5405 				writeback_registers(ctxt);
5406 				return EMULATION_RESTART;
5407 			}
5408 			goto done; /* skip rip writeback */
5409 		}
5410 		ctxt->eflags &= ~X86_EFLAGS_RF;
5411 	}
5412 
5413 	ctxt->eip = ctxt->_eip;
5414 	if (ctxt->mode != X86EMUL_MODE_PROT64)
5415 		ctxt->eip = (u32)ctxt->_eip;
5416 
5417 done:
5418 	if (rc == X86EMUL_PROPAGATE_FAULT) {
5419 		if (KVM_EMULATOR_BUG_ON(ctxt->exception.vector > 0x1f, ctxt))
5420 			return EMULATION_FAILED;
5421 		ctxt->have_exception = true;
5422 	}
5423 	if (rc == X86EMUL_INTERCEPTED)
5424 		return EMULATION_INTERCEPTED;
5425 
5426 	if (rc == X86EMUL_CONTINUE)
5427 		writeback_registers(ctxt);
5428 
5429 	return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK;
5430 
5431 twobyte_insn:
5432 	switch (ctxt->b) {
5433 	case 0x09:		/* wbinvd */
5434 		(ctxt->ops->wbinvd)(ctxt);
5435 		break;
5436 	case 0x08:		/* invd */
5437 	case 0x0d:		/* GrpP (prefetch) */
5438 	case 0x18:		/* Grp16 (prefetch/nop) */
5439 	case 0x1f:		/* nop */
5440 		break;
5441 	case 0x20: /* mov cr, reg */
5442 		ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg);
5443 		break;
5444 	case 0x21: /* mov from dr to reg */
5445 		ctxt->dst.val = ops->get_dr(ctxt, ctxt->modrm_reg);
5446 		break;
5447 	case 0x40 ... 0x4f:	/* cmov */
5448 		if (test_cc(ctxt->b, ctxt->eflags))
5449 			ctxt->dst.val = ctxt->src.val;
5450 		else if (ctxt->op_bytes != 4)
5451 			ctxt->dst.type = OP_NONE; /* no writeback */
5452 		break;
5453 	case 0x80 ... 0x8f: /* jnz rel, etc*/
5454 		if (test_cc(ctxt->b, ctxt->eflags))
5455 			rc = jmp_rel(ctxt, ctxt->src.val);
5456 		break;
5457 	case 0x90 ... 0x9f:     /* setcc r/m8 */
5458 		ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
5459 		break;
5460 	case 0xb6 ... 0xb7:	/* movzx */
5461 		ctxt->dst.bytes = ctxt->op_bytes;
5462 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val
5463 						       : (u16) ctxt->src.val;
5464 		break;
5465 	case 0xbe ... 0xbf:	/* movsx */
5466 		ctxt->dst.bytes = ctxt->op_bytes;
5467 		ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val :
5468 							(s16) ctxt->src.val;
5469 		break;
5470 	default:
5471 		goto cannot_emulate;
5472 	}
5473 
5474 threebyte_insn:
5475 
5476 	if (rc != X86EMUL_CONTINUE)
5477 		goto done;
5478 
5479 	goto writeback;
5480 
5481 cannot_emulate:
5482 	return EMULATION_FAILED;
5483 }
5484 
5485 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt)
5486 {
5487 	invalidate_registers(ctxt);
5488 }
5489 
5490 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt)
5491 {
5492 	writeback_registers(ctxt);
5493 }
5494 
5495 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt)
5496 {
5497 	if (ctxt->rep_prefix && (ctxt->d & String))
5498 		return false;
5499 
5500 	if (ctxt->d & TwoMemOp)
5501 		return false;
5502 
5503 	return true;
5504 }
5505