1 // SPDX-License-Identifier: GPL-2.0-only 2 /****************************************************************************** 3 * emulate.c 4 * 5 * Generic x86 (32-bit and 64-bit) instruction decoder and emulator. 6 * 7 * Copyright (c) 2005 Keir Fraser 8 * 9 * Linux coding style, mod r/m decoder, segment base fixes, real-mode 10 * privileged instructions: 11 * 12 * Copyright (C) 2006 Qumranet 13 * Copyright 2010 Red Hat, Inc. and/or its affiliates. 14 * 15 * Avi Kivity <avi@qumranet.com> 16 * Yaniv Kamay <yaniv@qumranet.com> 17 * 18 * From: xen-unstable 10676:af9809f51f81a3c43f276f00c81a52ef558afda4 19 */ 20 21 #include <linux/kvm_host.h> 22 #include "kvm_cache_regs.h" 23 #include "kvm_emulate.h" 24 #include <linux/stringify.h> 25 #include <asm/fpu/api.h> 26 #include <asm/debugreg.h> 27 #include <asm/nospec-branch.h> 28 29 #include "x86.h" 30 #include "tss.h" 31 #include "mmu.h" 32 #include "pmu.h" 33 34 /* 35 * Operand types 36 */ 37 #define OpNone 0ull 38 #define OpImplicit 1ull /* No generic decode */ 39 #define OpReg 2ull /* Register */ 40 #define OpMem 3ull /* Memory */ 41 #define OpAcc 4ull /* Accumulator: AL/AX/EAX/RAX */ 42 #define OpDI 5ull /* ES:DI/EDI/RDI */ 43 #define OpMem64 6ull /* Memory, 64-bit */ 44 #define OpImmUByte 7ull /* Zero-extended 8-bit immediate */ 45 #define OpDX 8ull /* DX register */ 46 #define OpCL 9ull /* CL register (for shifts) */ 47 #define OpImmByte 10ull /* 8-bit sign extended immediate */ 48 #define OpOne 11ull /* Implied 1 */ 49 #define OpImm 12ull /* Sign extended up to 32-bit immediate */ 50 #define OpMem16 13ull /* Memory operand (16-bit). */ 51 #define OpMem32 14ull /* Memory operand (32-bit). */ 52 #define OpImmU 15ull /* Immediate operand, zero extended */ 53 #define OpSI 16ull /* SI/ESI/RSI */ 54 #define OpImmFAddr 17ull /* Immediate far address */ 55 #define OpMemFAddr 18ull /* Far address in memory */ 56 #define OpImmU16 19ull /* Immediate operand, 16 bits, zero extended */ 57 #define OpES 20ull /* ES */ 58 #define OpCS 21ull /* CS */ 59 #define OpSS 22ull /* SS */ 60 #define OpDS 23ull /* DS */ 61 #define OpFS 24ull /* FS */ 62 #define OpGS 25ull /* GS */ 63 #define OpMem8 26ull /* 8-bit zero extended memory operand */ 64 #define OpImm64 27ull /* Sign extended 16/32/64-bit immediate */ 65 #define OpXLat 28ull /* memory at BX/EBX/RBX + zero-extended AL */ 66 #define OpAccLo 29ull /* Low part of extended acc (AX/AX/EAX/RAX) */ 67 #define OpAccHi 30ull /* High part of extended acc (-/DX/EDX/RDX) */ 68 69 #define OpBits 5 /* Width of operand field */ 70 #define OpMask ((1ull << OpBits) - 1) 71 72 /* 73 * Opcode effective-address decode tables. 74 * Note that we only emulate instructions that have at least one memory 75 * operand (excluding implicit stack references). We assume that stack 76 * references and instruction fetches will never occur in special memory 77 * areas that require emulation. So, for example, 'mov <imm>,<reg>' need 78 * not be handled. 79 */ 80 81 /* Operand sizes: 8-bit operands or specified/overridden size. */ 82 #define ByteOp (1<<0) /* 8-bit operands. */ 83 /* Destination operand type. */ 84 #define DstShift 1 85 #define ImplicitOps (OpImplicit << DstShift) 86 #define DstReg (OpReg << DstShift) 87 #define DstMem (OpMem << DstShift) 88 #define DstAcc (OpAcc << DstShift) 89 #define DstDI (OpDI << DstShift) 90 #define DstMem64 (OpMem64 << DstShift) 91 #define DstMem16 (OpMem16 << DstShift) 92 #define DstImmUByte (OpImmUByte << DstShift) 93 #define DstDX (OpDX << DstShift) 94 #define DstAccLo (OpAccLo << DstShift) 95 #define DstMask (OpMask << DstShift) 96 /* Source operand type. */ 97 #define SrcShift 6 98 #define SrcNone (OpNone << SrcShift) 99 #define SrcReg (OpReg << SrcShift) 100 #define SrcMem (OpMem << SrcShift) 101 #define SrcMem16 (OpMem16 << SrcShift) 102 #define SrcMem32 (OpMem32 << SrcShift) 103 #define SrcImm (OpImm << SrcShift) 104 #define SrcImmByte (OpImmByte << SrcShift) 105 #define SrcOne (OpOne << SrcShift) 106 #define SrcImmUByte (OpImmUByte << SrcShift) 107 #define SrcImmU (OpImmU << SrcShift) 108 #define SrcSI (OpSI << SrcShift) 109 #define SrcXLat (OpXLat << SrcShift) 110 #define SrcImmFAddr (OpImmFAddr << SrcShift) 111 #define SrcMemFAddr (OpMemFAddr << SrcShift) 112 #define SrcAcc (OpAcc << SrcShift) 113 #define SrcImmU16 (OpImmU16 << SrcShift) 114 #define SrcImm64 (OpImm64 << SrcShift) 115 #define SrcDX (OpDX << SrcShift) 116 #define SrcMem8 (OpMem8 << SrcShift) 117 #define SrcAccHi (OpAccHi << SrcShift) 118 #define SrcMask (OpMask << SrcShift) 119 #define BitOp (1<<11) 120 #define MemAbs (1<<12) /* Memory operand is absolute displacement */ 121 #define String (1<<13) /* String instruction (rep capable) */ 122 #define Stack (1<<14) /* Stack instruction (push/pop) */ 123 #define GroupMask (7<<15) /* Opcode uses one of the group mechanisms */ 124 #define Group (1<<15) /* Bits 3:5 of modrm byte extend opcode */ 125 #define GroupDual (2<<15) /* Alternate decoding of mod == 3 */ 126 #define Prefix (3<<15) /* Instruction varies with 66/f2/f3 prefix */ 127 #define RMExt (4<<15) /* Opcode extension in ModRM r/m if mod == 3 */ 128 #define Escape (5<<15) /* Escape to coprocessor instruction */ 129 #define InstrDual (6<<15) /* Alternate instruction decoding of mod == 3 */ 130 #define ModeDual (7<<15) /* Different instruction for 32/64 bit */ 131 #define Sse (1<<18) /* SSE Vector instruction */ 132 /* Generic ModRM decode. */ 133 #define ModRM (1<<19) 134 /* Destination is only written; never read. */ 135 #define Mov (1<<20) 136 /* Misc flags */ 137 #define Prot (1<<21) /* instruction generates #UD if not in prot-mode */ 138 #define EmulateOnUD (1<<22) /* Emulate if unsupported by the host */ 139 #define NoAccess (1<<23) /* Don't access memory (lea/invlpg/verr etc) */ 140 #define Op3264 (1<<24) /* Operand is 64b in long mode, 32b otherwise */ 141 #define Undefined (1<<25) /* No Such Instruction */ 142 #define Lock (1<<26) /* lock prefix is allowed for the instruction */ 143 #define Priv (1<<27) /* instruction generates #GP if current CPL != 0 */ 144 #define No64 (1<<28) 145 #define PageTable (1 << 29) /* instruction used to write page table */ 146 #define NotImpl (1 << 30) /* instruction is not implemented */ 147 /* Source 2 operand type */ 148 #define Src2Shift (31) 149 #define Src2None (OpNone << Src2Shift) 150 #define Src2Mem (OpMem << Src2Shift) 151 #define Src2CL (OpCL << Src2Shift) 152 #define Src2ImmByte (OpImmByte << Src2Shift) 153 #define Src2One (OpOne << Src2Shift) 154 #define Src2Imm (OpImm << Src2Shift) 155 #define Src2ES (OpES << Src2Shift) 156 #define Src2CS (OpCS << Src2Shift) 157 #define Src2SS (OpSS << Src2Shift) 158 #define Src2DS (OpDS << Src2Shift) 159 #define Src2FS (OpFS << Src2Shift) 160 #define Src2GS (OpGS << Src2Shift) 161 #define Src2Mask (OpMask << Src2Shift) 162 #define Mmx ((u64)1 << 40) /* MMX Vector instruction */ 163 #define AlignMask ((u64)7 << 41) 164 #define Aligned ((u64)1 << 41) /* Explicitly aligned (e.g. MOVDQA) */ 165 #define Unaligned ((u64)2 << 41) /* Explicitly unaligned (e.g. MOVDQU) */ 166 #define Avx ((u64)3 << 41) /* Advanced Vector Extensions */ 167 #define Aligned16 ((u64)4 << 41) /* Aligned to 16 byte boundary (e.g. FXSAVE) */ 168 #define Fastop ((u64)1 << 44) /* Use opcode::u.fastop */ 169 #define NoWrite ((u64)1 << 45) /* No writeback */ 170 #define SrcWrite ((u64)1 << 46) /* Write back src operand */ 171 #define NoMod ((u64)1 << 47) /* Mod field is ignored */ 172 #define Intercept ((u64)1 << 48) /* Has valid intercept field */ 173 #define CheckPerm ((u64)1 << 49) /* Has valid check_perm field */ 174 #define PrivUD ((u64)1 << 51) /* #UD instead of #GP on CPL > 0 */ 175 #define NearBranch ((u64)1 << 52) /* Near branches */ 176 #define No16 ((u64)1 << 53) /* No 16 bit operand */ 177 #define IncSP ((u64)1 << 54) /* SP is incremented before ModRM calc */ 178 #define TwoMemOp ((u64)1 << 55) /* Instruction has two memory operand */ 179 180 #define DstXacc (DstAccLo | SrcAccHi | SrcWrite) 181 182 #define X2(x...) x, x 183 #define X3(x...) X2(x), x 184 #define X4(x...) X2(x), X2(x) 185 #define X5(x...) X4(x), x 186 #define X6(x...) X4(x), X2(x) 187 #define X7(x...) X4(x), X3(x) 188 #define X8(x...) X4(x), X4(x) 189 #define X16(x...) X8(x), X8(x) 190 191 #define NR_FASTOP (ilog2(sizeof(ulong)) + 1) 192 #define FASTOP_SIZE 8 193 194 struct opcode { 195 u64 flags : 56; 196 u64 intercept : 8; 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 ulong reg_read(struct x86_emulate_ctxt *ctxt, unsigned nr) 247 { 248 if (!(ctxt->regs_valid & (1 << nr))) { 249 ctxt->regs_valid |= 1 << nr; 250 ctxt->_regs[nr] = ctxt->ops->read_gpr(ctxt, nr); 251 } 252 return ctxt->_regs[nr]; 253 } 254 255 static ulong *reg_write(struct x86_emulate_ctxt *ctxt, unsigned nr) 256 { 257 ctxt->regs_valid |= 1 << nr; 258 ctxt->regs_dirty |= 1 << nr; 259 return &ctxt->_regs[nr]; 260 } 261 262 static ulong *reg_rmw(struct x86_emulate_ctxt *ctxt, unsigned nr) 263 { 264 reg_read(ctxt, nr); 265 return reg_write(ctxt, nr); 266 } 267 268 static void writeback_registers(struct x86_emulate_ctxt *ctxt) 269 { 270 unsigned reg; 271 272 for_each_set_bit(reg, (ulong *)&ctxt->regs_dirty, 16) 273 ctxt->ops->write_gpr(ctxt, reg, ctxt->_regs[reg]); 274 } 275 276 static void invalidate_registers(struct x86_emulate_ctxt *ctxt) 277 { 278 ctxt->regs_dirty = 0; 279 ctxt->regs_valid = 0; 280 } 281 282 /* 283 * These EFLAGS bits are restored from saved value during emulation, and 284 * any changes are written back to the saved value after emulation. 285 */ 286 #define EFLAGS_MASK (X86_EFLAGS_OF|X86_EFLAGS_SF|X86_EFLAGS_ZF|X86_EFLAGS_AF|\ 287 X86_EFLAGS_PF|X86_EFLAGS_CF) 288 289 #ifdef CONFIG_X86_64 290 #define ON64(x) x 291 #else 292 #define ON64(x) 293 #endif 294 295 /* 296 * fastop functions have a special calling convention: 297 * 298 * dst: rax (in/out) 299 * src: rdx (in/out) 300 * src2: rcx (in) 301 * flags: rflags (in/out) 302 * ex: rsi (in:fastop pointer, out:zero if exception) 303 * 304 * Moreover, they are all exactly FASTOP_SIZE bytes long, so functions for 305 * different operand sizes can be reached by calculation, rather than a jump 306 * table (which would be bigger than the code). 307 */ 308 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop); 309 310 #define __FOP_FUNC(name) \ 311 ".align " __stringify(FASTOP_SIZE) " \n\t" \ 312 ".type " name ", @function \n\t" \ 313 name ":\n\t" 314 315 #define FOP_FUNC(name) \ 316 __FOP_FUNC(#name) 317 318 #define __FOP_RET(name) \ 319 "ret \n\t" \ 320 ".size " name ", .-" name "\n\t" 321 322 #define FOP_RET(name) \ 323 __FOP_RET(#name) 324 325 #define FOP_START(op) \ 326 extern void em_##op(struct fastop *fake); \ 327 asm(".pushsection .text, \"ax\" \n\t" \ 328 ".global em_" #op " \n\t" \ 329 ".align " __stringify(FASTOP_SIZE) " \n\t" \ 330 "em_" #op ":\n\t" 331 332 #define FOP_END \ 333 ".popsection") 334 335 #define __FOPNOP(name) \ 336 __FOP_FUNC(name) \ 337 __FOP_RET(name) 338 339 #define FOPNOP() \ 340 __FOPNOP(__stringify(__UNIQUE_ID(nop))) 341 342 #define FOP1E(op, dst) \ 343 __FOP_FUNC(#op "_" #dst) \ 344 "10: " #op " %" #dst " \n\t" \ 345 __FOP_RET(#op "_" #dst) 346 347 #define FOP1EEX(op, dst) \ 348 FOP1E(op, dst) _ASM_EXTABLE(10b, kvm_fastop_exception) 349 350 #define FASTOP1(op) \ 351 FOP_START(op) \ 352 FOP1E(op##b, al) \ 353 FOP1E(op##w, ax) \ 354 FOP1E(op##l, eax) \ 355 ON64(FOP1E(op##q, rax)) \ 356 FOP_END 357 358 /* 1-operand, using src2 (for MUL/DIV r/m) */ 359 #define FASTOP1SRC2(op, name) \ 360 FOP_START(name) \ 361 FOP1E(op, cl) \ 362 FOP1E(op, cx) \ 363 FOP1E(op, ecx) \ 364 ON64(FOP1E(op, rcx)) \ 365 FOP_END 366 367 /* 1-operand, using src2 (for MUL/DIV r/m), with exceptions */ 368 #define FASTOP1SRC2EX(op, name) \ 369 FOP_START(name) \ 370 FOP1EEX(op, cl) \ 371 FOP1EEX(op, cx) \ 372 FOP1EEX(op, ecx) \ 373 ON64(FOP1EEX(op, rcx)) \ 374 FOP_END 375 376 #define FOP2E(op, dst, src) \ 377 __FOP_FUNC(#op "_" #dst "_" #src) \ 378 #op " %" #src ", %" #dst " \n\t" \ 379 __FOP_RET(#op "_" #dst "_" #src) 380 381 #define FASTOP2(op) \ 382 FOP_START(op) \ 383 FOP2E(op##b, al, dl) \ 384 FOP2E(op##w, ax, dx) \ 385 FOP2E(op##l, eax, edx) \ 386 ON64(FOP2E(op##q, rax, rdx)) \ 387 FOP_END 388 389 /* 2 operand, word only */ 390 #define FASTOP2W(op) \ 391 FOP_START(op) \ 392 FOPNOP() \ 393 FOP2E(op##w, ax, dx) \ 394 FOP2E(op##l, eax, edx) \ 395 ON64(FOP2E(op##q, rax, rdx)) \ 396 FOP_END 397 398 /* 2 operand, src is CL */ 399 #define FASTOP2CL(op) \ 400 FOP_START(op) \ 401 FOP2E(op##b, al, cl) \ 402 FOP2E(op##w, ax, cl) \ 403 FOP2E(op##l, eax, cl) \ 404 ON64(FOP2E(op##q, rax, cl)) \ 405 FOP_END 406 407 /* 2 operand, src and dest are reversed */ 408 #define FASTOP2R(op, name) \ 409 FOP_START(name) \ 410 FOP2E(op##b, dl, al) \ 411 FOP2E(op##w, dx, ax) \ 412 FOP2E(op##l, edx, eax) \ 413 ON64(FOP2E(op##q, rdx, rax)) \ 414 FOP_END 415 416 #define FOP3E(op, dst, src, src2) \ 417 __FOP_FUNC(#op "_" #dst "_" #src "_" #src2) \ 418 #op " %" #src2 ", %" #src ", %" #dst " \n\t"\ 419 __FOP_RET(#op "_" #dst "_" #src "_" #src2) 420 421 /* 3-operand, word-only, src2=cl */ 422 #define FASTOP3WCL(op) \ 423 FOP_START(op) \ 424 FOPNOP() \ 425 FOP3E(op##w, ax, dx, cl) \ 426 FOP3E(op##l, eax, edx, cl) \ 427 ON64(FOP3E(op##q, rax, rdx, cl)) \ 428 FOP_END 429 430 /* Special case for SETcc - 1 instruction per cc */ 431 #define FOP_SETCC(op) \ 432 ".align 4 \n\t" \ 433 ".type " #op ", @function \n\t" \ 434 #op ": \n\t" \ 435 #op " %al \n\t" \ 436 __FOP_RET(#op) 437 438 asm(".pushsection .fixup, \"ax\"\n" 439 ".global kvm_fastop_exception \n" 440 "kvm_fastop_exception: xor %esi, %esi; ret\n" 441 ".popsection"); 442 443 FOP_START(setcc) 444 FOP_SETCC(seto) 445 FOP_SETCC(setno) 446 FOP_SETCC(setc) 447 FOP_SETCC(setnc) 448 FOP_SETCC(setz) 449 FOP_SETCC(setnz) 450 FOP_SETCC(setbe) 451 FOP_SETCC(setnbe) 452 FOP_SETCC(sets) 453 FOP_SETCC(setns) 454 FOP_SETCC(setp) 455 FOP_SETCC(setnp) 456 FOP_SETCC(setl) 457 FOP_SETCC(setnl) 458 FOP_SETCC(setle) 459 FOP_SETCC(setnle) 460 FOP_END; 461 462 FOP_START(salc) 463 FOP_FUNC(salc) 464 "pushf; sbb %al, %al; popf \n\t" 465 FOP_RET(salc) 466 FOP_END; 467 468 /* 469 * XXX: inoutclob user must know where the argument is being expanded. 470 * Relying on CONFIG_CC_HAS_ASM_GOTO would allow us to remove _fault. 471 */ 472 #define asm_safe(insn, inoutclob...) \ 473 ({ \ 474 int _fault = 0; \ 475 \ 476 asm volatile("1:" insn "\n" \ 477 "2:\n" \ 478 ".pushsection .fixup, \"ax\"\n" \ 479 "3: movl $1, %[_fault]\n" \ 480 " jmp 2b\n" \ 481 ".popsection\n" \ 482 _ASM_EXTABLE(1b, 3b) \ 483 : [_fault] "+qm"(_fault) inoutclob ); \ 484 \ 485 _fault ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; \ 486 }) 487 488 static int emulator_check_intercept(struct x86_emulate_ctxt *ctxt, 489 enum x86_intercept intercept, 490 enum x86_intercept_stage stage) 491 { 492 struct x86_instruction_info info = { 493 .intercept = intercept, 494 .rep_prefix = ctxt->rep_prefix, 495 .modrm_mod = ctxt->modrm_mod, 496 .modrm_reg = ctxt->modrm_reg, 497 .modrm_rm = ctxt->modrm_rm, 498 .src_val = ctxt->src.val64, 499 .dst_val = ctxt->dst.val64, 500 .src_bytes = ctxt->src.bytes, 501 .dst_bytes = ctxt->dst.bytes, 502 .ad_bytes = ctxt->ad_bytes, 503 .next_rip = ctxt->eip, 504 }; 505 506 return ctxt->ops->intercept(ctxt, &info, stage); 507 } 508 509 static void assign_masked(ulong *dest, ulong src, ulong mask) 510 { 511 *dest = (*dest & ~mask) | (src & mask); 512 } 513 514 static void assign_register(unsigned long *reg, u64 val, int bytes) 515 { 516 /* The 4-byte case *is* correct: in 64-bit mode we zero-extend. */ 517 switch (bytes) { 518 case 1: 519 *(u8 *)reg = (u8)val; 520 break; 521 case 2: 522 *(u16 *)reg = (u16)val; 523 break; 524 case 4: 525 *reg = (u32)val; 526 break; /* 64b: zero-extend */ 527 case 8: 528 *reg = val; 529 break; 530 } 531 } 532 533 static inline unsigned long ad_mask(struct x86_emulate_ctxt *ctxt) 534 { 535 return (1UL << (ctxt->ad_bytes << 3)) - 1; 536 } 537 538 static ulong stack_mask(struct x86_emulate_ctxt *ctxt) 539 { 540 u16 sel; 541 struct desc_struct ss; 542 543 if (ctxt->mode == X86EMUL_MODE_PROT64) 544 return ~0UL; 545 ctxt->ops->get_segment(ctxt, &sel, &ss, NULL, VCPU_SREG_SS); 546 return ~0U >> ((ss.d ^ 1) * 16); /* d=0: 0xffff; d=1: 0xffffffff */ 547 } 548 549 static int stack_size(struct x86_emulate_ctxt *ctxt) 550 { 551 return (__fls(stack_mask(ctxt)) + 1) >> 3; 552 } 553 554 /* Access/update address held in a register, based on addressing mode. */ 555 static inline unsigned long 556 address_mask(struct x86_emulate_ctxt *ctxt, unsigned long reg) 557 { 558 if (ctxt->ad_bytes == sizeof(unsigned long)) 559 return reg; 560 else 561 return reg & ad_mask(ctxt); 562 } 563 564 static inline unsigned long 565 register_address(struct x86_emulate_ctxt *ctxt, int reg) 566 { 567 return address_mask(ctxt, reg_read(ctxt, reg)); 568 } 569 570 static void masked_increment(ulong *reg, ulong mask, int inc) 571 { 572 assign_masked(reg, *reg + inc, mask); 573 } 574 575 static inline void 576 register_address_increment(struct x86_emulate_ctxt *ctxt, int reg, int inc) 577 { 578 ulong *preg = reg_rmw(ctxt, reg); 579 580 assign_register(preg, *preg + inc, ctxt->ad_bytes); 581 } 582 583 static void rsp_increment(struct x86_emulate_ctxt *ctxt, int inc) 584 { 585 masked_increment(reg_rmw(ctxt, VCPU_REGS_RSP), stack_mask(ctxt), inc); 586 } 587 588 static u32 desc_limit_scaled(struct desc_struct *desc) 589 { 590 u32 limit = get_desc_limit(desc); 591 592 return desc->g ? (limit << 12) | 0xfff : limit; 593 } 594 595 static unsigned long seg_base(struct x86_emulate_ctxt *ctxt, int seg) 596 { 597 if (ctxt->mode == X86EMUL_MODE_PROT64 && seg < VCPU_SREG_FS) 598 return 0; 599 600 return ctxt->ops->get_cached_segment_base(ctxt, seg); 601 } 602 603 static int emulate_exception(struct x86_emulate_ctxt *ctxt, int vec, 604 u32 error, bool valid) 605 { 606 WARN_ON(vec > 0x1f); 607 ctxt->exception.vector = vec; 608 ctxt->exception.error_code = error; 609 ctxt->exception.error_code_valid = valid; 610 return X86EMUL_PROPAGATE_FAULT; 611 } 612 613 static int emulate_db(struct x86_emulate_ctxt *ctxt) 614 { 615 return emulate_exception(ctxt, DB_VECTOR, 0, false); 616 } 617 618 static int emulate_gp(struct x86_emulate_ctxt *ctxt, int err) 619 { 620 return emulate_exception(ctxt, GP_VECTOR, err, true); 621 } 622 623 static int emulate_ss(struct x86_emulate_ctxt *ctxt, int err) 624 { 625 return emulate_exception(ctxt, SS_VECTOR, err, true); 626 } 627 628 static int emulate_ud(struct x86_emulate_ctxt *ctxt) 629 { 630 return emulate_exception(ctxt, UD_VECTOR, 0, false); 631 } 632 633 static int emulate_ts(struct x86_emulate_ctxt *ctxt, int err) 634 { 635 return emulate_exception(ctxt, TS_VECTOR, err, true); 636 } 637 638 static int emulate_de(struct x86_emulate_ctxt *ctxt) 639 { 640 return emulate_exception(ctxt, DE_VECTOR, 0, false); 641 } 642 643 static int emulate_nm(struct x86_emulate_ctxt *ctxt) 644 { 645 return emulate_exception(ctxt, NM_VECTOR, 0, false); 646 } 647 648 static u16 get_segment_selector(struct x86_emulate_ctxt *ctxt, unsigned seg) 649 { 650 u16 selector; 651 struct desc_struct desc; 652 653 ctxt->ops->get_segment(ctxt, &selector, &desc, NULL, seg); 654 return selector; 655 } 656 657 static void set_segment_selector(struct x86_emulate_ctxt *ctxt, u16 selector, 658 unsigned seg) 659 { 660 u16 dummy; 661 u32 base3; 662 struct desc_struct desc; 663 664 ctxt->ops->get_segment(ctxt, &dummy, &desc, &base3, seg); 665 ctxt->ops->set_segment(ctxt, selector, &desc, base3, seg); 666 } 667 668 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt) 669 { 670 return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48; 671 } 672 673 static inline bool emul_is_noncanonical_address(u64 la, 674 struct x86_emulate_ctxt *ctxt) 675 { 676 return get_canonical(la, ctxt_virt_addr_bits(ctxt)) != la; 677 } 678 679 /* 680 * x86 defines three classes of vector instructions: explicitly 681 * aligned, explicitly unaligned, and the rest, which change behaviour 682 * depending on whether they're AVX encoded or not. 683 * 684 * Also included is CMPXCHG16B which is not a vector instruction, yet it is 685 * subject to the same check. FXSAVE and FXRSTOR are checked here too as their 686 * 512 bytes of data must be aligned to a 16 byte boundary. 687 */ 688 static unsigned insn_alignment(struct x86_emulate_ctxt *ctxt, unsigned size) 689 { 690 u64 alignment = ctxt->d & AlignMask; 691 692 if (likely(size < 16)) 693 return 1; 694 695 switch (alignment) { 696 case Unaligned: 697 case Avx: 698 return 1; 699 case Aligned16: 700 return 16; 701 case Aligned: 702 default: 703 return size; 704 } 705 } 706 707 static __always_inline int __linearize(struct x86_emulate_ctxt *ctxt, 708 struct segmented_address addr, 709 unsigned *max_size, unsigned size, 710 bool write, bool fetch, 711 enum x86emul_mode mode, ulong *linear) 712 { 713 struct desc_struct desc; 714 bool usable; 715 ulong la; 716 u32 lim; 717 u16 sel; 718 u8 va_bits; 719 720 la = seg_base(ctxt, addr.seg) + addr.ea; 721 *max_size = 0; 722 switch (mode) { 723 case X86EMUL_MODE_PROT64: 724 *linear = la; 725 va_bits = ctxt_virt_addr_bits(ctxt); 726 if (get_canonical(la, va_bits) != la) 727 goto bad; 728 729 *max_size = min_t(u64, ~0u, (1ull << va_bits) - la); 730 if (size > *max_size) 731 goto bad; 732 break; 733 default: 734 *linear = la = (u32)la; 735 usable = ctxt->ops->get_segment(ctxt, &sel, &desc, NULL, 736 addr.seg); 737 if (!usable) 738 goto bad; 739 /* code segment in protected mode or read-only data segment */ 740 if ((((ctxt->mode != X86EMUL_MODE_REAL) && (desc.type & 8)) 741 || !(desc.type & 2)) && write) 742 goto bad; 743 /* unreadable code segment */ 744 if (!fetch && (desc.type & 8) && !(desc.type & 2)) 745 goto bad; 746 lim = desc_limit_scaled(&desc); 747 if (!(desc.type & 8) && (desc.type & 4)) { 748 /* expand-down segment */ 749 if (addr.ea <= lim) 750 goto bad; 751 lim = desc.d ? 0xffffffff : 0xffff; 752 } 753 if (addr.ea > lim) 754 goto bad; 755 if (lim == 0xffffffff) 756 *max_size = ~0u; 757 else { 758 *max_size = (u64)lim + 1 - addr.ea; 759 if (size > *max_size) 760 goto bad; 761 } 762 break; 763 } 764 if (la & (insn_alignment(ctxt, size) - 1)) 765 return emulate_gp(ctxt, 0); 766 return X86EMUL_CONTINUE; 767 bad: 768 if (addr.seg == VCPU_SREG_SS) 769 return emulate_ss(ctxt, 0); 770 else 771 return emulate_gp(ctxt, 0); 772 } 773 774 static int linearize(struct x86_emulate_ctxt *ctxt, 775 struct segmented_address addr, 776 unsigned size, bool write, 777 ulong *linear) 778 { 779 unsigned max_size; 780 return __linearize(ctxt, addr, &max_size, size, write, false, 781 ctxt->mode, linear); 782 } 783 784 static inline int assign_eip(struct x86_emulate_ctxt *ctxt, ulong dst, 785 enum x86emul_mode mode) 786 { 787 ulong linear; 788 int rc; 789 unsigned max_size; 790 struct segmented_address addr = { .seg = VCPU_SREG_CS, 791 .ea = dst }; 792 793 if (ctxt->op_bytes != sizeof(unsigned long)) 794 addr.ea = dst & ((1UL << (ctxt->op_bytes << 3)) - 1); 795 rc = __linearize(ctxt, addr, &max_size, 1, false, true, mode, &linear); 796 if (rc == X86EMUL_CONTINUE) 797 ctxt->_eip = addr.ea; 798 return rc; 799 } 800 801 static inline int assign_eip_near(struct x86_emulate_ctxt *ctxt, ulong dst) 802 { 803 return assign_eip(ctxt, dst, ctxt->mode); 804 } 805 806 static int assign_eip_far(struct x86_emulate_ctxt *ctxt, ulong dst, 807 const struct desc_struct *cs_desc) 808 { 809 enum x86emul_mode mode = ctxt->mode; 810 int rc; 811 812 #ifdef CONFIG_X86_64 813 if (ctxt->mode >= X86EMUL_MODE_PROT16) { 814 if (cs_desc->l) { 815 u64 efer = 0; 816 817 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 818 if (efer & EFER_LMA) 819 mode = X86EMUL_MODE_PROT64; 820 } else 821 mode = X86EMUL_MODE_PROT32; /* temporary value */ 822 } 823 #endif 824 if (mode == X86EMUL_MODE_PROT16 || mode == X86EMUL_MODE_PROT32) 825 mode = cs_desc->d ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16; 826 rc = assign_eip(ctxt, dst, mode); 827 if (rc == X86EMUL_CONTINUE) 828 ctxt->mode = mode; 829 return rc; 830 } 831 832 static inline int jmp_rel(struct x86_emulate_ctxt *ctxt, int rel) 833 { 834 return assign_eip_near(ctxt, ctxt->_eip + rel); 835 } 836 837 static int linear_read_system(struct x86_emulate_ctxt *ctxt, ulong linear, 838 void *data, unsigned size) 839 { 840 return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, true); 841 } 842 843 static int linear_write_system(struct x86_emulate_ctxt *ctxt, 844 ulong linear, void *data, 845 unsigned int size) 846 { 847 return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, true); 848 } 849 850 static int segmented_read_std(struct x86_emulate_ctxt *ctxt, 851 struct segmented_address addr, 852 void *data, 853 unsigned size) 854 { 855 int rc; 856 ulong linear; 857 858 rc = linearize(ctxt, addr, size, false, &linear); 859 if (rc != X86EMUL_CONTINUE) 860 return rc; 861 return ctxt->ops->read_std(ctxt, linear, data, size, &ctxt->exception, false); 862 } 863 864 static int segmented_write_std(struct x86_emulate_ctxt *ctxt, 865 struct segmented_address addr, 866 void *data, 867 unsigned int size) 868 { 869 int rc; 870 ulong linear; 871 872 rc = linearize(ctxt, addr, size, true, &linear); 873 if (rc != X86EMUL_CONTINUE) 874 return rc; 875 return ctxt->ops->write_std(ctxt, linear, data, size, &ctxt->exception, false); 876 } 877 878 /* 879 * Prefetch the remaining bytes of the instruction without crossing page 880 * boundary if they are not in fetch_cache yet. 881 */ 882 static int __do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, int op_size) 883 { 884 int rc; 885 unsigned size, max_size; 886 unsigned long linear; 887 int cur_size = ctxt->fetch.end - ctxt->fetch.data; 888 struct segmented_address addr = { .seg = VCPU_SREG_CS, 889 .ea = ctxt->eip + cur_size }; 890 891 /* 892 * We do not know exactly how many bytes will be needed, and 893 * __linearize is expensive, so fetch as much as possible. We 894 * just have to avoid going beyond the 15 byte limit, the end 895 * of the segment, or the end of the page. 896 * 897 * __linearize is called with size 0 so that it does not do any 898 * boundary check itself. Instead, we use max_size to check 899 * against op_size. 900 */ 901 rc = __linearize(ctxt, addr, &max_size, 0, false, true, ctxt->mode, 902 &linear); 903 if (unlikely(rc != X86EMUL_CONTINUE)) 904 return rc; 905 906 size = min_t(unsigned, 15UL ^ cur_size, max_size); 907 size = min_t(unsigned, size, PAGE_SIZE - offset_in_page(linear)); 908 909 /* 910 * One instruction can only straddle two pages, 911 * and one has been loaded at the beginning of 912 * x86_decode_insn. So, if not enough bytes 913 * still, we must have hit the 15-byte boundary. 914 */ 915 if (unlikely(size < op_size)) 916 return emulate_gp(ctxt, 0); 917 918 rc = ctxt->ops->fetch(ctxt, linear, ctxt->fetch.end, 919 size, &ctxt->exception); 920 if (unlikely(rc != X86EMUL_CONTINUE)) 921 return rc; 922 ctxt->fetch.end += size; 923 return X86EMUL_CONTINUE; 924 } 925 926 static __always_inline int do_insn_fetch_bytes(struct x86_emulate_ctxt *ctxt, 927 unsigned size) 928 { 929 unsigned done_size = ctxt->fetch.end - ctxt->fetch.ptr; 930 931 if (unlikely(done_size < size)) 932 return __do_insn_fetch_bytes(ctxt, size - done_size); 933 else 934 return X86EMUL_CONTINUE; 935 } 936 937 /* Fetch next part of the instruction being emulated. */ 938 #define insn_fetch(_type, _ctxt) \ 939 ({ _type _x; \ 940 \ 941 rc = do_insn_fetch_bytes(_ctxt, sizeof(_type)); \ 942 if (rc != X86EMUL_CONTINUE) \ 943 goto done; \ 944 ctxt->_eip += sizeof(_type); \ 945 memcpy(&_x, ctxt->fetch.ptr, sizeof(_type)); \ 946 ctxt->fetch.ptr += sizeof(_type); \ 947 _x; \ 948 }) 949 950 #define insn_fetch_arr(_arr, _size, _ctxt) \ 951 ({ \ 952 rc = do_insn_fetch_bytes(_ctxt, _size); \ 953 if (rc != X86EMUL_CONTINUE) \ 954 goto done; \ 955 ctxt->_eip += (_size); \ 956 memcpy(_arr, ctxt->fetch.ptr, _size); \ 957 ctxt->fetch.ptr += (_size); \ 958 }) 959 960 /* 961 * Given the 'reg' portion of a ModRM byte, and a register block, return a 962 * pointer into the block that addresses the relevant register. 963 * @highbyte_regs specifies whether to decode AH,CH,DH,BH. 964 */ 965 static void *decode_register(struct x86_emulate_ctxt *ctxt, u8 modrm_reg, 966 int byteop) 967 { 968 void *p; 969 int highbyte_regs = (ctxt->rex_prefix == 0) && byteop; 970 971 if (highbyte_regs && modrm_reg >= 4 && modrm_reg < 8) 972 p = (unsigned char *)reg_rmw(ctxt, modrm_reg & 3) + 1; 973 else 974 p = reg_rmw(ctxt, modrm_reg); 975 return p; 976 } 977 978 static int read_descriptor(struct x86_emulate_ctxt *ctxt, 979 struct segmented_address addr, 980 u16 *size, unsigned long *address, int op_bytes) 981 { 982 int rc; 983 984 if (op_bytes == 2) 985 op_bytes = 3; 986 *address = 0; 987 rc = segmented_read_std(ctxt, addr, size, 2); 988 if (rc != X86EMUL_CONTINUE) 989 return rc; 990 addr.ea += 2; 991 rc = segmented_read_std(ctxt, addr, address, op_bytes); 992 return rc; 993 } 994 995 FASTOP2(add); 996 FASTOP2(or); 997 FASTOP2(adc); 998 FASTOP2(sbb); 999 FASTOP2(and); 1000 FASTOP2(sub); 1001 FASTOP2(xor); 1002 FASTOP2(cmp); 1003 FASTOP2(test); 1004 1005 FASTOP1SRC2(mul, mul_ex); 1006 FASTOP1SRC2(imul, imul_ex); 1007 FASTOP1SRC2EX(div, div_ex); 1008 FASTOP1SRC2EX(idiv, idiv_ex); 1009 1010 FASTOP3WCL(shld); 1011 FASTOP3WCL(shrd); 1012 1013 FASTOP2W(imul); 1014 1015 FASTOP1(not); 1016 FASTOP1(neg); 1017 FASTOP1(inc); 1018 FASTOP1(dec); 1019 1020 FASTOP2CL(rol); 1021 FASTOP2CL(ror); 1022 FASTOP2CL(rcl); 1023 FASTOP2CL(rcr); 1024 FASTOP2CL(shl); 1025 FASTOP2CL(shr); 1026 FASTOP2CL(sar); 1027 1028 FASTOP2W(bsf); 1029 FASTOP2W(bsr); 1030 FASTOP2W(bt); 1031 FASTOP2W(bts); 1032 FASTOP2W(btr); 1033 FASTOP2W(btc); 1034 1035 FASTOP2(xadd); 1036 1037 FASTOP2R(cmp, cmp_r); 1038 1039 static int em_bsf_c(struct x86_emulate_ctxt *ctxt) 1040 { 1041 /* If src is zero, do not writeback, but update flags */ 1042 if (ctxt->src.val == 0) 1043 ctxt->dst.type = OP_NONE; 1044 return fastop(ctxt, em_bsf); 1045 } 1046 1047 static int em_bsr_c(struct x86_emulate_ctxt *ctxt) 1048 { 1049 /* If src is zero, do not writeback, but update flags */ 1050 if (ctxt->src.val == 0) 1051 ctxt->dst.type = OP_NONE; 1052 return fastop(ctxt, em_bsr); 1053 } 1054 1055 static __always_inline u8 test_cc(unsigned int condition, unsigned long flags) 1056 { 1057 u8 rc; 1058 void (*fop)(void) = (void *)em_setcc + 4 * (condition & 0xf); 1059 1060 flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF; 1061 asm("push %[flags]; popf; " CALL_NOSPEC 1062 : "=a"(rc) : [thunk_target]"r"(fop), [flags]"r"(flags)); 1063 return rc; 1064 } 1065 1066 static void fetch_register_operand(struct operand *op) 1067 { 1068 switch (op->bytes) { 1069 case 1: 1070 op->val = *(u8 *)op->addr.reg; 1071 break; 1072 case 2: 1073 op->val = *(u16 *)op->addr.reg; 1074 break; 1075 case 4: 1076 op->val = *(u32 *)op->addr.reg; 1077 break; 1078 case 8: 1079 op->val = *(u64 *)op->addr.reg; 1080 break; 1081 } 1082 } 1083 1084 static void emulator_get_fpu(void) 1085 { 1086 fpregs_lock(); 1087 1088 fpregs_assert_state_consistent(); 1089 if (test_thread_flag(TIF_NEED_FPU_LOAD)) 1090 switch_fpu_return(); 1091 } 1092 1093 static void emulator_put_fpu(void) 1094 { 1095 fpregs_unlock(); 1096 } 1097 1098 static void read_sse_reg(sse128_t *data, int reg) 1099 { 1100 emulator_get_fpu(); 1101 switch (reg) { 1102 case 0: asm("movdqa %%xmm0, %0" : "=m"(*data)); break; 1103 case 1: asm("movdqa %%xmm1, %0" : "=m"(*data)); break; 1104 case 2: asm("movdqa %%xmm2, %0" : "=m"(*data)); break; 1105 case 3: asm("movdqa %%xmm3, %0" : "=m"(*data)); break; 1106 case 4: asm("movdqa %%xmm4, %0" : "=m"(*data)); break; 1107 case 5: asm("movdqa %%xmm5, %0" : "=m"(*data)); break; 1108 case 6: asm("movdqa %%xmm6, %0" : "=m"(*data)); break; 1109 case 7: asm("movdqa %%xmm7, %0" : "=m"(*data)); break; 1110 #ifdef CONFIG_X86_64 1111 case 8: asm("movdqa %%xmm8, %0" : "=m"(*data)); break; 1112 case 9: asm("movdqa %%xmm9, %0" : "=m"(*data)); break; 1113 case 10: asm("movdqa %%xmm10, %0" : "=m"(*data)); break; 1114 case 11: asm("movdqa %%xmm11, %0" : "=m"(*data)); break; 1115 case 12: asm("movdqa %%xmm12, %0" : "=m"(*data)); break; 1116 case 13: asm("movdqa %%xmm13, %0" : "=m"(*data)); break; 1117 case 14: asm("movdqa %%xmm14, %0" : "=m"(*data)); break; 1118 case 15: asm("movdqa %%xmm15, %0" : "=m"(*data)); break; 1119 #endif 1120 default: BUG(); 1121 } 1122 emulator_put_fpu(); 1123 } 1124 1125 static void write_sse_reg(sse128_t *data, int reg) 1126 { 1127 emulator_get_fpu(); 1128 switch (reg) { 1129 case 0: asm("movdqa %0, %%xmm0" : : "m"(*data)); break; 1130 case 1: asm("movdqa %0, %%xmm1" : : "m"(*data)); break; 1131 case 2: asm("movdqa %0, %%xmm2" : : "m"(*data)); break; 1132 case 3: asm("movdqa %0, %%xmm3" : : "m"(*data)); break; 1133 case 4: asm("movdqa %0, %%xmm4" : : "m"(*data)); break; 1134 case 5: asm("movdqa %0, %%xmm5" : : "m"(*data)); break; 1135 case 6: asm("movdqa %0, %%xmm6" : : "m"(*data)); break; 1136 case 7: asm("movdqa %0, %%xmm7" : : "m"(*data)); break; 1137 #ifdef CONFIG_X86_64 1138 case 8: asm("movdqa %0, %%xmm8" : : "m"(*data)); break; 1139 case 9: asm("movdqa %0, %%xmm9" : : "m"(*data)); break; 1140 case 10: asm("movdqa %0, %%xmm10" : : "m"(*data)); break; 1141 case 11: asm("movdqa %0, %%xmm11" : : "m"(*data)); break; 1142 case 12: asm("movdqa %0, %%xmm12" : : "m"(*data)); break; 1143 case 13: asm("movdqa %0, %%xmm13" : : "m"(*data)); break; 1144 case 14: asm("movdqa %0, %%xmm14" : : "m"(*data)); break; 1145 case 15: asm("movdqa %0, %%xmm15" : : "m"(*data)); break; 1146 #endif 1147 default: BUG(); 1148 } 1149 emulator_put_fpu(); 1150 } 1151 1152 static void read_mmx_reg(u64 *data, int reg) 1153 { 1154 emulator_get_fpu(); 1155 switch (reg) { 1156 case 0: asm("movq %%mm0, %0" : "=m"(*data)); break; 1157 case 1: asm("movq %%mm1, %0" : "=m"(*data)); break; 1158 case 2: asm("movq %%mm2, %0" : "=m"(*data)); break; 1159 case 3: asm("movq %%mm3, %0" : "=m"(*data)); break; 1160 case 4: asm("movq %%mm4, %0" : "=m"(*data)); break; 1161 case 5: asm("movq %%mm5, %0" : "=m"(*data)); break; 1162 case 6: asm("movq %%mm6, %0" : "=m"(*data)); break; 1163 case 7: asm("movq %%mm7, %0" : "=m"(*data)); break; 1164 default: BUG(); 1165 } 1166 emulator_put_fpu(); 1167 } 1168 1169 static void write_mmx_reg(u64 *data, int reg) 1170 { 1171 emulator_get_fpu(); 1172 switch (reg) { 1173 case 0: asm("movq %0, %%mm0" : : "m"(*data)); break; 1174 case 1: asm("movq %0, %%mm1" : : "m"(*data)); break; 1175 case 2: asm("movq %0, %%mm2" : : "m"(*data)); break; 1176 case 3: asm("movq %0, %%mm3" : : "m"(*data)); break; 1177 case 4: asm("movq %0, %%mm4" : : "m"(*data)); break; 1178 case 5: asm("movq %0, %%mm5" : : "m"(*data)); break; 1179 case 6: asm("movq %0, %%mm6" : : "m"(*data)); break; 1180 case 7: asm("movq %0, %%mm7" : : "m"(*data)); break; 1181 default: BUG(); 1182 } 1183 emulator_put_fpu(); 1184 } 1185 1186 static int em_fninit(struct x86_emulate_ctxt *ctxt) 1187 { 1188 if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM)) 1189 return emulate_nm(ctxt); 1190 1191 emulator_get_fpu(); 1192 asm volatile("fninit"); 1193 emulator_put_fpu(); 1194 return X86EMUL_CONTINUE; 1195 } 1196 1197 static int em_fnstcw(struct x86_emulate_ctxt *ctxt) 1198 { 1199 u16 fcw; 1200 1201 if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM)) 1202 return emulate_nm(ctxt); 1203 1204 emulator_get_fpu(); 1205 asm volatile("fnstcw %0": "+m"(fcw)); 1206 emulator_put_fpu(); 1207 1208 ctxt->dst.val = fcw; 1209 1210 return X86EMUL_CONTINUE; 1211 } 1212 1213 static int em_fnstsw(struct x86_emulate_ctxt *ctxt) 1214 { 1215 u16 fsw; 1216 1217 if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM)) 1218 return emulate_nm(ctxt); 1219 1220 emulator_get_fpu(); 1221 asm volatile("fnstsw %0": "+m"(fsw)); 1222 emulator_put_fpu(); 1223 1224 ctxt->dst.val = fsw; 1225 1226 return X86EMUL_CONTINUE; 1227 } 1228 1229 static void decode_register_operand(struct x86_emulate_ctxt *ctxt, 1230 struct operand *op) 1231 { 1232 unsigned reg = ctxt->modrm_reg; 1233 1234 if (!(ctxt->d & ModRM)) 1235 reg = (ctxt->b & 7) | ((ctxt->rex_prefix & 1) << 3); 1236 1237 if (ctxt->d & Sse) { 1238 op->type = OP_XMM; 1239 op->bytes = 16; 1240 op->addr.xmm = reg; 1241 read_sse_reg(&op->vec_val, reg); 1242 return; 1243 } 1244 if (ctxt->d & Mmx) { 1245 reg &= 7; 1246 op->type = OP_MM; 1247 op->bytes = 8; 1248 op->addr.mm = reg; 1249 return; 1250 } 1251 1252 op->type = OP_REG; 1253 op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 1254 op->addr.reg = decode_register(ctxt, reg, ctxt->d & ByteOp); 1255 1256 fetch_register_operand(op); 1257 op->orig_val = op->val; 1258 } 1259 1260 static void adjust_modrm_seg(struct x86_emulate_ctxt *ctxt, int base_reg) 1261 { 1262 if (base_reg == VCPU_REGS_RSP || base_reg == VCPU_REGS_RBP) 1263 ctxt->modrm_seg = VCPU_SREG_SS; 1264 } 1265 1266 static int decode_modrm(struct x86_emulate_ctxt *ctxt, 1267 struct operand *op) 1268 { 1269 u8 sib; 1270 int index_reg, base_reg, scale; 1271 int rc = X86EMUL_CONTINUE; 1272 ulong modrm_ea = 0; 1273 1274 ctxt->modrm_reg = ((ctxt->rex_prefix << 1) & 8); /* REX.R */ 1275 index_reg = (ctxt->rex_prefix << 2) & 8; /* REX.X */ 1276 base_reg = (ctxt->rex_prefix << 3) & 8; /* REX.B */ 1277 1278 ctxt->modrm_mod = (ctxt->modrm & 0xc0) >> 6; 1279 ctxt->modrm_reg |= (ctxt->modrm & 0x38) >> 3; 1280 ctxt->modrm_rm = base_reg | (ctxt->modrm & 0x07); 1281 ctxt->modrm_seg = VCPU_SREG_DS; 1282 1283 if (ctxt->modrm_mod == 3 || (ctxt->d & NoMod)) { 1284 op->type = OP_REG; 1285 op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 1286 op->addr.reg = decode_register(ctxt, ctxt->modrm_rm, 1287 ctxt->d & ByteOp); 1288 if (ctxt->d & Sse) { 1289 op->type = OP_XMM; 1290 op->bytes = 16; 1291 op->addr.xmm = ctxt->modrm_rm; 1292 read_sse_reg(&op->vec_val, ctxt->modrm_rm); 1293 return rc; 1294 } 1295 if (ctxt->d & Mmx) { 1296 op->type = OP_MM; 1297 op->bytes = 8; 1298 op->addr.mm = ctxt->modrm_rm & 7; 1299 return rc; 1300 } 1301 fetch_register_operand(op); 1302 return rc; 1303 } 1304 1305 op->type = OP_MEM; 1306 1307 if (ctxt->ad_bytes == 2) { 1308 unsigned bx = reg_read(ctxt, VCPU_REGS_RBX); 1309 unsigned bp = reg_read(ctxt, VCPU_REGS_RBP); 1310 unsigned si = reg_read(ctxt, VCPU_REGS_RSI); 1311 unsigned di = reg_read(ctxt, VCPU_REGS_RDI); 1312 1313 /* 16-bit ModR/M decode. */ 1314 switch (ctxt->modrm_mod) { 1315 case 0: 1316 if (ctxt->modrm_rm == 6) 1317 modrm_ea += insn_fetch(u16, ctxt); 1318 break; 1319 case 1: 1320 modrm_ea += insn_fetch(s8, ctxt); 1321 break; 1322 case 2: 1323 modrm_ea += insn_fetch(u16, ctxt); 1324 break; 1325 } 1326 switch (ctxt->modrm_rm) { 1327 case 0: 1328 modrm_ea += bx + si; 1329 break; 1330 case 1: 1331 modrm_ea += bx + di; 1332 break; 1333 case 2: 1334 modrm_ea += bp + si; 1335 break; 1336 case 3: 1337 modrm_ea += bp + di; 1338 break; 1339 case 4: 1340 modrm_ea += si; 1341 break; 1342 case 5: 1343 modrm_ea += di; 1344 break; 1345 case 6: 1346 if (ctxt->modrm_mod != 0) 1347 modrm_ea += bp; 1348 break; 1349 case 7: 1350 modrm_ea += bx; 1351 break; 1352 } 1353 if (ctxt->modrm_rm == 2 || ctxt->modrm_rm == 3 || 1354 (ctxt->modrm_rm == 6 && ctxt->modrm_mod != 0)) 1355 ctxt->modrm_seg = VCPU_SREG_SS; 1356 modrm_ea = (u16)modrm_ea; 1357 } else { 1358 /* 32/64-bit ModR/M decode. */ 1359 if ((ctxt->modrm_rm & 7) == 4) { 1360 sib = insn_fetch(u8, ctxt); 1361 index_reg |= (sib >> 3) & 7; 1362 base_reg |= sib & 7; 1363 scale = sib >> 6; 1364 1365 if ((base_reg & 7) == 5 && ctxt->modrm_mod == 0) 1366 modrm_ea += insn_fetch(s32, ctxt); 1367 else { 1368 modrm_ea += reg_read(ctxt, base_reg); 1369 adjust_modrm_seg(ctxt, base_reg); 1370 /* Increment ESP on POP [ESP] */ 1371 if ((ctxt->d & IncSP) && 1372 base_reg == VCPU_REGS_RSP) 1373 modrm_ea += ctxt->op_bytes; 1374 } 1375 if (index_reg != 4) 1376 modrm_ea += reg_read(ctxt, index_reg) << scale; 1377 } else if ((ctxt->modrm_rm & 7) == 5 && ctxt->modrm_mod == 0) { 1378 modrm_ea += insn_fetch(s32, ctxt); 1379 if (ctxt->mode == X86EMUL_MODE_PROT64) 1380 ctxt->rip_relative = 1; 1381 } else { 1382 base_reg = ctxt->modrm_rm; 1383 modrm_ea += reg_read(ctxt, base_reg); 1384 adjust_modrm_seg(ctxt, base_reg); 1385 } 1386 switch (ctxt->modrm_mod) { 1387 case 1: 1388 modrm_ea += insn_fetch(s8, ctxt); 1389 break; 1390 case 2: 1391 modrm_ea += insn_fetch(s32, ctxt); 1392 break; 1393 } 1394 } 1395 op->addr.mem.ea = modrm_ea; 1396 if (ctxt->ad_bytes != 8) 1397 ctxt->memop.addr.mem.ea = (u32)ctxt->memop.addr.mem.ea; 1398 1399 done: 1400 return rc; 1401 } 1402 1403 static int decode_abs(struct x86_emulate_ctxt *ctxt, 1404 struct operand *op) 1405 { 1406 int rc = X86EMUL_CONTINUE; 1407 1408 op->type = OP_MEM; 1409 switch (ctxt->ad_bytes) { 1410 case 2: 1411 op->addr.mem.ea = insn_fetch(u16, ctxt); 1412 break; 1413 case 4: 1414 op->addr.mem.ea = insn_fetch(u32, ctxt); 1415 break; 1416 case 8: 1417 op->addr.mem.ea = insn_fetch(u64, ctxt); 1418 break; 1419 } 1420 done: 1421 return rc; 1422 } 1423 1424 static void fetch_bit_operand(struct x86_emulate_ctxt *ctxt) 1425 { 1426 long sv = 0, mask; 1427 1428 if (ctxt->dst.type == OP_MEM && ctxt->src.type == OP_REG) { 1429 mask = ~((long)ctxt->dst.bytes * 8 - 1); 1430 1431 if (ctxt->src.bytes == 2) 1432 sv = (s16)ctxt->src.val & (s16)mask; 1433 else if (ctxt->src.bytes == 4) 1434 sv = (s32)ctxt->src.val & (s32)mask; 1435 else 1436 sv = (s64)ctxt->src.val & (s64)mask; 1437 1438 ctxt->dst.addr.mem.ea = address_mask(ctxt, 1439 ctxt->dst.addr.mem.ea + (sv >> 3)); 1440 } 1441 1442 /* only subword offset */ 1443 ctxt->src.val &= (ctxt->dst.bytes << 3) - 1; 1444 } 1445 1446 static int read_emulated(struct x86_emulate_ctxt *ctxt, 1447 unsigned long addr, void *dest, unsigned size) 1448 { 1449 int rc; 1450 struct read_cache *mc = &ctxt->mem_read; 1451 1452 if (mc->pos < mc->end) 1453 goto read_cached; 1454 1455 WARN_ON((mc->end + size) >= sizeof(mc->data)); 1456 1457 rc = ctxt->ops->read_emulated(ctxt, addr, mc->data + mc->end, size, 1458 &ctxt->exception); 1459 if (rc != X86EMUL_CONTINUE) 1460 return rc; 1461 1462 mc->end += size; 1463 1464 read_cached: 1465 memcpy(dest, mc->data + mc->pos, size); 1466 mc->pos += size; 1467 return X86EMUL_CONTINUE; 1468 } 1469 1470 static int segmented_read(struct x86_emulate_ctxt *ctxt, 1471 struct segmented_address addr, 1472 void *data, 1473 unsigned size) 1474 { 1475 int rc; 1476 ulong linear; 1477 1478 rc = linearize(ctxt, addr, size, false, &linear); 1479 if (rc != X86EMUL_CONTINUE) 1480 return rc; 1481 return read_emulated(ctxt, linear, data, size); 1482 } 1483 1484 static int segmented_write(struct x86_emulate_ctxt *ctxt, 1485 struct segmented_address addr, 1486 const void *data, 1487 unsigned size) 1488 { 1489 int rc; 1490 ulong linear; 1491 1492 rc = linearize(ctxt, addr, size, true, &linear); 1493 if (rc != X86EMUL_CONTINUE) 1494 return rc; 1495 return ctxt->ops->write_emulated(ctxt, linear, data, size, 1496 &ctxt->exception); 1497 } 1498 1499 static int segmented_cmpxchg(struct x86_emulate_ctxt *ctxt, 1500 struct segmented_address addr, 1501 const void *orig_data, const void *data, 1502 unsigned size) 1503 { 1504 int rc; 1505 ulong linear; 1506 1507 rc = linearize(ctxt, addr, size, true, &linear); 1508 if (rc != X86EMUL_CONTINUE) 1509 return rc; 1510 return ctxt->ops->cmpxchg_emulated(ctxt, linear, orig_data, data, 1511 size, &ctxt->exception); 1512 } 1513 1514 static int pio_in_emulated(struct x86_emulate_ctxt *ctxt, 1515 unsigned int size, unsigned short port, 1516 void *dest) 1517 { 1518 struct read_cache *rc = &ctxt->io_read; 1519 1520 if (rc->pos == rc->end) { /* refill pio read ahead */ 1521 unsigned int in_page, n; 1522 unsigned int count = ctxt->rep_prefix ? 1523 address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) : 1; 1524 in_page = (ctxt->eflags & X86_EFLAGS_DF) ? 1525 offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)) : 1526 PAGE_SIZE - offset_in_page(reg_read(ctxt, VCPU_REGS_RDI)); 1527 n = min3(in_page, (unsigned int)sizeof(rc->data) / size, count); 1528 if (n == 0) 1529 n = 1; 1530 rc->pos = rc->end = 0; 1531 if (!ctxt->ops->pio_in_emulated(ctxt, size, port, rc->data, n)) 1532 return 0; 1533 rc->end = n * size; 1534 } 1535 1536 if (ctxt->rep_prefix && (ctxt->d & String) && 1537 !(ctxt->eflags & X86_EFLAGS_DF)) { 1538 ctxt->dst.data = rc->data + rc->pos; 1539 ctxt->dst.type = OP_MEM_STR; 1540 ctxt->dst.count = (rc->end - rc->pos) / size; 1541 rc->pos = rc->end; 1542 } else { 1543 memcpy(dest, rc->data + rc->pos, size); 1544 rc->pos += size; 1545 } 1546 return 1; 1547 } 1548 1549 static int read_interrupt_descriptor(struct x86_emulate_ctxt *ctxt, 1550 u16 index, struct desc_struct *desc) 1551 { 1552 struct desc_ptr dt; 1553 ulong addr; 1554 1555 ctxt->ops->get_idt(ctxt, &dt); 1556 1557 if (dt.size < index * 8 + 7) 1558 return emulate_gp(ctxt, index << 3 | 0x2); 1559 1560 addr = dt.address + index * 8; 1561 return linear_read_system(ctxt, addr, desc, sizeof(*desc)); 1562 } 1563 1564 static void get_descriptor_table_ptr(struct x86_emulate_ctxt *ctxt, 1565 u16 selector, struct desc_ptr *dt) 1566 { 1567 const struct x86_emulate_ops *ops = ctxt->ops; 1568 u32 base3 = 0; 1569 1570 if (selector & 1 << 2) { 1571 struct desc_struct desc; 1572 u16 sel; 1573 1574 memset(dt, 0, sizeof(*dt)); 1575 if (!ops->get_segment(ctxt, &sel, &desc, &base3, 1576 VCPU_SREG_LDTR)) 1577 return; 1578 1579 dt->size = desc_limit_scaled(&desc); /* what if limit > 65535? */ 1580 dt->address = get_desc_base(&desc) | ((u64)base3 << 32); 1581 } else 1582 ops->get_gdt(ctxt, dt); 1583 } 1584 1585 static int get_descriptor_ptr(struct x86_emulate_ctxt *ctxt, 1586 u16 selector, ulong *desc_addr_p) 1587 { 1588 struct desc_ptr dt; 1589 u16 index = selector >> 3; 1590 ulong addr; 1591 1592 get_descriptor_table_ptr(ctxt, selector, &dt); 1593 1594 if (dt.size < index * 8 + 7) 1595 return emulate_gp(ctxt, selector & 0xfffc); 1596 1597 addr = dt.address + index * 8; 1598 1599 #ifdef CONFIG_X86_64 1600 if (addr >> 32 != 0) { 1601 u64 efer = 0; 1602 1603 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 1604 if (!(efer & EFER_LMA)) 1605 addr &= (u32)-1; 1606 } 1607 #endif 1608 1609 *desc_addr_p = addr; 1610 return X86EMUL_CONTINUE; 1611 } 1612 1613 /* allowed just for 8 bytes segments */ 1614 static int read_segment_descriptor(struct x86_emulate_ctxt *ctxt, 1615 u16 selector, struct desc_struct *desc, 1616 ulong *desc_addr_p) 1617 { 1618 int rc; 1619 1620 rc = get_descriptor_ptr(ctxt, selector, desc_addr_p); 1621 if (rc != X86EMUL_CONTINUE) 1622 return rc; 1623 1624 return linear_read_system(ctxt, *desc_addr_p, desc, sizeof(*desc)); 1625 } 1626 1627 /* allowed just for 8 bytes segments */ 1628 static int write_segment_descriptor(struct x86_emulate_ctxt *ctxt, 1629 u16 selector, struct desc_struct *desc) 1630 { 1631 int rc; 1632 ulong addr; 1633 1634 rc = get_descriptor_ptr(ctxt, selector, &addr); 1635 if (rc != X86EMUL_CONTINUE) 1636 return rc; 1637 1638 return linear_write_system(ctxt, addr, desc, sizeof(*desc)); 1639 } 1640 1641 static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt, 1642 u16 selector, int seg, u8 cpl, 1643 enum x86_transfer_type transfer, 1644 struct desc_struct *desc) 1645 { 1646 struct desc_struct seg_desc, old_desc; 1647 u8 dpl, rpl; 1648 unsigned err_vec = GP_VECTOR; 1649 u32 err_code = 0; 1650 bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */ 1651 ulong desc_addr; 1652 int ret; 1653 u16 dummy; 1654 u32 base3 = 0; 1655 1656 memset(&seg_desc, 0, sizeof(seg_desc)); 1657 1658 if (ctxt->mode == X86EMUL_MODE_REAL) { 1659 /* set real mode segment descriptor (keep limit etc. for 1660 * unreal mode) */ 1661 ctxt->ops->get_segment(ctxt, &dummy, &seg_desc, NULL, seg); 1662 set_desc_base(&seg_desc, selector << 4); 1663 goto load; 1664 } else if (seg <= VCPU_SREG_GS && ctxt->mode == X86EMUL_MODE_VM86) { 1665 /* VM86 needs a clean new segment descriptor */ 1666 set_desc_base(&seg_desc, selector << 4); 1667 set_desc_limit(&seg_desc, 0xffff); 1668 seg_desc.type = 3; 1669 seg_desc.p = 1; 1670 seg_desc.s = 1; 1671 seg_desc.dpl = 3; 1672 goto load; 1673 } 1674 1675 rpl = selector & 3; 1676 1677 /* TR should be in GDT only */ 1678 if (seg == VCPU_SREG_TR && (selector & (1 << 2))) 1679 goto exception; 1680 1681 /* NULL selector is not valid for TR, CS and (except for long mode) SS */ 1682 if (null_selector) { 1683 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_TR) 1684 goto exception; 1685 1686 if (seg == VCPU_SREG_SS) { 1687 if (ctxt->mode != X86EMUL_MODE_PROT64 || rpl != cpl) 1688 goto exception; 1689 1690 /* 1691 * ctxt->ops->set_segment expects the CPL to be in 1692 * SS.DPL, so fake an expand-up 32-bit data segment. 1693 */ 1694 seg_desc.type = 3; 1695 seg_desc.p = 1; 1696 seg_desc.s = 1; 1697 seg_desc.dpl = cpl; 1698 seg_desc.d = 1; 1699 seg_desc.g = 1; 1700 } 1701 1702 /* Skip all following checks */ 1703 goto load; 1704 } 1705 1706 ret = read_segment_descriptor(ctxt, selector, &seg_desc, &desc_addr); 1707 if (ret != X86EMUL_CONTINUE) 1708 return ret; 1709 1710 err_code = selector & 0xfffc; 1711 err_vec = (transfer == X86_TRANSFER_TASK_SWITCH) ? TS_VECTOR : 1712 GP_VECTOR; 1713 1714 /* can't load system descriptor into segment selector */ 1715 if (seg <= VCPU_SREG_GS && !seg_desc.s) { 1716 if (transfer == X86_TRANSFER_CALL_JMP) 1717 return X86EMUL_UNHANDLEABLE; 1718 goto exception; 1719 } 1720 1721 if (!seg_desc.p) { 1722 err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR; 1723 goto exception; 1724 } 1725 1726 dpl = seg_desc.dpl; 1727 1728 switch (seg) { 1729 case VCPU_SREG_SS: 1730 /* 1731 * segment is not a writable data segment or segment 1732 * selector's RPL != CPL or segment selector's RPL != CPL 1733 */ 1734 if (rpl != cpl || (seg_desc.type & 0xa) != 0x2 || dpl != cpl) 1735 goto exception; 1736 break; 1737 case VCPU_SREG_CS: 1738 if (!(seg_desc.type & 8)) 1739 goto exception; 1740 1741 if (seg_desc.type & 4) { 1742 /* conforming */ 1743 if (dpl > cpl) 1744 goto exception; 1745 } else { 1746 /* nonconforming */ 1747 if (rpl > cpl || dpl != cpl) 1748 goto exception; 1749 } 1750 /* in long-mode d/b must be clear if l is set */ 1751 if (seg_desc.d && seg_desc.l) { 1752 u64 efer = 0; 1753 1754 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 1755 if (efer & EFER_LMA) 1756 goto exception; 1757 } 1758 1759 /* CS(RPL) <- CPL */ 1760 selector = (selector & 0xfffc) | cpl; 1761 break; 1762 case VCPU_SREG_TR: 1763 if (seg_desc.s || (seg_desc.type != 1 && seg_desc.type != 9)) 1764 goto exception; 1765 old_desc = seg_desc; 1766 seg_desc.type |= 2; /* busy */ 1767 ret = ctxt->ops->cmpxchg_emulated(ctxt, desc_addr, &old_desc, &seg_desc, 1768 sizeof(seg_desc), &ctxt->exception); 1769 if (ret != X86EMUL_CONTINUE) 1770 return ret; 1771 break; 1772 case VCPU_SREG_LDTR: 1773 if (seg_desc.s || seg_desc.type != 2) 1774 goto exception; 1775 break; 1776 default: /* DS, ES, FS, or GS */ 1777 /* 1778 * segment is not a data or readable code segment or 1779 * ((segment is a data or nonconforming code segment) 1780 * and (both RPL and CPL > DPL)) 1781 */ 1782 if ((seg_desc.type & 0xa) == 0x8 || 1783 (((seg_desc.type & 0xc) != 0xc) && 1784 (rpl > dpl && cpl > dpl))) 1785 goto exception; 1786 break; 1787 } 1788 1789 if (seg_desc.s) { 1790 /* mark segment as accessed */ 1791 if (!(seg_desc.type & 1)) { 1792 seg_desc.type |= 1; 1793 ret = write_segment_descriptor(ctxt, selector, 1794 &seg_desc); 1795 if (ret != X86EMUL_CONTINUE) 1796 return ret; 1797 } 1798 } else if (ctxt->mode == X86EMUL_MODE_PROT64) { 1799 ret = linear_read_system(ctxt, desc_addr+8, &base3, sizeof(base3)); 1800 if (ret != X86EMUL_CONTINUE) 1801 return ret; 1802 if (emul_is_noncanonical_address(get_desc_base(&seg_desc) | 1803 ((u64)base3 << 32), ctxt)) 1804 return emulate_gp(ctxt, 0); 1805 } 1806 load: 1807 ctxt->ops->set_segment(ctxt, selector, &seg_desc, base3, seg); 1808 if (desc) 1809 *desc = seg_desc; 1810 return X86EMUL_CONTINUE; 1811 exception: 1812 return emulate_exception(ctxt, err_vec, err_code, true); 1813 } 1814 1815 static int load_segment_descriptor(struct x86_emulate_ctxt *ctxt, 1816 u16 selector, int seg) 1817 { 1818 u8 cpl = ctxt->ops->cpl(ctxt); 1819 1820 /* 1821 * None of MOV, POP and LSS can load a NULL selector in CPL=3, but 1822 * they can load it at CPL<3 (Intel's manual says only LSS can, 1823 * but it's wrong). 1824 * 1825 * However, the Intel manual says that putting IST=1/DPL=3 in 1826 * an interrupt gate will result in SS=3 (the AMD manual instead 1827 * says it doesn't), so allow SS=3 in __load_segment_descriptor 1828 * and only forbid it here. 1829 */ 1830 if (seg == VCPU_SREG_SS && selector == 3 && 1831 ctxt->mode == X86EMUL_MODE_PROT64) 1832 return emulate_exception(ctxt, GP_VECTOR, 0, true); 1833 1834 return __load_segment_descriptor(ctxt, selector, seg, cpl, 1835 X86_TRANSFER_NONE, NULL); 1836 } 1837 1838 static void write_register_operand(struct operand *op) 1839 { 1840 return assign_register(op->addr.reg, op->val, op->bytes); 1841 } 1842 1843 static int writeback(struct x86_emulate_ctxt *ctxt, struct operand *op) 1844 { 1845 switch (op->type) { 1846 case OP_REG: 1847 write_register_operand(op); 1848 break; 1849 case OP_MEM: 1850 if (ctxt->lock_prefix) 1851 return segmented_cmpxchg(ctxt, 1852 op->addr.mem, 1853 &op->orig_val, 1854 &op->val, 1855 op->bytes); 1856 else 1857 return segmented_write(ctxt, 1858 op->addr.mem, 1859 &op->val, 1860 op->bytes); 1861 break; 1862 case OP_MEM_STR: 1863 return segmented_write(ctxt, 1864 op->addr.mem, 1865 op->data, 1866 op->bytes * op->count); 1867 break; 1868 case OP_XMM: 1869 write_sse_reg(&op->vec_val, op->addr.xmm); 1870 break; 1871 case OP_MM: 1872 write_mmx_reg(&op->mm_val, op->addr.mm); 1873 break; 1874 case OP_NONE: 1875 /* no writeback */ 1876 break; 1877 default: 1878 break; 1879 } 1880 return X86EMUL_CONTINUE; 1881 } 1882 1883 static int push(struct x86_emulate_ctxt *ctxt, void *data, int bytes) 1884 { 1885 struct segmented_address addr; 1886 1887 rsp_increment(ctxt, -bytes); 1888 addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt); 1889 addr.seg = VCPU_SREG_SS; 1890 1891 return segmented_write(ctxt, addr, data, bytes); 1892 } 1893 1894 static int em_push(struct x86_emulate_ctxt *ctxt) 1895 { 1896 /* Disable writeback. */ 1897 ctxt->dst.type = OP_NONE; 1898 return push(ctxt, &ctxt->src.val, ctxt->op_bytes); 1899 } 1900 1901 static int emulate_pop(struct x86_emulate_ctxt *ctxt, 1902 void *dest, int len) 1903 { 1904 int rc; 1905 struct segmented_address addr; 1906 1907 addr.ea = reg_read(ctxt, VCPU_REGS_RSP) & stack_mask(ctxt); 1908 addr.seg = VCPU_SREG_SS; 1909 rc = segmented_read(ctxt, addr, dest, len); 1910 if (rc != X86EMUL_CONTINUE) 1911 return rc; 1912 1913 rsp_increment(ctxt, len); 1914 return rc; 1915 } 1916 1917 static int em_pop(struct x86_emulate_ctxt *ctxt) 1918 { 1919 return emulate_pop(ctxt, &ctxt->dst.val, ctxt->op_bytes); 1920 } 1921 1922 static int emulate_popf(struct x86_emulate_ctxt *ctxt, 1923 void *dest, int len) 1924 { 1925 int rc; 1926 unsigned long val, change_mask; 1927 int iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT; 1928 int cpl = ctxt->ops->cpl(ctxt); 1929 1930 rc = emulate_pop(ctxt, &val, len); 1931 if (rc != X86EMUL_CONTINUE) 1932 return rc; 1933 1934 change_mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | 1935 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF | 1936 X86_EFLAGS_TF | X86_EFLAGS_DF | X86_EFLAGS_NT | 1937 X86_EFLAGS_AC | X86_EFLAGS_ID; 1938 1939 switch(ctxt->mode) { 1940 case X86EMUL_MODE_PROT64: 1941 case X86EMUL_MODE_PROT32: 1942 case X86EMUL_MODE_PROT16: 1943 if (cpl == 0) 1944 change_mask |= X86_EFLAGS_IOPL; 1945 if (cpl <= iopl) 1946 change_mask |= X86_EFLAGS_IF; 1947 break; 1948 case X86EMUL_MODE_VM86: 1949 if (iopl < 3) 1950 return emulate_gp(ctxt, 0); 1951 change_mask |= X86_EFLAGS_IF; 1952 break; 1953 default: /* real mode */ 1954 change_mask |= (X86_EFLAGS_IOPL | X86_EFLAGS_IF); 1955 break; 1956 } 1957 1958 *(unsigned long *)dest = 1959 (ctxt->eflags & ~change_mask) | (val & change_mask); 1960 1961 return rc; 1962 } 1963 1964 static int em_popf(struct x86_emulate_ctxt *ctxt) 1965 { 1966 ctxt->dst.type = OP_REG; 1967 ctxt->dst.addr.reg = &ctxt->eflags; 1968 ctxt->dst.bytes = ctxt->op_bytes; 1969 return emulate_popf(ctxt, &ctxt->dst.val, ctxt->op_bytes); 1970 } 1971 1972 static int em_enter(struct x86_emulate_ctxt *ctxt) 1973 { 1974 int rc; 1975 unsigned frame_size = ctxt->src.val; 1976 unsigned nesting_level = ctxt->src2.val & 31; 1977 ulong rbp; 1978 1979 if (nesting_level) 1980 return X86EMUL_UNHANDLEABLE; 1981 1982 rbp = reg_read(ctxt, VCPU_REGS_RBP); 1983 rc = push(ctxt, &rbp, stack_size(ctxt)); 1984 if (rc != X86EMUL_CONTINUE) 1985 return rc; 1986 assign_masked(reg_rmw(ctxt, VCPU_REGS_RBP), reg_read(ctxt, VCPU_REGS_RSP), 1987 stack_mask(ctxt)); 1988 assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), 1989 reg_read(ctxt, VCPU_REGS_RSP) - frame_size, 1990 stack_mask(ctxt)); 1991 return X86EMUL_CONTINUE; 1992 } 1993 1994 static int em_leave(struct x86_emulate_ctxt *ctxt) 1995 { 1996 assign_masked(reg_rmw(ctxt, VCPU_REGS_RSP), reg_read(ctxt, VCPU_REGS_RBP), 1997 stack_mask(ctxt)); 1998 return emulate_pop(ctxt, reg_rmw(ctxt, VCPU_REGS_RBP), ctxt->op_bytes); 1999 } 2000 2001 static int em_push_sreg(struct x86_emulate_ctxt *ctxt) 2002 { 2003 int seg = ctxt->src2.val; 2004 2005 ctxt->src.val = get_segment_selector(ctxt, seg); 2006 if (ctxt->op_bytes == 4) { 2007 rsp_increment(ctxt, -2); 2008 ctxt->op_bytes = 2; 2009 } 2010 2011 return em_push(ctxt); 2012 } 2013 2014 static int em_pop_sreg(struct x86_emulate_ctxt *ctxt) 2015 { 2016 int seg = ctxt->src2.val; 2017 unsigned long selector; 2018 int rc; 2019 2020 rc = emulate_pop(ctxt, &selector, 2); 2021 if (rc != X86EMUL_CONTINUE) 2022 return rc; 2023 2024 if (ctxt->modrm_reg == VCPU_SREG_SS) 2025 ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS; 2026 if (ctxt->op_bytes > 2) 2027 rsp_increment(ctxt, ctxt->op_bytes - 2); 2028 2029 rc = load_segment_descriptor(ctxt, (u16)selector, seg); 2030 return rc; 2031 } 2032 2033 static int em_pusha(struct x86_emulate_ctxt *ctxt) 2034 { 2035 unsigned long old_esp = reg_read(ctxt, VCPU_REGS_RSP); 2036 int rc = X86EMUL_CONTINUE; 2037 int reg = VCPU_REGS_RAX; 2038 2039 while (reg <= VCPU_REGS_RDI) { 2040 (reg == VCPU_REGS_RSP) ? 2041 (ctxt->src.val = old_esp) : (ctxt->src.val = reg_read(ctxt, reg)); 2042 2043 rc = em_push(ctxt); 2044 if (rc != X86EMUL_CONTINUE) 2045 return rc; 2046 2047 ++reg; 2048 } 2049 2050 return rc; 2051 } 2052 2053 static int em_pushf(struct x86_emulate_ctxt *ctxt) 2054 { 2055 ctxt->src.val = (unsigned long)ctxt->eflags & ~X86_EFLAGS_VM; 2056 return em_push(ctxt); 2057 } 2058 2059 static int em_popa(struct x86_emulate_ctxt *ctxt) 2060 { 2061 int rc = X86EMUL_CONTINUE; 2062 int reg = VCPU_REGS_RDI; 2063 u32 val; 2064 2065 while (reg >= VCPU_REGS_RAX) { 2066 if (reg == VCPU_REGS_RSP) { 2067 rsp_increment(ctxt, ctxt->op_bytes); 2068 --reg; 2069 } 2070 2071 rc = emulate_pop(ctxt, &val, ctxt->op_bytes); 2072 if (rc != X86EMUL_CONTINUE) 2073 break; 2074 assign_register(reg_rmw(ctxt, reg), val, ctxt->op_bytes); 2075 --reg; 2076 } 2077 return rc; 2078 } 2079 2080 static int __emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq) 2081 { 2082 const struct x86_emulate_ops *ops = ctxt->ops; 2083 int rc; 2084 struct desc_ptr dt; 2085 gva_t cs_addr; 2086 gva_t eip_addr; 2087 u16 cs, eip; 2088 2089 /* TODO: Add limit checks */ 2090 ctxt->src.val = ctxt->eflags; 2091 rc = em_push(ctxt); 2092 if (rc != X86EMUL_CONTINUE) 2093 return rc; 2094 2095 ctxt->eflags &= ~(X86_EFLAGS_IF | X86_EFLAGS_TF | X86_EFLAGS_AC); 2096 2097 ctxt->src.val = get_segment_selector(ctxt, VCPU_SREG_CS); 2098 rc = em_push(ctxt); 2099 if (rc != X86EMUL_CONTINUE) 2100 return rc; 2101 2102 ctxt->src.val = ctxt->_eip; 2103 rc = em_push(ctxt); 2104 if (rc != X86EMUL_CONTINUE) 2105 return rc; 2106 2107 ops->get_idt(ctxt, &dt); 2108 2109 eip_addr = dt.address + (irq << 2); 2110 cs_addr = dt.address + (irq << 2) + 2; 2111 2112 rc = linear_read_system(ctxt, cs_addr, &cs, 2); 2113 if (rc != X86EMUL_CONTINUE) 2114 return rc; 2115 2116 rc = linear_read_system(ctxt, eip_addr, &eip, 2); 2117 if (rc != X86EMUL_CONTINUE) 2118 return rc; 2119 2120 rc = load_segment_descriptor(ctxt, cs, VCPU_SREG_CS); 2121 if (rc != X86EMUL_CONTINUE) 2122 return rc; 2123 2124 ctxt->_eip = eip; 2125 2126 return rc; 2127 } 2128 2129 int emulate_int_real(struct x86_emulate_ctxt *ctxt, int irq) 2130 { 2131 int rc; 2132 2133 invalidate_registers(ctxt); 2134 rc = __emulate_int_real(ctxt, irq); 2135 if (rc == X86EMUL_CONTINUE) 2136 writeback_registers(ctxt); 2137 return rc; 2138 } 2139 2140 static int emulate_int(struct x86_emulate_ctxt *ctxt, int irq) 2141 { 2142 switch(ctxt->mode) { 2143 case X86EMUL_MODE_REAL: 2144 return __emulate_int_real(ctxt, irq); 2145 case X86EMUL_MODE_VM86: 2146 case X86EMUL_MODE_PROT16: 2147 case X86EMUL_MODE_PROT32: 2148 case X86EMUL_MODE_PROT64: 2149 default: 2150 /* Protected mode interrupts unimplemented yet */ 2151 return X86EMUL_UNHANDLEABLE; 2152 } 2153 } 2154 2155 static int emulate_iret_real(struct x86_emulate_ctxt *ctxt) 2156 { 2157 int rc = X86EMUL_CONTINUE; 2158 unsigned long temp_eip = 0; 2159 unsigned long temp_eflags = 0; 2160 unsigned long cs = 0; 2161 unsigned long mask = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | 2162 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_TF | 2163 X86_EFLAGS_IF | X86_EFLAGS_DF | X86_EFLAGS_OF | 2164 X86_EFLAGS_IOPL | X86_EFLAGS_NT | X86_EFLAGS_RF | 2165 X86_EFLAGS_AC | X86_EFLAGS_ID | 2166 X86_EFLAGS_FIXED; 2167 unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF | 2168 X86_EFLAGS_VIP; 2169 2170 /* TODO: Add stack limit check */ 2171 2172 rc = emulate_pop(ctxt, &temp_eip, ctxt->op_bytes); 2173 2174 if (rc != X86EMUL_CONTINUE) 2175 return rc; 2176 2177 if (temp_eip & ~0xffff) 2178 return emulate_gp(ctxt, 0); 2179 2180 rc = emulate_pop(ctxt, &cs, ctxt->op_bytes); 2181 2182 if (rc != X86EMUL_CONTINUE) 2183 return rc; 2184 2185 rc = emulate_pop(ctxt, &temp_eflags, ctxt->op_bytes); 2186 2187 if (rc != X86EMUL_CONTINUE) 2188 return rc; 2189 2190 rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS); 2191 2192 if (rc != X86EMUL_CONTINUE) 2193 return rc; 2194 2195 ctxt->_eip = temp_eip; 2196 2197 if (ctxt->op_bytes == 4) 2198 ctxt->eflags = ((temp_eflags & mask) | (ctxt->eflags & vm86_mask)); 2199 else if (ctxt->op_bytes == 2) { 2200 ctxt->eflags &= ~0xffff; 2201 ctxt->eflags |= temp_eflags; 2202 } 2203 2204 ctxt->eflags &= ~EFLG_RESERVED_ZEROS_MASK; /* Clear reserved zeros */ 2205 ctxt->eflags |= X86_EFLAGS_FIXED; 2206 ctxt->ops->set_nmi_mask(ctxt, false); 2207 2208 return rc; 2209 } 2210 2211 static int em_iret(struct x86_emulate_ctxt *ctxt) 2212 { 2213 switch(ctxt->mode) { 2214 case X86EMUL_MODE_REAL: 2215 return emulate_iret_real(ctxt); 2216 case X86EMUL_MODE_VM86: 2217 case X86EMUL_MODE_PROT16: 2218 case X86EMUL_MODE_PROT32: 2219 case X86EMUL_MODE_PROT64: 2220 default: 2221 /* iret from protected mode unimplemented yet */ 2222 return X86EMUL_UNHANDLEABLE; 2223 } 2224 } 2225 2226 static int em_jmp_far(struct x86_emulate_ctxt *ctxt) 2227 { 2228 int rc; 2229 unsigned short sel; 2230 struct desc_struct new_desc; 2231 u8 cpl = ctxt->ops->cpl(ctxt); 2232 2233 memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2); 2234 2235 rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl, 2236 X86_TRANSFER_CALL_JMP, 2237 &new_desc); 2238 if (rc != X86EMUL_CONTINUE) 2239 return rc; 2240 2241 rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc); 2242 /* Error handling is not implemented. */ 2243 if (rc != X86EMUL_CONTINUE) 2244 return X86EMUL_UNHANDLEABLE; 2245 2246 return rc; 2247 } 2248 2249 static int em_jmp_abs(struct x86_emulate_ctxt *ctxt) 2250 { 2251 return assign_eip_near(ctxt, ctxt->src.val); 2252 } 2253 2254 static int em_call_near_abs(struct x86_emulate_ctxt *ctxt) 2255 { 2256 int rc; 2257 long int old_eip; 2258 2259 old_eip = ctxt->_eip; 2260 rc = assign_eip_near(ctxt, ctxt->src.val); 2261 if (rc != X86EMUL_CONTINUE) 2262 return rc; 2263 ctxt->src.val = old_eip; 2264 rc = em_push(ctxt); 2265 return rc; 2266 } 2267 2268 static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt) 2269 { 2270 u64 old = ctxt->dst.orig_val64; 2271 2272 if (ctxt->dst.bytes == 16) 2273 return X86EMUL_UNHANDLEABLE; 2274 2275 if (((u32) (old >> 0) != (u32) reg_read(ctxt, VCPU_REGS_RAX)) || 2276 ((u32) (old >> 32) != (u32) reg_read(ctxt, VCPU_REGS_RDX))) { 2277 *reg_write(ctxt, VCPU_REGS_RAX) = (u32) (old >> 0); 2278 *reg_write(ctxt, VCPU_REGS_RDX) = (u32) (old >> 32); 2279 ctxt->eflags &= ~X86_EFLAGS_ZF; 2280 } else { 2281 ctxt->dst.val64 = ((u64)reg_read(ctxt, VCPU_REGS_RCX) << 32) | 2282 (u32) reg_read(ctxt, VCPU_REGS_RBX); 2283 2284 ctxt->eflags |= X86_EFLAGS_ZF; 2285 } 2286 return X86EMUL_CONTINUE; 2287 } 2288 2289 static int em_ret(struct x86_emulate_ctxt *ctxt) 2290 { 2291 int rc; 2292 unsigned long eip; 2293 2294 rc = emulate_pop(ctxt, &eip, ctxt->op_bytes); 2295 if (rc != X86EMUL_CONTINUE) 2296 return rc; 2297 2298 return assign_eip_near(ctxt, eip); 2299 } 2300 2301 static int em_ret_far(struct x86_emulate_ctxt *ctxt) 2302 { 2303 int rc; 2304 unsigned long eip, cs; 2305 int cpl = ctxt->ops->cpl(ctxt); 2306 struct desc_struct new_desc; 2307 2308 rc = emulate_pop(ctxt, &eip, ctxt->op_bytes); 2309 if (rc != X86EMUL_CONTINUE) 2310 return rc; 2311 rc = emulate_pop(ctxt, &cs, ctxt->op_bytes); 2312 if (rc != X86EMUL_CONTINUE) 2313 return rc; 2314 /* Outer-privilege level return is not implemented */ 2315 if (ctxt->mode >= X86EMUL_MODE_PROT16 && (cs & 3) > cpl) 2316 return X86EMUL_UNHANDLEABLE; 2317 rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl, 2318 X86_TRANSFER_RET, 2319 &new_desc); 2320 if (rc != X86EMUL_CONTINUE) 2321 return rc; 2322 rc = assign_eip_far(ctxt, eip, &new_desc); 2323 /* Error handling is not implemented. */ 2324 if (rc != X86EMUL_CONTINUE) 2325 return X86EMUL_UNHANDLEABLE; 2326 2327 return rc; 2328 } 2329 2330 static int em_ret_far_imm(struct x86_emulate_ctxt *ctxt) 2331 { 2332 int rc; 2333 2334 rc = em_ret_far(ctxt); 2335 if (rc != X86EMUL_CONTINUE) 2336 return rc; 2337 rsp_increment(ctxt, ctxt->src.val); 2338 return X86EMUL_CONTINUE; 2339 } 2340 2341 static int em_cmpxchg(struct x86_emulate_ctxt *ctxt) 2342 { 2343 /* Save real source value, then compare EAX against destination. */ 2344 ctxt->dst.orig_val = ctxt->dst.val; 2345 ctxt->dst.val = reg_read(ctxt, VCPU_REGS_RAX); 2346 ctxt->src.orig_val = ctxt->src.val; 2347 ctxt->src.val = ctxt->dst.orig_val; 2348 fastop(ctxt, em_cmp); 2349 2350 if (ctxt->eflags & X86_EFLAGS_ZF) { 2351 /* Success: write back to memory; no update of EAX */ 2352 ctxt->src.type = OP_NONE; 2353 ctxt->dst.val = ctxt->src.orig_val; 2354 } else { 2355 /* Failure: write the value we saw to EAX. */ 2356 ctxt->src.type = OP_REG; 2357 ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX); 2358 ctxt->src.val = ctxt->dst.orig_val; 2359 /* Create write-cycle to dest by writing the same value */ 2360 ctxt->dst.val = ctxt->dst.orig_val; 2361 } 2362 return X86EMUL_CONTINUE; 2363 } 2364 2365 static int em_lseg(struct x86_emulate_ctxt *ctxt) 2366 { 2367 int seg = ctxt->src2.val; 2368 unsigned short sel; 2369 int rc; 2370 2371 memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2); 2372 2373 rc = load_segment_descriptor(ctxt, sel, seg); 2374 if (rc != X86EMUL_CONTINUE) 2375 return rc; 2376 2377 ctxt->dst.val = ctxt->src.val; 2378 return rc; 2379 } 2380 2381 static int emulator_has_longmode(struct x86_emulate_ctxt *ctxt) 2382 { 2383 #ifdef CONFIG_X86_64 2384 return ctxt->ops->guest_has_long_mode(ctxt); 2385 #else 2386 return false; 2387 #endif 2388 } 2389 2390 static void rsm_set_desc_flags(struct desc_struct *desc, u32 flags) 2391 { 2392 desc->g = (flags >> 23) & 1; 2393 desc->d = (flags >> 22) & 1; 2394 desc->l = (flags >> 21) & 1; 2395 desc->avl = (flags >> 20) & 1; 2396 desc->p = (flags >> 15) & 1; 2397 desc->dpl = (flags >> 13) & 3; 2398 desc->s = (flags >> 12) & 1; 2399 desc->type = (flags >> 8) & 15; 2400 } 2401 2402 static int rsm_load_seg_32(struct x86_emulate_ctxt *ctxt, const char *smstate, 2403 int n) 2404 { 2405 struct desc_struct desc; 2406 int offset; 2407 u16 selector; 2408 2409 selector = GET_SMSTATE(u32, smstate, 0x7fa8 + n * 4); 2410 2411 if (n < 3) 2412 offset = 0x7f84 + n * 12; 2413 else 2414 offset = 0x7f2c + (n - 3) * 12; 2415 2416 set_desc_base(&desc, GET_SMSTATE(u32, smstate, offset + 8)); 2417 set_desc_limit(&desc, GET_SMSTATE(u32, smstate, offset + 4)); 2418 rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, offset)); 2419 ctxt->ops->set_segment(ctxt, selector, &desc, 0, n); 2420 return X86EMUL_CONTINUE; 2421 } 2422 2423 #ifdef CONFIG_X86_64 2424 static int rsm_load_seg_64(struct x86_emulate_ctxt *ctxt, const char *smstate, 2425 int n) 2426 { 2427 struct desc_struct desc; 2428 int offset; 2429 u16 selector; 2430 u32 base3; 2431 2432 offset = 0x7e00 + n * 16; 2433 2434 selector = GET_SMSTATE(u16, smstate, offset); 2435 rsm_set_desc_flags(&desc, GET_SMSTATE(u16, smstate, offset + 2) << 8); 2436 set_desc_limit(&desc, GET_SMSTATE(u32, smstate, offset + 4)); 2437 set_desc_base(&desc, GET_SMSTATE(u32, smstate, offset + 8)); 2438 base3 = GET_SMSTATE(u32, smstate, offset + 12); 2439 2440 ctxt->ops->set_segment(ctxt, selector, &desc, base3, n); 2441 return X86EMUL_CONTINUE; 2442 } 2443 #endif 2444 2445 static int rsm_enter_protected_mode(struct x86_emulate_ctxt *ctxt, 2446 u64 cr0, u64 cr3, u64 cr4) 2447 { 2448 int bad; 2449 u64 pcid; 2450 2451 /* In order to later set CR4.PCIDE, CR3[11:0] must be zero. */ 2452 pcid = 0; 2453 if (cr4 & X86_CR4_PCIDE) { 2454 pcid = cr3 & 0xfff; 2455 cr3 &= ~0xfff; 2456 } 2457 2458 bad = ctxt->ops->set_cr(ctxt, 3, cr3); 2459 if (bad) 2460 return X86EMUL_UNHANDLEABLE; 2461 2462 /* 2463 * First enable PAE, long mode needs it before CR0.PG = 1 is set. 2464 * Then enable protected mode. However, PCID cannot be enabled 2465 * if EFER.LMA=0, so set it separately. 2466 */ 2467 bad = ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE); 2468 if (bad) 2469 return X86EMUL_UNHANDLEABLE; 2470 2471 bad = ctxt->ops->set_cr(ctxt, 0, cr0); 2472 if (bad) 2473 return X86EMUL_UNHANDLEABLE; 2474 2475 if (cr4 & X86_CR4_PCIDE) { 2476 bad = ctxt->ops->set_cr(ctxt, 4, cr4); 2477 if (bad) 2478 return X86EMUL_UNHANDLEABLE; 2479 if (pcid) { 2480 bad = ctxt->ops->set_cr(ctxt, 3, cr3 | pcid); 2481 if (bad) 2482 return X86EMUL_UNHANDLEABLE; 2483 } 2484 2485 } 2486 2487 return X86EMUL_CONTINUE; 2488 } 2489 2490 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt, 2491 const char *smstate) 2492 { 2493 struct desc_struct desc; 2494 struct desc_ptr dt; 2495 u16 selector; 2496 u32 val, cr0, cr3, cr4; 2497 int i; 2498 2499 cr0 = GET_SMSTATE(u32, smstate, 0x7ffc); 2500 cr3 = GET_SMSTATE(u32, smstate, 0x7ff8); 2501 ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7ff4) | X86_EFLAGS_FIXED; 2502 ctxt->_eip = GET_SMSTATE(u32, smstate, 0x7ff0); 2503 2504 for (i = 0; i < 8; i++) 2505 *reg_write(ctxt, i) = GET_SMSTATE(u32, smstate, 0x7fd0 + i * 4); 2506 2507 val = GET_SMSTATE(u32, smstate, 0x7fcc); 2508 2509 if (ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1)) 2510 return X86EMUL_UNHANDLEABLE; 2511 2512 val = GET_SMSTATE(u32, smstate, 0x7fc8); 2513 2514 if (ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1)) 2515 return X86EMUL_UNHANDLEABLE; 2516 2517 selector = GET_SMSTATE(u32, smstate, 0x7fc4); 2518 set_desc_base(&desc, GET_SMSTATE(u32, smstate, 0x7f64)); 2519 set_desc_limit(&desc, GET_SMSTATE(u32, smstate, 0x7f60)); 2520 rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, 0x7f5c)); 2521 ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_TR); 2522 2523 selector = GET_SMSTATE(u32, smstate, 0x7fc0); 2524 set_desc_base(&desc, GET_SMSTATE(u32, smstate, 0x7f80)); 2525 set_desc_limit(&desc, GET_SMSTATE(u32, smstate, 0x7f7c)); 2526 rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, 0x7f78)); 2527 ctxt->ops->set_segment(ctxt, selector, &desc, 0, VCPU_SREG_LDTR); 2528 2529 dt.address = GET_SMSTATE(u32, smstate, 0x7f74); 2530 dt.size = GET_SMSTATE(u32, smstate, 0x7f70); 2531 ctxt->ops->set_gdt(ctxt, &dt); 2532 2533 dt.address = GET_SMSTATE(u32, smstate, 0x7f58); 2534 dt.size = GET_SMSTATE(u32, smstate, 0x7f54); 2535 ctxt->ops->set_idt(ctxt, &dt); 2536 2537 for (i = 0; i < 6; i++) { 2538 int r = rsm_load_seg_32(ctxt, smstate, i); 2539 if (r != X86EMUL_CONTINUE) 2540 return r; 2541 } 2542 2543 cr4 = GET_SMSTATE(u32, smstate, 0x7f14); 2544 2545 ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7ef8)); 2546 2547 return rsm_enter_protected_mode(ctxt, cr0, cr3, cr4); 2548 } 2549 2550 #ifdef CONFIG_X86_64 2551 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt, 2552 const char *smstate) 2553 { 2554 struct desc_struct desc; 2555 struct desc_ptr dt; 2556 u64 val, cr0, cr3, cr4; 2557 u32 base3; 2558 u16 selector; 2559 int i, r; 2560 2561 for (i = 0; i < 16; i++) 2562 *reg_write(ctxt, i) = GET_SMSTATE(u64, smstate, 0x7ff8 - i * 8); 2563 2564 ctxt->_eip = GET_SMSTATE(u64, smstate, 0x7f78); 2565 ctxt->eflags = GET_SMSTATE(u32, smstate, 0x7f70) | X86_EFLAGS_FIXED; 2566 2567 val = GET_SMSTATE(u32, smstate, 0x7f68); 2568 2569 if (ctxt->ops->set_dr(ctxt, 6, (val & DR6_VOLATILE) | DR6_FIXED_1)) 2570 return X86EMUL_UNHANDLEABLE; 2571 2572 val = GET_SMSTATE(u32, smstate, 0x7f60); 2573 2574 if (ctxt->ops->set_dr(ctxt, 7, (val & DR7_VOLATILE) | DR7_FIXED_1)) 2575 return X86EMUL_UNHANDLEABLE; 2576 2577 cr0 = GET_SMSTATE(u64, smstate, 0x7f58); 2578 cr3 = GET_SMSTATE(u64, smstate, 0x7f50); 2579 cr4 = GET_SMSTATE(u64, smstate, 0x7f48); 2580 ctxt->ops->set_smbase(ctxt, GET_SMSTATE(u32, smstate, 0x7f00)); 2581 val = GET_SMSTATE(u64, smstate, 0x7ed0); 2582 2583 if (ctxt->ops->set_msr(ctxt, MSR_EFER, val & ~EFER_LMA)) 2584 return X86EMUL_UNHANDLEABLE; 2585 2586 selector = GET_SMSTATE(u32, smstate, 0x7e90); 2587 rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, 0x7e92) << 8); 2588 set_desc_limit(&desc, GET_SMSTATE(u32, smstate, 0x7e94)); 2589 set_desc_base(&desc, GET_SMSTATE(u32, smstate, 0x7e98)); 2590 base3 = GET_SMSTATE(u32, smstate, 0x7e9c); 2591 ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_TR); 2592 2593 dt.size = GET_SMSTATE(u32, smstate, 0x7e84); 2594 dt.address = GET_SMSTATE(u64, smstate, 0x7e88); 2595 ctxt->ops->set_idt(ctxt, &dt); 2596 2597 selector = GET_SMSTATE(u32, smstate, 0x7e70); 2598 rsm_set_desc_flags(&desc, GET_SMSTATE(u32, smstate, 0x7e72) << 8); 2599 set_desc_limit(&desc, GET_SMSTATE(u32, smstate, 0x7e74)); 2600 set_desc_base(&desc, GET_SMSTATE(u32, smstate, 0x7e78)); 2601 base3 = GET_SMSTATE(u32, smstate, 0x7e7c); 2602 ctxt->ops->set_segment(ctxt, selector, &desc, base3, VCPU_SREG_LDTR); 2603 2604 dt.size = GET_SMSTATE(u32, smstate, 0x7e64); 2605 dt.address = GET_SMSTATE(u64, smstate, 0x7e68); 2606 ctxt->ops->set_gdt(ctxt, &dt); 2607 2608 r = rsm_enter_protected_mode(ctxt, cr0, cr3, cr4); 2609 if (r != X86EMUL_CONTINUE) 2610 return r; 2611 2612 for (i = 0; i < 6; i++) { 2613 r = rsm_load_seg_64(ctxt, smstate, i); 2614 if (r != X86EMUL_CONTINUE) 2615 return r; 2616 } 2617 2618 return X86EMUL_CONTINUE; 2619 } 2620 #endif 2621 2622 static int em_rsm(struct x86_emulate_ctxt *ctxt) 2623 { 2624 unsigned long cr0, cr4, efer; 2625 char buf[512]; 2626 u64 smbase; 2627 int ret; 2628 2629 if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_MASK) == 0) 2630 return emulate_ud(ctxt); 2631 2632 smbase = ctxt->ops->get_smbase(ctxt); 2633 2634 ret = ctxt->ops->read_phys(ctxt, smbase + 0xfe00, buf, sizeof(buf)); 2635 if (ret != X86EMUL_CONTINUE) 2636 return X86EMUL_UNHANDLEABLE; 2637 2638 if ((ctxt->ops->get_hflags(ctxt) & X86EMUL_SMM_INSIDE_NMI_MASK) == 0) 2639 ctxt->ops->set_nmi_mask(ctxt, false); 2640 2641 ctxt->ops->set_hflags(ctxt, ctxt->ops->get_hflags(ctxt) & 2642 ~(X86EMUL_SMM_INSIDE_NMI_MASK | X86EMUL_SMM_MASK)); 2643 2644 /* 2645 * Get back to real mode, to prepare a safe state in which to load 2646 * CR0/CR3/CR4/EFER. It's all a bit more complicated if the vCPU 2647 * supports long mode. 2648 */ 2649 if (emulator_has_longmode(ctxt)) { 2650 struct desc_struct cs_desc; 2651 2652 /* Zero CR4.PCIDE before CR0.PG. */ 2653 cr4 = ctxt->ops->get_cr(ctxt, 4); 2654 if (cr4 & X86_CR4_PCIDE) 2655 ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE); 2656 2657 /* A 32-bit code segment is required to clear EFER.LMA. */ 2658 memset(&cs_desc, 0, sizeof(cs_desc)); 2659 cs_desc.type = 0xb; 2660 cs_desc.s = cs_desc.g = cs_desc.p = 1; 2661 ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS); 2662 } 2663 2664 /* For the 64-bit case, this will clear EFER.LMA. */ 2665 cr0 = ctxt->ops->get_cr(ctxt, 0); 2666 if (cr0 & X86_CR0_PE) 2667 ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE)); 2668 2669 if (emulator_has_longmode(ctxt)) { 2670 /* Clear CR4.PAE before clearing EFER.LME. */ 2671 cr4 = ctxt->ops->get_cr(ctxt, 4); 2672 if (cr4 & X86_CR4_PAE) 2673 ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE); 2674 2675 /* And finally go back to 32-bit mode. */ 2676 efer = 0; 2677 ctxt->ops->set_msr(ctxt, MSR_EFER, efer); 2678 } 2679 2680 /* 2681 * Give pre_leave_smm() a chance to make ISA-specific changes to the 2682 * vCPU state (e.g. enter guest mode) before loading state from the SMM 2683 * state-save area. 2684 */ 2685 if (ctxt->ops->pre_leave_smm(ctxt, buf)) 2686 return X86EMUL_UNHANDLEABLE; 2687 2688 #ifdef CONFIG_X86_64 2689 if (emulator_has_longmode(ctxt)) 2690 ret = rsm_load_state_64(ctxt, buf); 2691 else 2692 #endif 2693 ret = rsm_load_state_32(ctxt, buf); 2694 2695 if (ret != X86EMUL_CONTINUE) { 2696 /* FIXME: should triple fault */ 2697 return X86EMUL_UNHANDLEABLE; 2698 } 2699 2700 ctxt->ops->post_leave_smm(ctxt); 2701 2702 return X86EMUL_CONTINUE; 2703 } 2704 2705 static void 2706 setup_syscalls_segments(struct x86_emulate_ctxt *ctxt, 2707 struct desc_struct *cs, struct desc_struct *ss) 2708 { 2709 cs->l = 0; /* will be adjusted later */ 2710 set_desc_base(cs, 0); /* flat segment */ 2711 cs->g = 1; /* 4kb granularity */ 2712 set_desc_limit(cs, 0xfffff); /* 4GB limit */ 2713 cs->type = 0x0b; /* Read, Execute, Accessed */ 2714 cs->s = 1; 2715 cs->dpl = 0; /* will be adjusted later */ 2716 cs->p = 1; 2717 cs->d = 1; 2718 cs->avl = 0; 2719 2720 set_desc_base(ss, 0); /* flat segment */ 2721 set_desc_limit(ss, 0xfffff); /* 4GB limit */ 2722 ss->g = 1; /* 4kb granularity */ 2723 ss->s = 1; 2724 ss->type = 0x03; /* Read/Write, Accessed */ 2725 ss->d = 1; /* 32bit stack segment */ 2726 ss->dpl = 0; 2727 ss->p = 1; 2728 ss->l = 0; 2729 ss->avl = 0; 2730 } 2731 2732 static bool vendor_intel(struct x86_emulate_ctxt *ctxt) 2733 { 2734 u32 eax, ebx, ecx, edx; 2735 2736 eax = ecx = 0; 2737 ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true); 2738 return is_guest_vendor_intel(ebx, ecx, edx); 2739 } 2740 2741 static bool em_syscall_is_enabled(struct x86_emulate_ctxt *ctxt) 2742 { 2743 const struct x86_emulate_ops *ops = ctxt->ops; 2744 u32 eax, ebx, ecx, edx; 2745 2746 /* 2747 * syscall should always be enabled in longmode - so only become 2748 * vendor specific (cpuid) if other modes are active... 2749 */ 2750 if (ctxt->mode == X86EMUL_MODE_PROT64) 2751 return true; 2752 2753 eax = 0x00000000; 2754 ecx = 0x00000000; 2755 ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, true); 2756 /* 2757 * remark: Intel CPUs only support "syscall" in 64bit longmode. Also a 2758 * 64bit guest with a 32bit compat-app running will #UD !! While this 2759 * behaviour can be fixed (by emulating) into AMD response - CPUs of 2760 * AMD can't behave like Intel. 2761 */ 2762 if (is_guest_vendor_intel(ebx, ecx, edx)) 2763 return false; 2764 2765 if (is_guest_vendor_amd(ebx, ecx, edx) || 2766 is_guest_vendor_hygon(ebx, ecx, edx)) 2767 return true; 2768 2769 /* 2770 * default: (not Intel, not AMD, not Hygon), apply Intel's 2771 * stricter rules... 2772 */ 2773 return false; 2774 } 2775 2776 static int em_syscall(struct x86_emulate_ctxt *ctxt) 2777 { 2778 const struct x86_emulate_ops *ops = ctxt->ops; 2779 struct desc_struct cs, ss; 2780 u64 msr_data; 2781 u16 cs_sel, ss_sel; 2782 u64 efer = 0; 2783 2784 /* syscall is not available in real mode */ 2785 if (ctxt->mode == X86EMUL_MODE_REAL || 2786 ctxt->mode == X86EMUL_MODE_VM86) 2787 return emulate_ud(ctxt); 2788 2789 if (!(em_syscall_is_enabled(ctxt))) 2790 return emulate_ud(ctxt); 2791 2792 ops->get_msr(ctxt, MSR_EFER, &efer); 2793 if (!(efer & EFER_SCE)) 2794 return emulate_ud(ctxt); 2795 2796 setup_syscalls_segments(ctxt, &cs, &ss); 2797 ops->get_msr(ctxt, MSR_STAR, &msr_data); 2798 msr_data >>= 32; 2799 cs_sel = (u16)(msr_data & 0xfffc); 2800 ss_sel = (u16)(msr_data + 8); 2801 2802 if (efer & EFER_LMA) { 2803 cs.d = 0; 2804 cs.l = 1; 2805 } 2806 ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS); 2807 ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS); 2808 2809 *reg_write(ctxt, VCPU_REGS_RCX) = ctxt->_eip; 2810 if (efer & EFER_LMA) { 2811 #ifdef CONFIG_X86_64 2812 *reg_write(ctxt, VCPU_REGS_R11) = ctxt->eflags; 2813 2814 ops->get_msr(ctxt, 2815 ctxt->mode == X86EMUL_MODE_PROT64 ? 2816 MSR_LSTAR : MSR_CSTAR, &msr_data); 2817 ctxt->_eip = msr_data; 2818 2819 ops->get_msr(ctxt, MSR_SYSCALL_MASK, &msr_data); 2820 ctxt->eflags &= ~msr_data; 2821 ctxt->eflags |= X86_EFLAGS_FIXED; 2822 #endif 2823 } else { 2824 /* legacy mode */ 2825 ops->get_msr(ctxt, MSR_STAR, &msr_data); 2826 ctxt->_eip = (u32)msr_data; 2827 2828 ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF); 2829 } 2830 2831 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0; 2832 return X86EMUL_CONTINUE; 2833 } 2834 2835 static int em_sysenter(struct x86_emulate_ctxt *ctxt) 2836 { 2837 const struct x86_emulate_ops *ops = ctxt->ops; 2838 struct desc_struct cs, ss; 2839 u64 msr_data; 2840 u16 cs_sel, ss_sel; 2841 u64 efer = 0; 2842 2843 ops->get_msr(ctxt, MSR_EFER, &efer); 2844 /* inject #GP if in real mode */ 2845 if (ctxt->mode == X86EMUL_MODE_REAL) 2846 return emulate_gp(ctxt, 0); 2847 2848 /* 2849 * Not recognized on AMD in compat mode (but is recognized in legacy 2850 * mode). 2851 */ 2852 if ((ctxt->mode != X86EMUL_MODE_PROT64) && (efer & EFER_LMA) 2853 && !vendor_intel(ctxt)) 2854 return emulate_ud(ctxt); 2855 2856 /* sysenter/sysexit have not been tested in 64bit mode. */ 2857 if (ctxt->mode == X86EMUL_MODE_PROT64) 2858 return X86EMUL_UNHANDLEABLE; 2859 2860 ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data); 2861 if ((msr_data & 0xfffc) == 0x0) 2862 return emulate_gp(ctxt, 0); 2863 2864 setup_syscalls_segments(ctxt, &cs, &ss); 2865 ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF); 2866 cs_sel = (u16)msr_data & ~SEGMENT_RPL_MASK; 2867 ss_sel = cs_sel + 8; 2868 if (efer & EFER_LMA) { 2869 cs.d = 0; 2870 cs.l = 1; 2871 } 2872 2873 ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS); 2874 ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS); 2875 2876 ops->get_msr(ctxt, MSR_IA32_SYSENTER_EIP, &msr_data); 2877 ctxt->_eip = (efer & EFER_LMA) ? msr_data : (u32)msr_data; 2878 2879 ops->get_msr(ctxt, MSR_IA32_SYSENTER_ESP, &msr_data); 2880 *reg_write(ctxt, VCPU_REGS_RSP) = (efer & EFER_LMA) ? msr_data : 2881 (u32)msr_data; 2882 2883 return X86EMUL_CONTINUE; 2884 } 2885 2886 static int em_sysexit(struct x86_emulate_ctxt *ctxt) 2887 { 2888 const struct x86_emulate_ops *ops = ctxt->ops; 2889 struct desc_struct cs, ss; 2890 u64 msr_data, rcx, rdx; 2891 int usermode; 2892 u16 cs_sel = 0, ss_sel = 0; 2893 2894 /* inject #GP if in real mode or Virtual 8086 mode */ 2895 if (ctxt->mode == X86EMUL_MODE_REAL || 2896 ctxt->mode == X86EMUL_MODE_VM86) 2897 return emulate_gp(ctxt, 0); 2898 2899 setup_syscalls_segments(ctxt, &cs, &ss); 2900 2901 if ((ctxt->rex_prefix & 0x8) != 0x0) 2902 usermode = X86EMUL_MODE_PROT64; 2903 else 2904 usermode = X86EMUL_MODE_PROT32; 2905 2906 rcx = reg_read(ctxt, VCPU_REGS_RCX); 2907 rdx = reg_read(ctxt, VCPU_REGS_RDX); 2908 2909 cs.dpl = 3; 2910 ss.dpl = 3; 2911 ops->get_msr(ctxt, MSR_IA32_SYSENTER_CS, &msr_data); 2912 switch (usermode) { 2913 case X86EMUL_MODE_PROT32: 2914 cs_sel = (u16)(msr_data + 16); 2915 if ((msr_data & 0xfffc) == 0x0) 2916 return emulate_gp(ctxt, 0); 2917 ss_sel = (u16)(msr_data + 24); 2918 rcx = (u32)rcx; 2919 rdx = (u32)rdx; 2920 break; 2921 case X86EMUL_MODE_PROT64: 2922 cs_sel = (u16)(msr_data + 32); 2923 if (msr_data == 0x0) 2924 return emulate_gp(ctxt, 0); 2925 ss_sel = cs_sel + 8; 2926 cs.d = 0; 2927 cs.l = 1; 2928 if (emul_is_noncanonical_address(rcx, ctxt) || 2929 emul_is_noncanonical_address(rdx, ctxt)) 2930 return emulate_gp(ctxt, 0); 2931 break; 2932 } 2933 cs_sel |= SEGMENT_RPL_MASK; 2934 ss_sel |= SEGMENT_RPL_MASK; 2935 2936 ops->set_segment(ctxt, cs_sel, &cs, 0, VCPU_SREG_CS); 2937 ops->set_segment(ctxt, ss_sel, &ss, 0, VCPU_SREG_SS); 2938 2939 ctxt->_eip = rdx; 2940 *reg_write(ctxt, VCPU_REGS_RSP) = rcx; 2941 2942 return X86EMUL_CONTINUE; 2943 } 2944 2945 static bool emulator_bad_iopl(struct x86_emulate_ctxt *ctxt) 2946 { 2947 int iopl; 2948 if (ctxt->mode == X86EMUL_MODE_REAL) 2949 return false; 2950 if (ctxt->mode == X86EMUL_MODE_VM86) 2951 return true; 2952 iopl = (ctxt->eflags & X86_EFLAGS_IOPL) >> X86_EFLAGS_IOPL_BIT; 2953 return ctxt->ops->cpl(ctxt) > iopl; 2954 } 2955 2956 #define VMWARE_PORT_VMPORT (0x5658) 2957 #define VMWARE_PORT_VMRPC (0x5659) 2958 2959 static bool emulator_io_port_access_allowed(struct x86_emulate_ctxt *ctxt, 2960 u16 port, u16 len) 2961 { 2962 const struct x86_emulate_ops *ops = ctxt->ops; 2963 struct desc_struct tr_seg; 2964 u32 base3; 2965 int r; 2966 u16 tr, io_bitmap_ptr, perm, bit_idx = port & 0x7; 2967 unsigned mask = (1 << len) - 1; 2968 unsigned long base; 2969 2970 /* 2971 * VMware allows access to these ports even if denied 2972 * by TSS I/O permission bitmap. Mimic behavior. 2973 */ 2974 if (enable_vmware_backdoor && 2975 ((port == VMWARE_PORT_VMPORT) || (port == VMWARE_PORT_VMRPC))) 2976 return true; 2977 2978 ops->get_segment(ctxt, &tr, &tr_seg, &base3, VCPU_SREG_TR); 2979 if (!tr_seg.p) 2980 return false; 2981 if (desc_limit_scaled(&tr_seg) < 103) 2982 return false; 2983 base = get_desc_base(&tr_seg); 2984 #ifdef CONFIG_X86_64 2985 base |= ((u64)base3) << 32; 2986 #endif 2987 r = ops->read_std(ctxt, base + 102, &io_bitmap_ptr, 2, NULL, true); 2988 if (r != X86EMUL_CONTINUE) 2989 return false; 2990 if (io_bitmap_ptr + port/8 > desc_limit_scaled(&tr_seg)) 2991 return false; 2992 r = ops->read_std(ctxt, base + io_bitmap_ptr + port/8, &perm, 2, NULL, true); 2993 if (r != X86EMUL_CONTINUE) 2994 return false; 2995 if ((perm >> bit_idx) & mask) 2996 return false; 2997 return true; 2998 } 2999 3000 static bool emulator_io_permited(struct x86_emulate_ctxt *ctxt, 3001 u16 port, u16 len) 3002 { 3003 if (ctxt->perm_ok) 3004 return true; 3005 3006 if (emulator_bad_iopl(ctxt)) 3007 if (!emulator_io_port_access_allowed(ctxt, port, len)) 3008 return false; 3009 3010 ctxt->perm_ok = true; 3011 3012 return true; 3013 } 3014 3015 static void string_registers_quirk(struct x86_emulate_ctxt *ctxt) 3016 { 3017 /* 3018 * Intel CPUs mask the counter and pointers in quite strange 3019 * manner when ECX is zero due to REP-string optimizations. 3020 */ 3021 #ifdef CONFIG_X86_64 3022 if (ctxt->ad_bytes != 4 || !vendor_intel(ctxt)) 3023 return; 3024 3025 *reg_write(ctxt, VCPU_REGS_RCX) = 0; 3026 3027 switch (ctxt->b) { 3028 case 0xa4: /* movsb */ 3029 case 0xa5: /* movsd/w */ 3030 *reg_rmw(ctxt, VCPU_REGS_RSI) &= (u32)-1; 3031 fallthrough; 3032 case 0xaa: /* stosb */ 3033 case 0xab: /* stosd/w */ 3034 *reg_rmw(ctxt, VCPU_REGS_RDI) &= (u32)-1; 3035 } 3036 #endif 3037 } 3038 3039 static void save_state_to_tss16(struct x86_emulate_ctxt *ctxt, 3040 struct tss_segment_16 *tss) 3041 { 3042 tss->ip = ctxt->_eip; 3043 tss->flag = ctxt->eflags; 3044 tss->ax = reg_read(ctxt, VCPU_REGS_RAX); 3045 tss->cx = reg_read(ctxt, VCPU_REGS_RCX); 3046 tss->dx = reg_read(ctxt, VCPU_REGS_RDX); 3047 tss->bx = reg_read(ctxt, VCPU_REGS_RBX); 3048 tss->sp = reg_read(ctxt, VCPU_REGS_RSP); 3049 tss->bp = reg_read(ctxt, VCPU_REGS_RBP); 3050 tss->si = reg_read(ctxt, VCPU_REGS_RSI); 3051 tss->di = reg_read(ctxt, VCPU_REGS_RDI); 3052 3053 tss->es = get_segment_selector(ctxt, VCPU_SREG_ES); 3054 tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS); 3055 tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS); 3056 tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS); 3057 tss->ldt = get_segment_selector(ctxt, VCPU_SREG_LDTR); 3058 } 3059 3060 static int load_state_from_tss16(struct x86_emulate_ctxt *ctxt, 3061 struct tss_segment_16 *tss) 3062 { 3063 int ret; 3064 u8 cpl; 3065 3066 ctxt->_eip = tss->ip; 3067 ctxt->eflags = tss->flag | 2; 3068 *reg_write(ctxt, VCPU_REGS_RAX) = tss->ax; 3069 *reg_write(ctxt, VCPU_REGS_RCX) = tss->cx; 3070 *reg_write(ctxt, VCPU_REGS_RDX) = tss->dx; 3071 *reg_write(ctxt, VCPU_REGS_RBX) = tss->bx; 3072 *reg_write(ctxt, VCPU_REGS_RSP) = tss->sp; 3073 *reg_write(ctxt, VCPU_REGS_RBP) = tss->bp; 3074 *reg_write(ctxt, VCPU_REGS_RSI) = tss->si; 3075 *reg_write(ctxt, VCPU_REGS_RDI) = tss->di; 3076 3077 /* 3078 * SDM says that segment selectors are loaded before segment 3079 * descriptors 3080 */ 3081 set_segment_selector(ctxt, tss->ldt, VCPU_SREG_LDTR); 3082 set_segment_selector(ctxt, tss->es, VCPU_SREG_ES); 3083 set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS); 3084 set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS); 3085 set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS); 3086 3087 cpl = tss->cs & 3; 3088 3089 /* 3090 * Now load segment descriptors. If fault happens at this stage 3091 * it is handled in a context of new task 3092 */ 3093 ret = __load_segment_descriptor(ctxt, tss->ldt, VCPU_SREG_LDTR, cpl, 3094 X86_TRANSFER_TASK_SWITCH, NULL); 3095 if (ret != X86EMUL_CONTINUE) 3096 return ret; 3097 ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl, 3098 X86_TRANSFER_TASK_SWITCH, NULL); 3099 if (ret != X86EMUL_CONTINUE) 3100 return ret; 3101 ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl, 3102 X86_TRANSFER_TASK_SWITCH, NULL); 3103 if (ret != X86EMUL_CONTINUE) 3104 return ret; 3105 ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl, 3106 X86_TRANSFER_TASK_SWITCH, NULL); 3107 if (ret != X86EMUL_CONTINUE) 3108 return ret; 3109 ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl, 3110 X86_TRANSFER_TASK_SWITCH, NULL); 3111 if (ret != X86EMUL_CONTINUE) 3112 return ret; 3113 3114 return X86EMUL_CONTINUE; 3115 } 3116 3117 static int task_switch_16(struct x86_emulate_ctxt *ctxt, 3118 u16 tss_selector, u16 old_tss_sel, 3119 ulong old_tss_base, struct desc_struct *new_desc) 3120 { 3121 struct tss_segment_16 tss_seg; 3122 int ret; 3123 u32 new_tss_base = get_desc_base(new_desc); 3124 3125 ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg)); 3126 if (ret != X86EMUL_CONTINUE) 3127 return ret; 3128 3129 save_state_to_tss16(ctxt, &tss_seg); 3130 3131 ret = linear_write_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg)); 3132 if (ret != X86EMUL_CONTINUE) 3133 return ret; 3134 3135 ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg)); 3136 if (ret != X86EMUL_CONTINUE) 3137 return ret; 3138 3139 if (old_tss_sel != 0xffff) { 3140 tss_seg.prev_task_link = old_tss_sel; 3141 3142 ret = linear_write_system(ctxt, new_tss_base, 3143 &tss_seg.prev_task_link, 3144 sizeof(tss_seg.prev_task_link)); 3145 if (ret != X86EMUL_CONTINUE) 3146 return ret; 3147 } 3148 3149 return load_state_from_tss16(ctxt, &tss_seg); 3150 } 3151 3152 static void save_state_to_tss32(struct x86_emulate_ctxt *ctxt, 3153 struct tss_segment_32 *tss) 3154 { 3155 /* CR3 and ldt selector are not saved intentionally */ 3156 tss->eip = ctxt->_eip; 3157 tss->eflags = ctxt->eflags; 3158 tss->eax = reg_read(ctxt, VCPU_REGS_RAX); 3159 tss->ecx = reg_read(ctxt, VCPU_REGS_RCX); 3160 tss->edx = reg_read(ctxt, VCPU_REGS_RDX); 3161 tss->ebx = reg_read(ctxt, VCPU_REGS_RBX); 3162 tss->esp = reg_read(ctxt, VCPU_REGS_RSP); 3163 tss->ebp = reg_read(ctxt, VCPU_REGS_RBP); 3164 tss->esi = reg_read(ctxt, VCPU_REGS_RSI); 3165 tss->edi = reg_read(ctxt, VCPU_REGS_RDI); 3166 3167 tss->es = get_segment_selector(ctxt, VCPU_SREG_ES); 3168 tss->cs = get_segment_selector(ctxt, VCPU_SREG_CS); 3169 tss->ss = get_segment_selector(ctxt, VCPU_SREG_SS); 3170 tss->ds = get_segment_selector(ctxt, VCPU_SREG_DS); 3171 tss->fs = get_segment_selector(ctxt, VCPU_SREG_FS); 3172 tss->gs = get_segment_selector(ctxt, VCPU_SREG_GS); 3173 } 3174 3175 static int load_state_from_tss32(struct x86_emulate_ctxt *ctxt, 3176 struct tss_segment_32 *tss) 3177 { 3178 int ret; 3179 u8 cpl; 3180 3181 if (ctxt->ops->set_cr(ctxt, 3, tss->cr3)) 3182 return emulate_gp(ctxt, 0); 3183 ctxt->_eip = tss->eip; 3184 ctxt->eflags = tss->eflags | 2; 3185 3186 /* General purpose registers */ 3187 *reg_write(ctxt, VCPU_REGS_RAX) = tss->eax; 3188 *reg_write(ctxt, VCPU_REGS_RCX) = tss->ecx; 3189 *reg_write(ctxt, VCPU_REGS_RDX) = tss->edx; 3190 *reg_write(ctxt, VCPU_REGS_RBX) = tss->ebx; 3191 *reg_write(ctxt, VCPU_REGS_RSP) = tss->esp; 3192 *reg_write(ctxt, VCPU_REGS_RBP) = tss->ebp; 3193 *reg_write(ctxt, VCPU_REGS_RSI) = tss->esi; 3194 *reg_write(ctxt, VCPU_REGS_RDI) = tss->edi; 3195 3196 /* 3197 * SDM says that segment selectors are loaded before segment 3198 * descriptors. This is important because CPL checks will 3199 * use CS.RPL. 3200 */ 3201 set_segment_selector(ctxt, tss->ldt_selector, VCPU_SREG_LDTR); 3202 set_segment_selector(ctxt, tss->es, VCPU_SREG_ES); 3203 set_segment_selector(ctxt, tss->cs, VCPU_SREG_CS); 3204 set_segment_selector(ctxt, tss->ss, VCPU_SREG_SS); 3205 set_segment_selector(ctxt, tss->ds, VCPU_SREG_DS); 3206 set_segment_selector(ctxt, tss->fs, VCPU_SREG_FS); 3207 set_segment_selector(ctxt, tss->gs, VCPU_SREG_GS); 3208 3209 /* 3210 * If we're switching between Protected Mode and VM86, we need to make 3211 * sure to update the mode before loading the segment descriptors so 3212 * that the selectors are interpreted correctly. 3213 */ 3214 if (ctxt->eflags & X86_EFLAGS_VM) { 3215 ctxt->mode = X86EMUL_MODE_VM86; 3216 cpl = 3; 3217 } else { 3218 ctxt->mode = X86EMUL_MODE_PROT32; 3219 cpl = tss->cs & 3; 3220 } 3221 3222 /* 3223 * Now load segment descriptors. If fault happenes at this stage 3224 * it is handled in a context of new task 3225 */ 3226 ret = __load_segment_descriptor(ctxt, tss->ldt_selector, VCPU_SREG_LDTR, 3227 cpl, X86_TRANSFER_TASK_SWITCH, NULL); 3228 if (ret != X86EMUL_CONTINUE) 3229 return ret; 3230 ret = __load_segment_descriptor(ctxt, tss->es, VCPU_SREG_ES, cpl, 3231 X86_TRANSFER_TASK_SWITCH, NULL); 3232 if (ret != X86EMUL_CONTINUE) 3233 return ret; 3234 ret = __load_segment_descriptor(ctxt, tss->cs, VCPU_SREG_CS, cpl, 3235 X86_TRANSFER_TASK_SWITCH, NULL); 3236 if (ret != X86EMUL_CONTINUE) 3237 return ret; 3238 ret = __load_segment_descriptor(ctxt, tss->ss, VCPU_SREG_SS, cpl, 3239 X86_TRANSFER_TASK_SWITCH, NULL); 3240 if (ret != X86EMUL_CONTINUE) 3241 return ret; 3242 ret = __load_segment_descriptor(ctxt, tss->ds, VCPU_SREG_DS, cpl, 3243 X86_TRANSFER_TASK_SWITCH, NULL); 3244 if (ret != X86EMUL_CONTINUE) 3245 return ret; 3246 ret = __load_segment_descriptor(ctxt, tss->fs, VCPU_SREG_FS, cpl, 3247 X86_TRANSFER_TASK_SWITCH, NULL); 3248 if (ret != X86EMUL_CONTINUE) 3249 return ret; 3250 ret = __load_segment_descriptor(ctxt, tss->gs, VCPU_SREG_GS, cpl, 3251 X86_TRANSFER_TASK_SWITCH, NULL); 3252 3253 return ret; 3254 } 3255 3256 static int task_switch_32(struct x86_emulate_ctxt *ctxt, 3257 u16 tss_selector, u16 old_tss_sel, 3258 ulong old_tss_base, struct desc_struct *new_desc) 3259 { 3260 struct tss_segment_32 tss_seg; 3261 int ret; 3262 u32 new_tss_base = get_desc_base(new_desc); 3263 u32 eip_offset = offsetof(struct tss_segment_32, eip); 3264 u32 ldt_sel_offset = offsetof(struct tss_segment_32, ldt_selector); 3265 3266 ret = linear_read_system(ctxt, old_tss_base, &tss_seg, sizeof(tss_seg)); 3267 if (ret != X86EMUL_CONTINUE) 3268 return ret; 3269 3270 save_state_to_tss32(ctxt, &tss_seg); 3271 3272 /* Only GP registers and segment selectors are saved */ 3273 ret = linear_write_system(ctxt, old_tss_base + eip_offset, &tss_seg.eip, 3274 ldt_sel_offset - eip_offset); 3275 if (ret != X86EMUL_CONTINUE) 3276 return ret; 3277 3278 ret = linear_read_system(ctxt, new_tss_base, &tss_seg, sizeof(tss_seg)); 3279 if (ret != X86EMUL_CONTINUE) 3280 return ret; 3281 3282 if (old_tss_sel != 0xffff) { 3283 tss_seg.prev_task_link = old_tss_sel; 3284 3285 ret = linear_write_system(ctxt, new_tss_base, 3286 &tss_seg.prev_task_link, 3287 sizeof(tss_seg.prev_task_link)); 3288 if (ret != X86EMUL_CONTINUE) 3289 return ret; 3290 } 3291 3292 return load_state_from_tss32(ctxt, &tss_seg); 3293 } 3294 3295 static int emulator_do_task_switch(struct x86_emulate_ctxt *ctxt, 3296 u16 tss_selector, int idt_index, int reason, 3297 bool has_error_code, u32 error_code) 3298 { 3299 const struct x86_emulate_ops *ops = ctxt->ops; 3300 struct desc_struct curr_tss_desc, next_tss_desc; 3301 int ret; 3302 u16 old_tss_sel = get_segment_selector(ctxt, VCPU_SREG_TR); 3303 ulong old_tss_base = 3304 ops->get_cached_segment_base(ctxt, VCPU_SREG_TR); 3305 u32 desc_limit; 3306 ulong desc_addr, dr7; 3307 3308 /* FIXME: old_tss_base == ~0 ? */ 3309 3310 ret = read_segment_descriptor(ctxt, tss_selector, &next_tss_desc, &desc_addr); 3311 if (ret != X86EMUL_CONTINUE) 3312 return ret; 3313 ret = read_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc, &desc_addr); 3314 if (ret != X86EMUL_CONTINUE) 3315 return ret; 3316 3317 /* FIXME: check that next_tss_desc is tss */ 3318 3319 /* 3320 * Check privileges. The three cases are task switch caused by... 3321 * 3322 * 1. jmp/call/int to task gate: Check against DPL of the task gate 3323 * 2. Exception/IRQ/iret: No check is performed 3324 * 3. jmp/call to TSS/task-gate: No check is performed since the 3325 * hardware checks it before exiting. 3326 */ 3327 if (reason == TASK_SWITCH_GATE) { 3328 if (idt_index != -1) { 3329 /* Software interrupts */ 3330 struct desc_struct task_gate_desc; 3331 int dpl; 3332 3333 ret = read_interrupt_descriptor(ctxt, idt_index, 3334 &task_gate_desc); 3335 if (ret != X86EMUL_CONTINUE) 3336 return ret; 3337 3338 dpl = task_gate_desc.dpl; 3339 if ((tss_selector & 3) > dpl || ops->cpl(ctxt) > dpl) 3340 return emulate_gp(ctxt, (idt_index << 3) | 0x2); 3341 } 3342 } 3343 3344 desc_limit = desc_limit_scaled(&next_tss_desc); 3345 if (!next_tss_desc.p || 3346 ((desc_limit < 0x67 && (next_tss_desc.type & 8)) || 3347 desc_limit < 0x2b)) { 3348 return emulate_ts(ctxt, tss_selector & 0xfffc); 3349 } 3350 3351 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) { 3352 curr_tss_desc.type &= ~(1 << 1); /* clear busy flag */ 3353 write_segment_descriptor(ctxt, old_tss_sel, &curr_tss_desc); 3354 } 3355 3356 if (reason == TASK_SWITCH_IRET) 3357 ctxt->eflags = ctxt->eflags & ~X86_EFLAGS_NT; 3358 3359 /* set back link to prev task only if NT bit is set in eflags 3360 note that old_tss_sel is not used after this point */ 3361 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE) 3362 old_tss_sel = 0xffff; 3363 3364 if (next_tss_desc.type & 8) 3365 ret = task_switch_32(ctxt, tss_selector, old_tss_sel, 3366 old_tss_base, &next_tss_desc); 3367 else 3368 ret = task_switch_16(ctxt, tss_selector, old_tss_sel, 3369 old_tss_base, &next_tss_desc); 3370 if (ret != X86EMUL_CONTINUE) 3371 return ret; 3372 3373 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) 3374 ctxt->eflags = ctxt->eflags | X86_EFLAGS_NT; 3375 3376 if (reason != TASK_SWITCH_IRET) { 3377 next_tss_desc.type |= (1 << 1); /* set busy flag */ 3378 write_segment_descriptor(ctxt, tss_selector, &next_tss_desc); 3379 } 3380 3381 ops->set_cr(ctxt, 0, ops->get_cr(ctxt, 0) | X86_CR0_TS); 3382 ops->set_segment(ctxt, tss_selector, &next_tss_desc, 0, VCPU_SREG_TR); 3383 3384 if (has_error_code) { 3385 ctxt->op_bytes = ctxt->ad_bytes = (next_tss_desc.type & 8) ? 4 : 2; 3386 ctxt->lock_prefix = 0; 3387 ctxt->src.val = (unsigned long) error_code; 3388 ret = em_push(ctxt); 3389 } 3390 3391 ops->get_dr(ctxt, 7, &dr7); 3392 ops->set_dr(ctxt, 7, dr7 & ~(DR_LOCAL_ENABLE_MASK | DR_LOCAL_SLOWDOWN)); 3393 3394 return ret; 3395 } 3396 3397 int emulator_task_switch(struct x86_emulate_ctxt *ctxt, 3398 u16 tss_selector, int idt_index, int reason, 3399 bool has_error_code, u32 error_code) 3400 { 3401 int rc; 3402 3403 invalidate_registers(ctxt); 3404 ctxt->_eip = ctxt->eip; 3405 ctxt->dst.type = OP_NONE; 3406 3407 rc = emulator_do_task_switch(ctxt, tss_selector, idt_index, reason, 3408 has_error_code, error_code); 3409 3410 if (rc == X86EMUL_CONTINUE) { 3411 ctxt->eip = ctxt->_eip; 3412 writeback_registers(ctxt); 3413 } 3414 3415 return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK; 3416 } 3417 3418 static void string_addr_inc(struct x86_emulate_ctxt *ctxt, int reg, 3419 struct operand *op) 3420 { 3421 int df = (ctxt->eflags & X86_EFLAGS_DF) ? -op->count : op->count; 3422 3423 register_address_increment(ctxt, reg, df * op->bytes); 3424 op->addr.mem.ea = register_address(ctxt, reg); 3425 } 3426 3427 static int em_das(struct x86_emulate_ctxt *ctxt) 3428 { 3429 u8 al, old_al; 3430 bool af, cf, old_cf; 3431 3432 cf = ctxt->eflags & X86_EFLAGS_CF; 3433 al = ctxt->dst.val; 3434 3435 old_al = al; 3436 old_cf = cf; 3437 cf = false; 3438 af = ctxt->eflags & X86_EFLAGS_AF; 3439 if ((al & 0x0f) > 9 || af) { 3440 al -= 6; 3441 cf = old_cf | (al >= 250); 3442 af = true; 3443 } else { 3444 af = false; 3445 } 3446 if (old_al > 0x99 || old_cf) { 3447 al -= 0x60; 3448 cf = true; 3449 } 3450 3451 ctxt->dst.val = al; 3452 /* Set PF, ZF, SF */ 3453 ctxt->src.type = OP_IMM; 3454 ctxt->src.val = 0; 3455 ctxt->src.bytes = 1; 3456 fastop(ctxt, em_or); 3457 ctxt->eflags &= ~(X86_EFLAGS_AF | X86_EFLAGS_CF); 3458 if (cf) 3459 ctxt->eflags |= X86_EFLAGS_CF; 3460 if (af) 3461 ctxt->eflags |= X86_EFLAGS_AF; 3462 return X86EMUL_CONTINUE; 3463 } 3464 3465 static int em_aam(struct x86_emulate_ctxt *ctxt) 3466 { 3467 u8 al, ah; 3468 3469 if (ctxt->src.val == 0) 3470 return emulate_de(ctxt); 3471 3472 al = ctxt->dst.val & 0xff; 3473 ah = al / ctxt->src.val; 3474 al %= ctxt->src.val; 3475 3476 ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al | (ah << 8); 3477 3478 /* Set PF, ZF, SF */ 3479 ctxt->src.type = OP_IMM; 3480 ctxt->src.val = 0; 3481 ctxt->src.bytes = 1; 3482 fastop(ctxt, em_or); 3483 3484 return X86EMUL_CONTINUE; 3485 } 3486 3487 static int em_aad(struct x86_emulate_ctxt *ctxt) 3488 { 3489 u8 al = ctxt->dst.val & 0xff; 3490 u8 ah = (ctxt->dst.val >> 8) & 0xff; 3491 3492 al = (al + (ah * ctxt->src.val)) & 0xff; 3493 3494 ctxt->dst.val = (ctxt->dst.val & 0xffff0000) | al; 3495 3496 /* Set PF, ZF, SF */ 3497 ctxt->src.type = OP_IMM; 3498 ctxt->src.val = 0; 3499 ctxt->src.bytes = 1; 3500 fastop(ctxt, em_or); 3501 3502 return X86EMUL_CONTINUE; 3503 } 3504 3505 static int em_call(struct x86_emulate_ctxt *ctxt) 3506 { 3507 int rc; 3508 long rel = ctxt->src.val; 3509 3510 ctxt->src.val = (unsigned long)ctxt->_eip; 3511 rc = jmp_rel(ctxt, rel); 3512 if (rc != X86EMUL_CONTINUE) 3513 return rc; 3514 return em_push(ctxt); 3515 } 3516 3517 static int em_call_far(struct x86_emulate_ctxt *ctxt) 3518 { 3519 u16 sel, old_cs; 3520 ulong old_eip; 3521 int rc; 3522 struct desc_struct old_desc, new_desc; 3523 const struct x86_emulate_ops *ops = ctxt->ops; 3524 int cpl = ctxt->ops->cpl(ctxt); 3525 enum x86emul_mode prev_mode = ctxt->mode; 3526 3527 old_eip = ctxt->_eip; 3528 ops->get_segment(ctxt, &old_cs, &old_desc, NULL, VCPU_SREG_CS); 3529 3530 memcpy(&sel, ctxt->src.valptr + ctxt->op_bytes, 2); 3531 rc = __load_segment_descriptor(ctxt, sel, VCPU_SREG_CS, cpl, 3532 X86_TRANSFER_CALL_JMP, &new_desc); 3533 if (rc != X86EMUL_CONTINUE) 3534 return rc; 3535 3536 rc = assign_eip_far(ctxt, ctxt->src.val, &new_desc); 3537 if (rc != X86EMUL_CONTINUE) 3538 goto fail; 3539 3540 ctxt->src.val = old_cs; 3541 rc = em_push(ctxt); 3542 if (rc != X86EMUL_CONTINUE) 3543 goto fail; 3544 3545 ctxt->src.val = old_eip; 3546 rc = em_push(ctxt); 3547 /* If we failed, we tainted the memory, but the very least we should 3548 restore cs */ 3549 if (rc != X86EMUL_CONTINUE) { 3550 pr_warn_once("faulting far call emulation tainted memory\n"); 3551 goto fail; 3552 } 3553 return rc; 3554 fail: 3555 ops->set_segment(ctxt, old_cs, &old_desc, 0, VCPU_SREG_CS); 3556 ctxt->mode = prev_mode; 3557 return rc; 3558 3559 } 3560 3561 static int em_ret_near_imm(struct x86_emulate_ctxt *ctxt) 3562 { 3563 int rc; 3564 unsigned long eip; 3565 3566 rc = emulate_pop(ctxt, &eip, ctxt->op_bytes); 3567 if (rc != X86EMUL_CONTINUE) 3568 return rc; 3569 rc = assign_eip_near(ctxt, eip); 3570 if (rc != X86EMUL_CONTINUE) 3571 return rc; 3572 rsp_increment(ctxt, ctxt->src.val); 3573 return X86EMUL_CONTINUE; 3574 } 3575 3576 static int em_xchg(struct x86_emulate_ctxt *ctxt) 3577 { 3578 /* Write back the register source. */ 3579 ctxt->src.val = ctxt->dst.val; 3580 write_register_operand(&ctxt->src); 3581 3582 /* Write back the memory destination with implicit LOCK prefix. */ 3583 ctxt->dst.val = ctxt->src.orig_val; 3584 ctxt->lock_prefix = 1; 3585 return X86EMUL_CONTINUE; 3586 } 3587 3588 static int em_imul_3op(struct x86_emulate_ctxt *ctxt) 3589 { 3590 ctxt->dst.val = ctxt->src2.val; 3591 return fastop(ctxt, em_imul); 3592 } 3593 3594 static int em_cwd(struct x86_emulate_ctxt *ctxt) 3595 { 3596 ctxt->dst.type = OP_REG; 3597 ctxt->dst.bytes = ctxt->src.bytes; 3598 ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX); 3599 ctxt->dst.val = ~((ctxt->src.val >> (ctxt->src.bytes * 8 - 1)) - 1); 3600 3601 return X86EMUL_CONTINUE; 3602 } 3603 3604 static int em_rdpid(struct x86_emulate_ctxt *ctxt) 3605 { 3606 u64 tsc_aux = 0; 3607 3608 if (ctxt->ops->get_msr(ctxt, MSR_TSC_AUX, &tsc_aux)) 3609 return emulate_ud(ctxt); 3610 ctxt->dst.val = tsc_aux; 3611 return X86EMUL_CONTINUE; 3612 } 3613 3614 static int em_rdtsc(struct x86_emulate_ctxt *ctxt) 3615 { 3616 u64 tsc = 0; 3617 3618 ctxt->ops->get_msr(ctxt, MSR_IA32_TSC, &tsc); 3619 *reg_write(ctxt, VCPU_REGS_RAX) = (u32)tsc; 3620 *reg_write(ctxt, VCPU_REGS_RDX) = tsc >> 32; 3621 return X86EMUL_CONTINUE; 3622 } 3623 3624 static int em_rdpmc(struct x86_emulate_ctxt *ctxt) 3625 { 3626 u64 pmc; 3627 3628 if (ctxt->ops->read_pmc(ctxt, reg_read(ctxt, VCPU_REGS_RCX), &pmc)) 3629 return emulate_gp(ctxt, 0); 3630 *reg_write(ctxt, VCPU_REGS_RAX) = (u32)pmc; 3631 *reg_write(ctxt, VCPU_REGS_RDX) = pmc >> 32; 3632 return X86EMUL_CONTINUE; 3633 } 3634 3635 static int em_mov(struct x86_emulate_ctxt *ctxt) 3636 { 3637 memcpy(ctxt->dst.valptr, ctxt->src.valptr, sizeof(ctxt->src.valptr)); 3638 return X86EMUL_CONTINUE; 3639 } 3640 3641 static int em_movbe(struct x86_emulate_ctxt *ctxt) 3642 { 3643 u16 tmp; 3644 3645 if (!ctxt->ops->guest_has_movbe(ctxt)) 3646 return emulate_ud(ctxt); 3647 3648 switch (ctxt->op_bytes) { 3649 case 2: 3650 /* 3651 * From MOVBE definition: "...When the operand size is 16 bits, 3652 * the upper word of the destination register remains unchanged 3653 * ..." 3654 * 3655 * Both casting ->valptr and ->val to u16 breaks strict aliasing 3656 * rules so we have to do the operation almost per hand. 3657 */ 3658 tmp = (u16)ctxt->src.val; 3659 ctxt->dst.val &= ~0xffffUL; 3660 ctxt->dst.val |= (unsigned long)swab16(tmp); 3661 break; 3662 case 4: 3663 ctxt->dst.val = swab32((u32)ctxt->src.val); 3664 break; 3665 case 8: 3666 ctxt->dst.val = swab64(ctxt->src.val); 3667 break; 3668 default: 3669 BUG(); 3670 } 3671 return X86EMUL_CONTINUE; 3672 } 3673 3674 static int em_cr_write(struct x86_emulate_ctxt *ctxt) 3675 { 3676 if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val)) 3677 return emulate_gp(ctxt, 0); 3678 3679 /* Disable writeback. */ 3680 ctxt->dst.type = OP_NONE; 3681 return X86EMUL_CONTINUE; 3682 } 3683 3684 static int em_dr_write(struct x86_emulate_ctxt *ctxt) 3685 { 3686 unsigned long val; 3687 3688 if (ctxt->mode == X86EMUL_MODE_PROT64) 3689 val = ctxt->src.val & ~0ULL; 3690 else 3691 val = ctxt->src.val & ~0U; 3692 3693 /* #UD condition is already handled. */ 3694 if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0) 3695 return emulate_gp(ctxt, 0); 3696 3697 /* Disable writeback. */ 3698 ctxt->dst.type = OP_NONE; 3699 return X86EMUL_CONTINUE; 3700 } 3701 3702 static int em_wrmsr(struct x86_emulate_ctxt *ctxt) 3703 { 3704 u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX); 3705 u64 msr_data; 3706 int r; 3707 3708 msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX) 3709 | ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32); 3710 r = ctxt->ops->set_msr(ctxt, msr_index, msr_data); 3711 3712 if (r == X86EMUL_IO_NEEDED) 3713 return r; 3714 3715 if (r > 0) 3716 return emulate_gp(ctxt, 0); 3717 3718 return r < 0 ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE; 3719 } 3720 3721 static int em_rdmsr(struct x86_emulate_ctxt *ctxt) 3722 { 3723 u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX); 3724 u64 msr_data; 3725 int r; 3726 3727 r = ctxt->ops->get_msr(ctxt, msr_index, &msr_data); 3728 3729 if (r == X86EMUL_IO_NEEDED) 3730 return r; 3731 3732 if (r) 3733 return emulate_gp(ctxt, 0); 3734 3735 *reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data; 3736 *reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32; 3737 return X86EMUL_CONTINUE; 3738 } 3739 3740 static int em_store_sreg(struct x86_emulate_ctxt *ctxt, int segment) 3741 { 3742 if (segment > VCPU_SREG_GS && 3743 (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) && 3744 ctxt->ops->cpl(ctxt) > 0) 3745 return emulate_gp(ctxt, 0); 3746 3747 ctxt->dst.val = get_segment_selector(ctxt, segment); 3748 if (ctxt->dst.bytes == 4 && ctxt->dst.type == OP_MEM) 3749 ctxt->dst.bytes = 2; 3750 return X86EMUL_CONTINUE; 3751 } 3752 3753 static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt) 3754 { 3755 if (ctxt->modrm_reg > VCPU_SREG_GS) 3756 return emulate_ud(ctxt); 3757 3758 return em_store_sreg(ctxt, ctxt->modrm_reg); 3759 } 3760 3761 static int em_mov_sreg_rm(struct x86_emulate_ctxt *ctxt) 3762 { 3763 u16 sel = ctxt->src.val; 3764 3765 if (ctxt->modrm_reg == VCPU_SREG_CS || ctxt->modrm_reg > VCPU_SREG_GS) 3766 return emulate_ud(ctxt); 3767 3768 if (ctxt->modrm_reg == VCPU_SREG_SS) 3769 ctxt->interruptibility = KVM_X86_SHADOW_INT_MOV_SS; 3770 3771 /* Disable writeback. */ 3772 ctxt->dst.type = OP_NONE; 3773 return load_segment_descriptor(ctxt, sel, ctxt->modrm_reg); 3774 } 3775 3776 static int em_sldt(struct x86_emulate_ctxt *ctxt) 3777 { 3778 return em_store_sreg(ctxt, VCPU_SREG_LDTR); 3779 } 3780 3781 static int em_lldt(struct x86_emulate_ctxt *ctxt) 3782 { 3783 u16 sel = ctxt->src.val; 3784 3785 /* Disable writeback. */ 3786 ctxt->dst.type = OP_NONE; 3787 return load_segment_descriptor(ctxt, sel, VCPU_SREG_LDTR); 3788 } 3789 3790 static int em_str(struct x86_emulate_ctxt *ctxt) 3791 { 3792 return em_store_sreg(ctxt, VCPU_SREG_TR); 3793 } 3794 3795 static int em_ltr(struct x86_emulate_ctxt *ctxt) 3796 { 3797 u16 sel = ctxt->src.val; 3798 3799 /* Disable writeback. */ 3800 ctxt->dst.type = OP_NONE; 3801 return load_segment_descriptor(ctxt, sel, VCPU_SREG_TR); 3802 } 3803 3804 static int em_invlpg(struct x86_emulate_ctxt *ctxt) 3805 { 3806 int rc; 3807 ulong linear; 3808 3809 rc = linearize(ctxt, ctxt->src.addr.mem, 1, false, &linear); 3810 if (rc == X86EMUL_CONTINUE) 3811 ctxt->ops->invlpg(ctxt, linear); 3812 /* Disable writeback. */ 3813 ctxt->dst.type = OP_NONE; 3814 return X86EMUL_CONTINUE; 3815 } 3816 3817 static int em_clts(struct x86_emulate_ctxt *ctxt) 3818 { 3819 ulong cr0; 3820 3821 cr0 = ctxt->ops->get_cr(ctxt, 0); 3822 cr0 &= ~X86_CR0_TS; 3823 ctxt->ops->set_cr(ctxt, 0, cr0); 3824 return X86EMUL_CONTINUE; 3825 } 3826 3827 static int em_hypercall(struct x86_emulate_ctxt *ctxt) 3828 { 3829 int rc = ctxt->ops->fix_hypercall(ctxt); 3830 3831 if (rc != X86EMUL_CONTINUE) 3832 return rc; 3833 3834 /* Let the processor re-execute the fixed hypercall */ 3835 ctxt->_eip = ctxt->eip; 3836 /* Disable writeback. */ 3837 ctxt->dst.type = OP_NONE; 3838 return X86EMUL_CONTINUE; 3839 } 3840 3841 static int emulate_store_desc_ptr(struct x86_emulate_ctxt *ctxt, 3842 void (*get)(struct x86_emulate_ctxt *ctxt, 3843 struct desc_ptr *ptr)) 3844 { 3845 struct desc_ptr desc_ptr; 3846 3847 if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) && 3848 ctxt->ops->cpl(ctxt) > 0) 3849 return emulate_gp(ctxt, 0); 3850 3851 if (ctxt->mode == X86EMUL_MODE_PROT64) 3852 ctxt->op_bytes = 8; 3853 get(ctxt, &desc_ptr); 3854 if (ctxt->op_bytes == 2) { 3855 ctxt->op_bytes = 4; 3856 desc_ptr.address &= 0x00ffffff; 3857 } 3858 /* Disable writeback. */ 3859 ctxt->dst.type = OP_NONE; 3860 return segmented_write_std(ctxt, ctxt->dst.addr.mem, 3861 &desc_ptr, 2 + ctxt->op_bytes); 3862 } 3863 3864 static int em_sgdt(struct x86_emulate_ctxt *ctxt) 3865 { 3866 return emulate_store_desc_ptr(ctxt, ctxt->ops->get_gdt); 3867 } 3868 3869 static int em_sidt(struct x86_emulate_ctxt *ctxt) 3870 { 3871 return emulate_store_desc_ptr(ctxt, ctxt->ops->get_idt); 3872 } 3873 3874 static int em_lgdt_lidt(struct x86_emulate_ctxt *ctxt, bool lgdt) 3875 { 3876 struct desc_ptr desc_ptr; 3877 int rc; 3878 3879 if (ctxt->mode == X86EMUL_MODE_PROT64) 3880 ctxt->op_bytes = 8; 3881 rc = read_descriptor(ctxt, ctxt->src.addr.mem, 3882 &desc_ptr.size, &desc_ptr.address, 3883 ctxt->op_bytes); 3884 if (rc != X86EMUL_CONTINUE) 3885 return rc; 3886 if (ctxt->mode == X86EMUL_MODE_PROT64 && 3887 emul_is_noncanonical_address(desc_ptr.address, ctxt)) 3888 return emulate_gp(ctxt, 0); 3889 if (lgdt) 3890 ctxt->ops->set_gdt(ctxt, &desc_ptr); 3891 else 3892 ctxt->ops->set_idt(ctxt, &desc_ptr); 3893 /* Disable writeback. */ 3894 ctxt->dst.type = OP_NONE; 3895 return X86EMUL_CONTINUE; 3896 } 3897 3898 static int em_lgdt(struct x86_emulate_ctxt *ctxt) 3899 { 3900 return em_lgdt_lidt(ctxt, true); 3901 } 3902 3903 static int em_lidt(struct x86_emulate_ctxt *ctxt) 3904 { 3905 return em_lgdt_lidt(ctxt, false); 3906 } 3907 3908 static int em_smsw(struct x86_emulate_ctxt *ctxt) 3909 { 3910 if ((ctxt->ops->get_cr(ctxt, 4) & X86_CR4_UMIP) && 3911 ctxt->ops->cpl(ctxt) > 0) 3912 return emulate_gp(ctxt, 0); 3913 3914 if (ctxt->dst.type == OP_MEM) 3915 ctxt->dst.bytes = 2; 3916 ctxt->dst.val = ctxt->ops->get_cr(ctxt, 0); 3917 return X86EMUL_CONTINUE; 3918 } 3919 3920 static int em_lmsw(struct x86_emulate_ctxt *ctxt) 3921 { 3922 ctxt->ops->set_cr(ctxt, 0, (ctxt->ops->get_cr(ctxt, 0) & ~0x0eul) 3923 | (ctxt->src.val & 0x0f)); 3924 ctxt->dst.type = OP_NONE; 3925 return X86EMUL_CONTINUE; 3926 } 3927 3928 static int em_loop(struct x86_emulate_ctxt *ctxt) 3929 { 3930 int rc = X86EMUL_CONTINUE; 3931 3932 register_address_increment(ctxt, VCPU_REGS_RCX, -1); 3933 if ((address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) != 0) && 3934 (ctxt->b == 0xe2 || test_cc(ctxt->b ^ 0x5, ctxt->eflags))) 3935 rc = jmp_rel(ctxt, ctxt->src.val); 3936 3937 return rc; 3938 } 3939 3940 static int em_jcxz(struct x86_emulate_ctxt *ctxt) 3941 { 3942 int rc = X86EMUL_CONTINUE; 3943 3944 if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) 3945 rc = jmp_rel(ctxt, ctxt->src.val); 3946 3947 return rc; 3948 } 3949 3950 static int em_in(struct x86_emulate_ctxt *ctxt) 3951 { 3952 if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val, 3953 &ctxt->dst.val)) 3954 return X86EMUL_IO_NEEDED; 3955 3956 return X86EMUL_CONTINUE; 3957 } 3958 3959 static int em_out(struct x86_emulate_ctxt *ctxt) 3960 { 3961 ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val, 3962 &ctxt->src.val, 1); 3963 /* Disable writeback. */ 3964 ctxt->dst.type = OP_NONE; 3965 return X86EMUL_CONTINUE; 3966 } 3967 3968 static int em_cli(struct x86_emulate_ctxt *ctxt) 3969 { 3970 if (emulator_bad_iopl(ctxt)) 3971 return emulate_gp(ctxt, 0); 3972 3973 ctxt->eflags &= ~X86_EFLAGS_IF; 3974 return X86EMUL_CONTINUE; 3975 } 3976 3977 static int em_sti(struct x86_emulate_ctxt *ctxt) 3978 { 3979 if (emulator_bad_iopl(ctxt)) 3980 return emulate_gp(ctxt, 0); 3981 3982 ctxt->interruptibility = KVM_X86_SHADOW_INT_STI; 3983 ctxt->eflags |= X86_EFLAGS_IF; 3984 return X86EMUL_CONTINUE; 3985 } 3986 3987 static int em_cpuid(struct x86_emulate_ctxt *ctxt) 3988 { 3989 u32 eax, ebx, ecx, edx; 3990 u64 msr = 0; 3991 3992 ctxt->ops->get_msr(ctxt, MSR_MISC_FEATURES_ENABLES, &msr); 3993 if (msr & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT && 3994 ctxt->ops->cpl(ctxt)) { 3995 return emulate_gp(ctxt, 0); 3996 } 3997 3998 eax = reg_read(ctxt, VCPU_REGS_RAX); 3999 ecx = reg_read(ctxt, VCPU_REGS_RCX); 4000 ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, &edx, false); 4001 *reg_write(ctxt, VCPU_REGS_RAX) = eax; 4002 *reg_write(ctxt, VCPU_REGS_RBX) = ebx; 4003 *reg_write(ctxt, VCPU_REGS_RCX) = ecx; 4004 *reg_write(ctxt, VCPU_REGS_RDX) = edx; 4005 return X86EMUL_CONTINUE; 4006 } 4007 4008 static int em_sahf(struct x86_emulate_ctxt *ctxt) 4009 { 4010 u32 flags; 4011 4012 flags = X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF | 4013 X86_EFLAGS_SF; 4014 flags &= *reg_rmw(ctxt, VCPU_REGS_RAX) >> 8; 4015 4016 ctxt->eflags &= ~0xffUL; 4017 ctxt->eflags |= flags | X86_EFLAGS_FIXED; 4018 return X86EMUL_CONTINUE; 4019 } 4020 4021 static int em_lahf(struct x86_emulate_ctxt *ctxt) 4022 { 4023 *reg_rmw(ctxt, VCPU_REGS_RAX) &= ~0xff00UL; 4024 *reg_rmw(ctxt, VCPU_REGS_RAX) |= (ctxt->eflags & 0xff) << 8; 4025 return X86EMUL_CONTINUE; 4026 } 4027 4028 static int em_bswap(struct x86_emulate_ctxt *ctxt) 4029 { 4030 switch (ctxt->op_bytes) { 4031 #ifdef CONFIG_X86_64 4032 case 8: 4033 asm("bswap %0" : "+r"(ctxt->dst.val)); 4034 break; 4035 #endif 4036 default: 4037 asm("bswap %0" : "+r"(*(u32 *)&ctxt->dst.val)); 4038 break; 4039 } 4040 return X86EMUL_CONTINUE; 4041 } 4042 4043 static int em_clflush(struct x86_emulate_ctxt *ctxt) 4044 { 4045 /* emulating clflush regardless of cpuid */ 4046 return X86EMUL_CONTINUE; 4047 } 4048 4049 static int em_clflushopt(struct x86_emulate_ctxt *ctxt) 4050 { 4051 /* emulating clflushopt regardless of cpuid */ 4052 return X86EMUL_CONTINUE; 4053 } 4054 4055 static int em_movsxd(struct x86_emulate_ctxt *ctxt) 4056 { 4057 ctxt->dst.val = (s32) ctxt->src.val; 4058 return X86EMUL_CONTINUE; 4059 } 4060 4061 static int check_fxsr(struct x86_emulate_ctxt *ctxt) 4062 { 4063 if (!ctxt->ops->guest_has_fxsr(ctxt)) 4064 return emulate_ud(ctxt); 4065 4066 if (ctxt->ops->get_cr(ctxt, 0) & (X86_CR0_TS | X86_CR0_EM)) 4067 return emulate_nm(ctxt); 4068 4069 /* 4070 * Don't emulate a case that should never be hit, instead of working 4071 * around a lack of fxsave64/fxrstor64 on old compilers. 4072 */ 4073 if (ctxt->mode >= X86EMUL_MODE_PROT64) 4074 return X86EMUL_UNHANDLEABLE; 4075 4076 return X86EMUL_CONTINUE; 4077 } 4078 4079 /* 4080 * Hardware doesn't save and restore XMM 0-7 without CR4.OSFXSR, but does save 4081 * and restore MXCSR. 4082 */ 4083 static size_t __fxstate_size(int nregs) 4084 { 4085 return offsetof(struct fxregs_state, xmm_space[0]) + nregs * 16; 4086 } 4087 4088 static inline size_t fxstate_size(struct x86_emulate_ctxt *ctxt) 4089 { 4090 bool cr4_osfxsr; 4091 if (ctxt->mode == X86EMUL_MODE_PROT64) 4092 return __fxstate_size(16); 4093 4094 cr4_osfxsr = ctxt->ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR; 4095 return __fxstate_size(cr4_osfxsr ? 8 : 0); 4096 } 4097 4098 /* 4099 * FXSAVE and FXRSTOR have 4 different formats depending on execution mode, 4100 * 1) 16 bit mode 4101 * 2) 32 bit mode 4102 * - like (1), but FIP and FDP (foo) are only 16 bit. At least Intel CPUs 4103 * preserve whole 32 bit values, though, so (1) and (2) are the same wrt. 4104 * save and restore 4105 * 3) 64-bit mode with REX.W prefix 4106 * - like (2), but XMM 8-15 are being saved and restored 4107 * 4) 64-bit mode without REX.W prefix 4108 * - like (3), but FIP and FDP are 64 bit 4109 * 4110 * Emulation uses (3) for (1) and (2) and preserves XMM 8-15 to reach the 4111 * desired result. (4) is not emulated. 4112 * 4113 * Note: Guest and host CPUID.(EAX=07H,ECX=0H):EBX[bit 13] (deprecate FPU CS 4114 * and FPU DS) should match. 4115 */ 4116 static int em_fxsave(struct x86_emulate_ctxt *ctxt) 4117 { 4118 struct fxregs_state fx_state; 4119 int rc; 4120 4121 rc = check_fxsr(ctxt); 4122 if (rc != X86EMUL_CONTINUE) 4123 return rc; 4124 4125 emulator_get_fpu(); 4126 4127 rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_state)); 4128 4129 emulator_put_fpu(); 4130 4131 if (rc != X86EMUL_CONTINUE) 4132 return rc; 4133 4134 return segmented_write_std(ctxt, ctxt->memop.addr.mem, &fx_state, 4135 fxstate_size(ctxt)); 4136 } 4137 4138 /* 4139 * FXRSTOR might restore XMM registers not provided by the guest. Fill 4140 * in the host registers (via FXSAVE) instead, so they won't be modified. 4141 * (preemption has to stay disabled until FXRSTOR). 4142 * 4143 * Use noinline to keep the stack for other functions called by callers small. 4144 */ 4145 static noinline int fxregs_fixup(struct fxregs_state *fx_state, 4146 const size_t used_size) 4147 { 4148 struct fxregs_state fx_tmp; 4149 int rc; 4150 4151 rc = asm_safe("fxsave %[fx]", , [fx] "+m"(fx_tmp)); 4152 memcpy((void *)fx_state + used_size, (void *)&fx_tmp + used_size, 4153 __fxstate_size(16) - used_size); 4154 4155 return rc; 4156 } 4157 4158 static int em_fxrstor(struct x86_emulate_ctxt *ctxt) 4159 { 4160 struct fxregs_state fx_state; 4161 int rc; 4162 size_t size; 4163 4164 rc = check_fxsr(ctxt); 4165 if (rc != X86EMUL_CONTINUE) 4166 return rc; 4167 4168 size = fxstate_size(ctxt); 4169 rc = segmented_read_std(ctxt, ctxt->memop.addr.mem, &fx_state, size); 4170 if (rc != X86EMUL_CONTINUE) 4171 return rc; 4172 4173 emulator_get_fpu(); 4174 4175 if (size < __fxstate_size(16)) { 4176 rc = fxregs_fixup(&fx_state, size); 4177 if (rc != X86EMUL_CONTINUE) 4178 goto out; 4179 } 4180 4181 if (fx_state.mxcsr >> 16) { 4182 rc = emulate_gp(ctxt, 0); 4183 goto out; 4184 } 4185 4186 if (rc == X86EMUL_CONTINUE) 4187 rc = asm_safe("fxrstor %[fx]", : [fx] "m"(fx_state)); 4188 4189 out: 4190 emulator_put_fpu(); 4191 4192 return rc; 4193 } 4194 4195 static int em_xsetbv(struct x86_emulate_ctxt *ctxt) 4196 { 4197 u32 eax, ecx, edx; 4198 4199 eax = reg_read(ctxt, VCPU_REGS_RAX); 4200 edx = reg_read(ctxt, VCPU_REGS_RDX); 4201 ecx = reg_read(ctxt, VCPU_REGS_RCX); 4202 4203 if (ctxt->ops->set_xcr(ctxt, ecx, ((u64)edx << 32) | eax)) 4204 return emulate_gp(ctxt, 0); 4205 4206 return X86EMUL_CONTINUE; 4207 } 4208 4209 static bool valid_cr(int nr) 4210 { 4211 switch (nr) { 4212 case 0: 4213 case 2 ... 4: 4214 case 8: 4215 return true; 4216 default: 4217 return false; 4218 } 4219 } 4220 4221 static int check_cr_read(struct x86_emulate_ctxt *ctxt) 4222 { 4223 if (!valid_cr(ctxt->modrm_reg)) 4224 return emulate_ud(ctxt); 4225 4226 return X86EMUL_CONTINUE; 4227 } 4228 4229 static int check_cr_write(struct x86_emulate_ctxt *ctxt) 4230 { 4231 u64 new_val = ctxt->src.val64; 4232 int cr = ctxt->modrm_reg; 4233 u64 efer = 0; 4234 4235 static u64 cr_reserved_bits[] = { 4236 0xffffffff00000000ULL, 4237 0, 0, 0, /* CR3 checked later */ 4238 CR4_RESERVED_BITS, 4239 0, 0, 0, 4240 CR8_RESERVED_BITS, 4241 }; 4242 4243 if (!valid_cr(cr)) 4244 return emulate_ud(ctxt); 4245 4246 if (new_val & cr_reserved_bits[cr]) 4247 return emulate_gp(ctxt, 0); 4248 4249 switch (cr) { 4250 case 0: { 4251 u64 cr4; 4252 if (((new_val & X86_CR0_PG) && !(new_val & X86_CR0_PE)) || 4253 ((new_val & X86_CR0_NW) && !(new_val & X86_CR0_CD))) 4254 return emulate_gp(ctxt, 0); 4255 4256 cr4 = ctxt->ops->get_cr(ctxt, 4); 4257 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 4258 4259 if ((new_val & X86_CR0_PG) && (efer & EFER_LME) && 4260 !(cr4 & X86_CR4_PAE)) 4261 return emulate_gp(ctxt, 0); 4262 4263 break; 4264 } 4265 case 3: { 4266 u64 rsvd = 0; 4267 4268 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 4269 if (efer & EFER_LMA) { 4270 u64 maxphyaddr; 4271 u32 eax, ebx, ecx, edx; 4272 4273 eax = 0x80000008; 4274 ecx = 0; 4275 if (ctxt->ops->get_cpuid(ctxt, &eax, &ebx, &ecx, 4276 &edx, true)) 4277 maxphyaddr = eax & 0xff; 4278 else 4279 maxphyaddr = 36; 4280 rsvd = rsvd_bits(maxphyaddr, 63); 4281 if (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_PCIDE) 4282 rsvd &= ~X86_CR3_PCID_NOFLUSH; 4283 } 4284 4285 if (new_val & rsvd) 4286 return emulate_gp(ctxt, 0); 4287 4288 break; 4289 } 4290 case 4: { 4291 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 4292 4293 if ((efer & EFER_LMA) && !(new_val & X86_CR4_PAE)) 4294 return emulate_gp(ctxt, 0); 4295 4296 break; 4297 } 4298 } 4299 4300 return X86EMUL_CONTINUE; 4301 } 4302 4303 static int check_dr7_gd(struct x86_emulate_ctxt *ctxt) 4304 { 4305 unsigned long dr7; 4306 4307 ctxt->ops->get_dr(ctxt, 7, &dr7); 4308 4309 /* Check if DR7.Global_Enable is set */ 4310 return dr7 & (1 << 13); 4311 } 4312 4313 static int check_dr_read(struct x86_emulate_ctxt *ctxt) 4314 { 4315 int dr = ctxt->modrm_reg; 4316 u64 cr4; 4317 4318 if (dr > 7) 4319 return emulate_ud(ctxt); 4320 4321 cr4 = ctxt->ops->get_cr(ctxt, 4); 4322 if ((cr4 & X86_CR4_DE) && (dr == 4 || dr == 5)) 4323 return emulate_ud(ctxt); 4324 4325 if (check_dr7_gd(ctxt)) { 4326 ulong dr6; 4327 4328 ctxt->ops->get_dr(ctxt, 6, &dr6); 4329 dr6 &= ~DR_TRAP_BITS; 4330 dr6 |= DR6_BD | DR6_RTM; 4331 ctxt->ops->set_dr(ctxt, 6, dr6); 4332 return emulate_db(ctxt); 4333 } 4334 4335 return X86EMUL_CONTINUE; 4336 } 4337 4338 static int check_dr_write(struct x86_emulate_ctxt *ctxt) 4339 { 4340 u64 new_val = ctxt->src.val64; 4341 int dr = ctxt->modrm_reg; 4342 4343 if ((dr == 6 || dr == 7) && (new_val & 0xffffffff00000000ULL)) 4344 return emulate_gp(ctxt, 0); 4345 4346 return check_dr_read(ctxt); 4347 } 4348 4349 static int check_svme(struct x86_emulate_ctxt *ctxt) 4350 { 4351 u64 efer = 0; 4352 4353 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 4354 4355 if (!(efer & EFER_SVME)) 4356 return emulate_ud(ctxt); 4357 4358 return X86EMUL_CONTINUE; 4359 } 4360 4361 static int check_svme_pa(struct x86_emulate_ctxt *ctxt) 4362 { 4363 u64 rax = reg_read(ctxt, VCPU_REGS_RAX); 4364 4365 /* Valid physical address? */ 4366 if (rax & 0xffff000000000000ULL) 4367 return emulate_gp(ctxt, 0); 4368 4369 return check_svme(ctxt); 4370 } 4371 4372 static int check_rdtsc(struct x86_emulate_ctxt *ctxt) 4373 { 4374 u64 cr4 = ctxt->ops->get_cr(ctxt, 4); 4375 4376 if (cr4 & X86_CR4_TSD && ctxt->ops->cpl(ctxt)) 4377 return emulate_ud(ctxt); 4378 4379 return X86EMUL_CONTINUE; 4380 } 4381 4382 static int check_rdpmc(struct x86_emulate_ctxt *ctxt) 4383 { 4384 u64 cr4 = ctxt->ops->get_cr(ctxt, 4); 4385 u64 rcx = reg_read(ctxt, VCPU_REGS_RCX); 4386 4387 /* 4388 * VMware allows access to these Pseduo-PMCs even when read via RDPMC 4389 * in Ring3 when CR4.PCE=0. 4390 */ 4391 if (enable_vmware_backdoor && is_vmware_backdoor_pmc(rcx)) 4392 return X86EMUL_CONTINUE; 4393 4394 if ((!(cr4 & X86_CR4_PCE) && ctxt->ops->cpl(ctxt)) || 4395 ctxt->ops->check_pmc(ctxt, rcx)) 4396 return emulate_gp(ctxt, 0); 4397 4398 return X86EMUL_CONTINUE; 4399 } 4400 4401 static int check_perm_in(struct x86_emulate_ctxt *ctxt) 4402 { 4403 ctxt->dst.bytes = min(ctxt->dst.bytes, 4u); 4404 if (!emulator_io_permited(ctxt, ctxt->src.val, ctxt->dst.bytes)) 4405 return emulate_gp(ctxt, 0); 4406 4407 return X86EMUL_CONTINUE; 4408 } 4409 4410 static int check_perm_out(struct x86_emulate_ctxt *ctxt) 4411 { 4412 ctxt->src.bytes = min(ctxt->src.bytes, 4u); 4413 if (!emulator_io_permited(ctxt, ctxt->dst.val, ctxt->src.bytes)) 4414 return emulate_gp(ctxt, 0); 4415 4416 return X86EMUL_CONTINUE; 4417 } 4418 4419 #define D(_y) { .flags = (_y) } 4420 #define DI(_y, _i) { .flags = (_y)|Intercept, .intercept = x86_intercept_##_i } 4421 #define DIP(_y, _i, _p) { .flags = (_y)|Intercept|CheckPerm, \ 4422 .intercept = x86_intercept_##_i, .check_perm = (_p) } 4423 #define N D(NotImpl) 4424 #define EXT(_f, _e) { .flags = ((_f) | RMExt), .u.group = (_e) } 4425 #define G(_f, _g) { .flags = ((_f) | Group | ModRM), .u.group = (_g) } 4426 #define GD(_f, _g) { .flags = ((_f) | GroupDual | ModRM), .u.gdual = (_g) } 4427 #define ID(_f, _i) { .flags = ((_f) | InstrDual | ModRM), .u.idual = (_i) } 4428 #define MD(_f, _m) { .flags = ((_f) | ModeDual), .u.mdual = (_m) } 4429 #define E(_f, _e) { .flags = ((_f) | Escape | ModRM), .u.esc = (_e) } 4430 #define I(_f, _e) { .flags = (_f), .u.execute = (_e) } 4431 #define F(_f, _e) { .flags = (_f) | Fastop, .u.fastop = (_e) } 4432 #define II(_f, _e, _i) \ 4433 { .flags = (_f)|Intercept, .u.execute = (_e), .intercept = x86_intercept_##_i } 4434 #define IIP(_f, _e, _i, _p) \ 4435 { .flags = (_f)|Intercept|CheckPerm, .u.execute = (_e), \ 4436 .intercept = x86_intercept_##_i, .check_perm = (_p) } 4437 #define GP(_f, _g) { .flags = ((_f) | Prefix), .u.gprefix = (_g) } 4438 4439 #define D2bv(_f) D((_f) | ByteOp), D(_f) 4440 #define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p) 4441 #define I2bv(_f, _e) I((_f) | ByteOp, _e), I(_f, _e) 4442 #define F2bv(_f, _e) F((_f) | ByteOp, _e), F(_f, _e) 4443 #define I2bvIP(_f, _e, _i, _p) \ 4444 IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p) 4445 4446 #define F6ALU(_f, _e) F2bv((_f) | DstMem | SrcReg | ModRM, _e), \ 4447 F2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e), \ 4448 F2bv(((_f) & ~Lock) | DstAcc | SrcImm, _e) 4449 4450 static const struct opcode group7_rm0[] = { 4451 N, 4452 I(SrcNone | Priv | EmulateOnUD, em_hypercall), 4453 N, N, N, N, N, N, 4454 }; 4455 4456 static const struct opcode group7_rm1[] = { 4457 DI(SrcNone | Priv, monitor), 4458 DI(SrcNone | Priv, mwait), 4459 N, N, N, N, N, N, 4460 }; 4461 4462 static const struct opcode group7_rm2[] = { 4463 N, 4464 II(ImplicitOps | Priv, em_xsetbv, xsetbv), 4465 N, N, N, N, N, N, 4466 }; 4467 4468 static const struct opcode group7_rm3[] = { 4469 DIP(SrcNone | Prot | Priv, vmrun, check_svme_pa), 4470 II(SrcNone | Prot | EmulateOnUD, em_hypercall, vmmcall), 4471 DIP(SrcNone | Prot | Priv, vmload, check_svme_pa), 4472 DIP(SrcNone | Prot | Priv, vmsave, check_svme_pa), 4473 DIP(SrcNone | Prot | Priv, stgi, check_svme), 4474 DIP(SrcNone | Prot | Priv, clgi, check_svme), 4475 DIP(SrcNone | Prot | Priv, skinit, check_svme), 4476 DIP(SrcNone | Prot | Priv, invlpga, check_svme), 4477 }; 4478 4479 static const struct opcode group7_rm7[] = { 4480 N, 4481 DIP(SrcNone, rdtscp, check_rdtsc), 4482 N, N, N, N, N, N, 4483 }; 4484 4485 static const struct opcode group1[] = { 4486 F(Lock, em_add), 4487 F(Lock | PageTable, em_or), 4488 F(Lock, em_adc), 4489 F(Lock, em_sbb), 4490 F(Lock | PageTable, em_and), 4491 F(Lock, em_sub), 4492 F(Lock, em_xor), 4493 F(NoWrite, em_cmp), 4494 }; 4495 4496 static const struct opcode group1A[] = { 4497 I(DstMem | SrcNone | Mov | Stack | IncSP | TwoMemOp, em_pop), N, N, N, N, N, N, N, 4498 }; 4499 4500 static const struct opcode group2[] = { 4501 F(DstMem | ModRM, em_rol), 4502 F(DstMem | ModRM, em_ror), 4503 F(DstMem | ModRM, em_rcl), 4504 F(DstMem | ModRM, em_rcr), 4505 F(DstMem | ModRM, em_shl), 4506 F(DstMem | ModRM, em_shr), 4507 F(DstMem | ModRM, em_shl), 4508 F(DstMem | ModRM, em_sar), 4509 }; 4510 4511 static const struct opcode group3[] = { 4512 F(DstMem | SrcImm | NoWrite, em_test), 4513 F(DstMem | SrcImm | NoWrite, em_test), 4514 F(DstMem | SrcNone | Lock, em_not), 4515 F(DstMem | SrcNone | Lock, em_neg), 4516 F(DstXacc | Src2Mem, em_mul_ex), 4517 F(DstXacc | Src2Mem, em_imul_ex), 4518 F(DstXacc | Src2Mem, em_div_ex), 4519 F(DstXacc | Src2Mem, em_idiv_ex), 4520 }; 4521 4522 static const struct opcode group4[] = { 4523 F(ByteOp | DstMem | SrcNone | Lock, em_inc), 4524 F(ByteOp | DstMem | SrcNone | Lock, em_dec), 4525 N, N, N, N, N, N, 4526 }; 4527 4528 static const struct opcode group5[] = { 4529 F(DstMem | SrcNone | Lock, em_inc), 4530 F(DstMem | SrcNone | Lock, em_dec), 4531 I(SrcMem | NearBranch, em_call_near_abs), 4532 I(SrcMemFAddr | ImplicitOps, em_call_far), 4533 I(SrcMem | NearBranch, em_jmp_abs), 4534 I(SrcMemFAddr | ImplicitOps, em_jmp_far), 4535 I(SrcMem | Stack | TwoMemOp, em_push), D(Undefined), 4536 }; 4537 4538 static const struct opcode group6[] = { 4539 II(Prot | DstMem, em_sldt, sldt), 4540 II(Prot | DstMem, em_str, str), 4541 II(Prot | Priv | SrcMem16, em_lldt, lldt), 4542 II(Prot | Priv | SrcMem16, em_ltr, ltr), 4543 N, N, N, N, 4544 }; 4545 4546 static const struct group_dual group7 = { { 4547 II(Mov | DstMem, em_sgdt, sgdt), 4548 II(Mov | DstMem, em_sidt, sidt), 4549 II(SrcMem | Priv, em_lgdt, lgdt), 4550 II(SrcMem | Priv, em_lidt, lidt), 4551 II(SrcNone | DstMem | Mov, em_smsw, smsw), N, 4552 II(SrcMem16 | Mov | Priv, em_lmsw, lmsw), 4553 II(SrcMem | ByteOp | Priv | NoAccess, em_invlpg, invlpg), 4554 }, { 4555 EXT(0, group7_rm0), 4556 EXT(0, group7_rm1), 4557 EXT(0, group7_rm2), 4558 EXT(0, group7_rm3), 4559 II(SrcNone | DstMem | Mov, em_smsw, smsw), N, 4560 II(SrcMem16 | Mov | Priv, em_lmsw, lmsw), 4561 EXT(0, group7_rm7), 4562 } }; 4563 4564 static const struct opcode group8[] = { 4565 N, N, N, N, 4566 F(DstMem | SrcImmByte | NoWrite, em_bt), 4567 F(DstMem | SrcImmByte | Lock | PageTable, em_bts), 4568 F(DstMem | SrcImmByte | Lock, em_btr), 4569 F(DstMem | SrcImmByte | Lock | PageTable, em_btc), 4570 }; 4571 4572 /* 4573 * The "memory" destination is actually always a register, since we come 4574 * from the register case of group9. 4575 */ 4576 static const struct gprefix pfx_0f_c7_7 = { 4577 N, N, N, II(DstMem | ModRM | Op3264 | EmulateOnUD, em_rdpid, rdtscp), 4578 }; 4579 4580 4581 static const struct group_dual group9 = { { 4582 N, I(DstMem64 | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N, 4583 }, { 4584 N, N, N, N, N, N, N, 4585 GP(0, &pfx_0f_c7_7), 4586 } }; 4587 4588 static const struct opcode group11[] = { 4589 I(DstMem | SrcImm | Mov | PageTable, em_mov), 4590 X7(D(Undefined)), 4591 }; 4592 4593 static const struct gprefix pfx_0f_ae_7 = { 4594 I(SrcMem | ByteOp, em_clflush), I(SrcMem | ByteOp, em_clflushopt), N, N, 4595 }; 4596 4597 static const struct group_dual group15 = { { 4598 I(ModRM | Aligned16, em_fxsave), 4599 I(ModRM | Aligned16, em_fxrstor), 4600 N, N, N, N, N, GP(0, &pfx_0f_ae_7), 4601 }, { 4602 N, N, N, N, N, N, N, N, 4603 } }; 4604 4605 static const struct gprefix pfx_0f_6f_0f_7f = { 4606 I(Mmx, em_mov), I(Sse | Aligned, em_mov), N, I(Sse | Unaligned, em_mov), 4607 }; 4608 4609 static const struct instr_dual instr_dual_0f_2b = { 4610 I(0, em_mov), N 4611 }; 4612 4613 static const struct gprefix pfx_0f_2b = { 4614 ID(0, &instr_dual_0f_2b), ID(0, &instr_dual_0f_2b), N, N, 4615 }; 4616 4617 static const struct gprefix pfx_0f_10_0f_11 = { 4618 I(Unaligned, em_mov), I(Unaligned, em_mov), N, N, 4619 }; 4620 4621 static const struct gprefix pfx_0f_28_0f_29 = { 4622 I(Aligned, em_mov), I(Aligned, em_mov), N, N, 4623 }; 4624 4625 static const struct gprefix pfx_0f_e7 = { 4626 N, I(Sse, em_mov), N, N, 4627 }; 4628 4629 static const struct escape escape_d9 = { { 4630 N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstcw), 4631 }, { 4632 /* 0xC0 - 0xC7 */ 4633 N, N, N, N, N, N, N, N, 4634 /* 0xC8 - 0xCF */ 4635 N, N, N, N, N, N, N, N, 4636 /* 0xD0 - 0xC7 */ 4637 N, N, N, N, N, N, N, N, 4638 /* 0xD8 - 0xDF */ 4639 N, N, N, N, N, N, N, N, 4640 /* 0xE0 - 0xE7 */ 4641 N, N, N, N, N, N, N, N, 4642 /* 0xE8 - 0xEF */ 4643 N, N, N, N, N, N, N, N, 4644 /* 0xF0 - 0xF7 */ 4645 N, N, N, N, N, N, N, N, 4646 /* 0xF8 - 0xFF */ 4647 N, N, N, N, N, N, N, N, 4648 } }; 4649 4650 static const struct escape escape_db = { { 4651 N, N, N, N, N, N, N, N, 4652 }, { 4653 /* 0xC0 - 0xC7 */ 4654 N, N, N, N, N, N, N, N, 4655 /* 0xC8 - 0xCF */ 4656 N, N, N, N, N, N, N, N, 4657 /* 0xD0 - 0xC7 */ 4658 N, N, N, N, N, N, N, N, 4659 /* 0xD8 - 0xDF */ 4660 N, N, N, N, N, N, N, N, 4661 /* 0xE0 - 0xE7 */ 4662 N, N, N, I(ImplicitOps, em_fninit), N, N, N, N, 4663 /* 0xE8 - 0xEF */ 4664 N, N, N, N, N, N, N, N, 4665 /* 0xF0 - 0xF7 */ 4666 N, N, N, N, N, N, N, N, 4667 /* 0xF8 - 0xFF */ 4668 N, N, N, N, N, N, N, N, 4669 } }; 4670 4671 static const struct escape escape_dd = { { 4672 N, N, N, N, N, N, N, I(DstMem16 | Mov, em_fnstsw), 4673 }, { 4674 /* 0xC0 - 0xC7 */ 4675 N, N, N, N, N, N, N, N, 4676 /* 0xC8 - 0xCF */ 4677 N, N, N, N, N, N, N, N, 4678 /* 0xD0 - 0xC7 */ 4679 N, N, N, N, N, N, N, N, 4680 /* 0xD8 - 0xDF */ 4681 N, N, N, N, N, N, N, N, 4682 /* 0xE0 - 0xE7 */ 4683 N, N, N, N, N, N, N, N, 4684 /* 0xE8 - 0xEF */ 4685 N, N, N, N, N, N, N, N, 4686 /* 0xF0 - 0xF7 */ 4687 N, N, N, N, N, N, N, N, 4688 /* 0xF8 - 0xFF */ 4689 N, N, N, N, N, N, N, N, 4690 } }; 4691 4692 static const struct instr_dual instr_dual_0f_c3 = { 4693 I(DstMem | SrcReg | ModRM | No16 | Mov, em_mov), N 4694 }; 4695 4696 static const struct mode_dual mode_dual_63 = { 4697 N, I(DstReg | SrcMem32 | ModRM | Mov, em_movsxd) 4698 }; 4699 4700 static const struct opcode opcode_table[256] = { 4701 /* 0x00 - 0x07 */ 4702 F6ALU(Lock, em_add), 4703 I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg), 4704 I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg), 4705 /* 0x08 - 0x0F */ 4706 F6ALU(Lock | PageTable, em_or), 4707 I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg), 4708 N, 4709 /* 0x10 - 0x17 */ 4710 F6ALU(Lock, em_adc), 4711 I(ImplicitOps | Stack | No64 | Src2SS, em_push_sreg), 4712 I(ImplicitOps | Stack | No64 | Src2SS, em_pop_sreg), 4713 /* 0x18 - 0x1F */ 4714 F6ALU(Lock, em_sbb), 4715 I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg), 4716 I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg), 4717 /* 0x20 - 0x27 */ 4718 F6ALU(Lock | PageTable, em_and), N, N, 4719 /* 0x28 - 0x2F */ 4720 F6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das), 4721 /* 0x30 - 0x37 */ 4722 F6ALU(Lock, em_xor), N, N, 4723 /* 0x38 - 0x3F */ 4724 F6ALU(NoWrite, em_cmp), N, N, 4725 /* 0x40 - 0x4F */ 4726 X8(F(DstReg, em_inc)), X8(F(DstReg, em_dec)), 4727 /* 0x50 - 0x57 */ 4728 X8(I(SrcReg | Stack, em_push)), 4729 /* 0x58 - 0x5F */ 4730 X8(I(DstReg | Stack, em_pop)), 4731 /* 0x60 - 0x67 */ 4732 I(ImplicitOps | Stack | No64, em_pusha), 4733 I(ImplicitOps | Stack | No64, em_popa), 4734 N, MD(ModRM, &mode_dual_63), 4735 N, N, N, N, 4736 /* 0x68 - 0x6F */ 4737 I(SrcImm | Mov | Stack, em_push), 4738 I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op), 4739 I(SrcImmByte | Mov | Stack, em_push), 4740 I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op), 4741 I2bvIP(DstDI | SrcDX | Mov | String | Unaligned, em_in, ins, check_perm_in), /* insb, insw/insd */ 4742 I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */ 4743 /* 0x70 - 0x7F */ 4744 X16(D(SrcImmByte | NearBranch)), 4745 /* 0x80 - 0x87 */ 4746 G(ByteOp | DstMem | SrcImm, group1), 4747 G(DstMem | SrcImm, group1), 4748 G(ByteOp | DstMem | SrcImm | No64, group1), 4749 G(DstMem | SrcImmByte, group1), 4750 F2bv(DstMem | SrcReg | ModRM | NoWrite, em_test), 4751 I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg), 4752 /* 0x88 - 0x8F */ 4753 I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov), 4754 I2bv(DstReg | SrcMem | ModRM | Mov, em_mov), 4755 I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg), 4756 D(ModRM | SrcMem | NoAccess | DstReg), 4757 I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm), 4758 G(0, group1A), 4759 /* 0x90 - 0x97 */ 4760 DI(SrcAcc | DstReg, pause), X7(D(SrcAcc | DstReg)), 4761 /* 0x98 - 0x9F */ 4762 D(DstAcc | SrcNone), I(ImplicitOps | SrcAcc, em_cwd), 4763 I(SrcImmFAddr | No64, em_call_far), N, 4764 II(ImplicitOps | Stack, em_pushf, pushf), 4765 II(ImplicitOps | Stack, em_popf, popf), 4766 I(ImplicitOps, em_sahf), I(ImplicitOps, em_lahf), 4767 /* 0xA0 - 0xA7 */ 4768 I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov), 4769 I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov), 4770 I2bv(SrcSI | DstDI | Mov | String | TwoMemOp, em_mov), 4771 F2bv(SrcSI | DstDI | String | NoWrite | TwoMemOp, em_cmp_r), 4772 /* 0xA8 - 0xAF */ 4773 F2bv(DstAcc | SrcImm | NoWrite, em_test), 4774 I2bv(SrcAcc | DstDI | Mov | String, em_mov), 4775 I2bv(SrcSI | DstAcc | Mov | String, em_mov), 4776 F2bv(SrcAcc | DstDI | String | NoWrite, em_cmp_r), 4777 /* 0xB0 - 0xB7 */ 4778 X8(I(ByteOp | DstReg | SrcImm | Mov, em_mov)), 4779 /* 0xB8 - 0xBF */ 4780 X8(I(DstReg | SrcImm64 | Mov, em_mov)), 4781 /* 0xC0 - 0xC7 */ 4782 G(ByteOp | Src2ImmByte, group2), G(Src2ImmByte, group2), 4783 I(ImplicitOps | NearBranch | SrcImmU16, em_ret_near_imm), 4784 I(ImplicitOps | NearBranch, em_ret), 4785 I(DstReg | SrcMemFAddr | ModRM | No64 | Src2ES, em_lseg), 4786 I(DstReg | SrcMemFAddr | ModRM | No64 | Src2DS, em_lseg), 4787 G(ByteOp, group11), G(0, group11), 4788 /* 0xC8 - 0xCF */ 4789 I(Stack | SrcImmU16 | Src2ImmByte, em_enter), I(Stack, em_leave), 4790 I(ImplicitOps | SrcImmU16, em_ret_far_imm), 4791 I(ImplicitOps, em_ret_far), 4792 D(ImplicitOps), DI(SrcImmByte, intn), 4793 D(ImplicitOps | No64), II(ImplicitOps, em_iret, iret), 4794 /* 0xD0 - 0xD7 */ 4795 G(Src2One | ByteOp, group2), G(Src2One, group2), 4796 G(Src2CL | ByteOp, group2), G(Src2CL, group2), 4797 I(DstAcc | SrcImmUByte | No64, em_aam), 4798 I(DstAcc | SrcImmUByte | No64, em_aad), 4799 F(DstAcc | ByteOp | No64, em_salc), 4800 I(DstAcc | SrcXLat | ByteOp, em_mov), 4801 /* 0xD8 - 0xDF */ 4802 N, E(0, &escape_d9), N, E(0, &escape_db), N, E(0, &escape_dd), N, N, 4803 /* 0xE0 - 0xE7 */ 4804 X3(I(SrcImmByte | NearBranch, em_loop)), 4805 I(SrcImmByte | NearBranch, em_jcxz), 4806 I2bvIP(SrcImmUByte | DstAcc, em_in, in, check_perm_in), 4807 I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out), 4808 /* 0xE8 - 0xEF */ 4809 I(SrcImm | NearBranch, em_call), D(SrcImm | ImplicitOps | NearBranch), 4810 I(SrcImmFAddr | No64, em_jmp_far), 4811 D(SrcImmByte | ImplicitOps | NearBranch), 4812 I2bvIP(SrcDX | DstAcc, em_in, in, check_perm_in), 4813 I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out), 4814 /* 0xF0 - 0xF7 */ 4815 N, DI(ImplicitOps, icebp), N, N, 4816 DI(ImplicitOps | Priv, hlt), D(ImplicitOps), 4817 G(ByteOp, group3), G(0, group3), 4818 /* 0xF8 - 0xFF */ 4819 D(ImplicitOps), D(ImplicitOps), 4820 I(ImplicitOps, em_cli), I(ImplicitOps, em_sti), 4821 D(ImplicitOps), D(ImplicitOps), G(0, group4), G(0, group5), 4822 }; 4823 4824 static const struct opcode twobyte_table[256] = { 4825 /* 0x00 - 0x0F */ 4826 G(0, group6), GD(0, &group7), N, N, 4827 N, I(ImplicitOps | EmulateOnUD, em_syscall), 4828 II(ImplicitOps | Priv, em_clts, clts), N, 4829 DI(ImplicitOps | Priv, invd), DI(ImplicitOps | Priv, wbinvd), N, N, 4830 N, D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N, 4831 /* 0x10 - 0x1F */ 4832 GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_10_0f_11), 4833 GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_10_0f_11), 4834 N, N, N, N, N, N, 4835 D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 4 * prefetch + 4 * reserved NOP */ 4836 D(ImplicitOps | ModRM | SrcMem | NoAccess), N, N, 4837 D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */ 4838 D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */ 4839 D(ImplicitOps | ModRM | SrcMem | NoAccess), /* 8 * reserved NOP */ 4840 D(ImplicitOps | ModRM | SrcMem | NoAccess), /* NOP + 7 * reserved NOP */ 4841 /* 0x20 - 0x2F */ 4842 DIP(ModRM | DstMem | Priv | Op3264 | NoMod, cr_read, check_cr_read), 4843 DIP(ModRM | DstMem | Priv | Op3264 | NoMod, dr_read, check_dr_read), 4844 IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_cr_write, cr_write, 4845 check_cr_write), 4846 IIP(ModRM | SrcMem | Priv | Op3264 | NoMod, em_dr_write, dr_write, 4847 check_dr_write), 4848 N, N, N, N, 4849 GP(ModRM | DstReg | SrcMem | Mov | Sse, &pfx_0f_28_0f_29), 4850 GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_28_0f_29), 4851 N, GP(ModRM | DstMem | SrcReg | Mov | Sse, &pfx_0f_2b), 4852 N, N, N, N, 4853 /* 0x30 - 0x3F */ 4854 II(ImplicitOps | Priv, em_wrmsr, wrmsr), 4855 IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc), 4856 II(ImplicitOps | Priv, em_rdmsr, rdmsr), 4857 IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc), 4858 I(ImplicitOps | EmulateOnUD, em_sysenter), 4859 I(ImplicitOps | Priv | EmulateOnUD, em_sysexit), 4860 N, N, 4861 N, N, N, N, N, N, N, N, 4862 /* 0x40 - 0x4F */ 4863 X16(D(DstReg | SrcMem | ModRM)), 4864 /* 0x50 - 0x5F */ 4865 N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, 4866 /* 0x60 - 0x6F */ 4867 N, N, N, N, 4868 N, N, N, N, 4869 N, N, N, N, 4870 N, N, N, GP(SrcMem | DstReg | ModRM | Mov, &pfx_0f_6f_0f_7f), 4871 /* 0x70 - 0x7F */ 4872 N, N, N, N, 4873 N, N, N, N, 4874 N, N, N, N, 4875 N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_6f_0f_7f), 4876 /* 0x80 - 0x8F */ 4877 X16(D(SrcImm | NearBranch)), 4878 /* 0x90 - 0x9F */ 4879 X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)), 4880 /* 0xA0 - 0xA7 */ 4881 I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg), 4882 II(ImplicitOps, em_cpuid, cpuid), 4883 F(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt), 4884 F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld), 4885 F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N, 4886 /* 0xA8 - 0xAF */ 4887 I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg), 4888 II(EmulateOnUD | ImplicitOps, em_rsm, rsm), 4889 F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts), 4890 F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd), 4891 F(DstMem | SrcReg | Src2CL | ModRM, em_shrd), 4892 GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul), 4893 /* 0xB0 - 0xB7 */ 4894 I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg), 4895 I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg), 4896 F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr), 4897 I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg), 4898 I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg), 4899 D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov), 4900 /* 0xB8 - 0xBF */ 4901 N, N, 4902 G(BitOp, group8), 4903 F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc), 4904 I(DstReg | SrcMem | ModRM, em_bsf_c), 4905 I(DstReg | SrcMem | ModRM, em_bsr_c), 4906 D(DstReg | SrcMem8 | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov), 4907 /* 0xC0 - 0xC7 */ 4908 F2bv(DstMem | SrcReg | ModRM | SrcWrite | Lock, em_xadd), 4909 N, ID(0, &instr_dual_0f_c3), 4910 N, N, N, GD(0, &group9), 4911 /* 0xC8 - 0xCF */ 4912 X8(I(DstReg, em_bswap)), 4913 /* 0xD0 - 0xDF */ 4914 N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, 4915 /* 0xE0 - 0xEF */ 4916 N, N, N, N, N, N, N, GP(SrcReg | DstMem | ModRM | Mov, &pfx_0f_e7), 4917 N, N, N, N, N, N, N, N, 4918 /* 0xF0 - 0xFF */ 4919 N, N, N, N, N, N, N, N, N, N, N, N, N, N, N, N 4920 }; 4921 4922 static const struct instr_dual instr_dual_0f_38_f0 = { 4923 I(DstReg | SrcMem | Mov, em_movbe), N 4924 }; 4925 4926 static const struct instr_dual instr_dual_0f_38_f1 = { 4927 I(DstMem | SrcReg | Mov, em_movbe), N 4928 }; 4929 4930 static const struct gprefix three_byte_0f_38_f0 = { 4931 ID(0, &instr_dual_0f_38_f0), N, N, N 4932 }; 4933 4934 static const struct gprefix three_byte_0f_38_f1 = { 4935 ID(0, &instr_dual_0f_38_f1), N, N, N 4936 }; 4937 4938 /* 4939 * Insns below are selected by the prefix which indexed by the third opcode 4940 * byte. 4941 */ 4942 static const struct opcode opcode_map_0f_38[256] = { 4943 /* 0x00 - 0x7f */ 4944 X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), 4945 /* 0x80 - 0xef */ 4946 X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), X16(N), 4947 /* 0xf0 - 0xf1 */ 4948 GP(EmulateOnUD | ModRM, &three_byte_0f_38_f0), 4949 GP(EmulateOnUD | ModRM, &three_byte_0f_38_f1), 4950 /* 0xf2 - 0xff */ 4951 N, N, X4(N), X8(N) 4952 }; 4953 4954 #undef D 4955 #undef N 4956 #undef G 4957 #undef GD 4958 #undef I 4959 #undef GP 4960 #undef EXT 4961 #undef MD 4962 #undef ID 4963 4964 #undef D2bv 4965 #undef D2bvIP 4966 #undef I2bv 4967 #undef I2bvIP 4968 #undef I6ALU 4969 4970 static unsigned imm_size(struct x86_emulate_ctxt *ctxt) 4971 { 4972 unsigned size; 4973 4974 size = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 4975 if (size == 8) 4976 size = 4; 4977 return size; 4978 } 4979 4980 static int decode_imm(struct x86_emulate_ctxt *ctxt, struct operand *op, 4981 unsigned size, bool sign_extension) 4982 { 4983 int rc = X86EMUL_CONTINUE; 4984 4985 op->type = OP_IMM; 4986 op->bytes = size; 4987 op->addr.mem.ea = ctxt->_eip; 4988 /* NB. Immediates are sign-extended as necessary. */ 4989 switch (op->bytes) { 4990 case 1: 4991 op->val = insn_fetch(s8, ctxt); 4992 break; 4993 case 2: 4994 op->val = insn_fetch(s16, ctxt); 4995 break; 4996 case 4: 4997 op->val = insn_fetch(s32, ctxt); 4998 break; 4999 case 8: 5000 op->val = insn_fetch(s64, ctxt); 5001 break; 5002 } 5003 if (!sign_extension) { 5004 switch (op->bytes) { 5005 case 1: 5006 op->val &= 0xff; 5007 break; 5008 case 2: 5009 op->val &= 0xffff; 5010 break; 5011 case 4: 5012 op->val &= 0xffffffff; 5013 break; 5014 } 5015 } 5016 done: 5017 return rc; 5018 } 5019 5020 static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op, 5021 unsigned d) 5022 { 5023 int rc = X86EMUL_CONTINUE; 5024 5025 switch (d) { 5026 case OpReg: 5027 decode_register_operand(ctxt, op); 5028 break; 5029 case OpImmUByte: 5030 rc = decode_imm(ctxt, op, 1, false); 5031 break; 5032 case OpMem: 5033 ctxt->memop.bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 5034 mem_common: 5035 *op = ctxt->memop; 5036 ctxt->memopp = op; 5037 if (ctxt->d & BitOp) 5038 fetch_bit_operand(ctxt); 5039 op->orig_val = op->val; 5040 break; 5041 case OpMem64: 5042 ctxt->memop.bytes = (ctxt->op_bytes == 8) ? 16 : 8; 5043 goto mem_common; 5044 case OpAcc: 5045 op->type = OP_REG; 5046 op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 5047 op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX); 5048 fetch_register_operand(op); 5049 op->orig_val = op->val; 5050 break; 5051 case OpAccLo: 5052 op->type = OP_REG; 5053 op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes; 5054 op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX); 5055 fetch_register_operand(op); 5056 op->orig_val = op->val; 5057 break; 5058 case OpAccHi: 5059 if (ctxt->d & ByteOp) { 5060 op->type = OP_NONE; 5061 break; 5062 } 5063 op->type = OP_REG; 5064 op->bytes = ctxt->op_bytes; 5065 op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX); 5066 fetch_register_operand(op); 5067 op->orig_val = op->val; 5068 break; 5069 case OpDI: 5070 op->type = OP_MEM; 5071 op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 5072 op->addr.mem.ea = 5073 register_address(ctxt, VCPU_REGS_RDI); 5074 op->addr.mem.seg = VCPU_SREG_ES; 5075 op->val = 0; 5076 op->count = 1; 5077 break; 5078 case OpDX: 5079 op->type = OP_REG; 5080 op->bytes = 2; 5081 op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX); 5082 fetch_register_operand(op); 5083 break; 5084 case OpCL: 5085 op->type = OP_IMM; 5086 op->bytes = 1; 5087 op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff; 5088 break; 5089 case OpImmByte: 5090 rc = decode_imm(ctxt, op, 1, true); 5091 break; 5092 case OpOne: 5093 op->type = OP_IMM; 5094 op->bytes = 1; 5095 op->val = 1; 5096 break; 5097 case OpImm: 5098 rc = decode_imm(ctxt, op, imm_size(ctxt), true); 5099 break; 5100 case OpImm64: 5101 rc = decode_imm(ctxt, op, ctxt->op_bytes, true); 5102 break; 5103 case OpMem8: 5104 ctxt->memop.bytes = 1; 5105 if (ctxt->memop.type == OP_REG) { 5106 ctxt->memop.addr.reg = decode_register(ctxt, 5107 ctxt->modrm_rm, true); 5108 fetch_register_operand(&ctxt->memop); 5109 } 5110 goto mem_common; 5111 case OpMem16: 5112 ctxt->memop.bytes = 2; 5113 goto mem_common; 5114 case OpMem32: 5115 ctxt->memop.bytes = 4; 5116 goto mem_common; 5117 case OpImmU16: 5118 rc = decode_imm(ctxt, op, 2, false); 5119 break; 5120 case OpImmU: 5121 rc = decode_imm(ctxt, op, imm_size(ctxt), false); 5122 break; 5123 case OpSI: 5124 op->type = OP_MEM; 5125 op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 5126 op->addr.mem.ea = 5127 register_address(ctxt, VCPU_REGS_RSI); 5128 op->addr.mem.seg = ctxt->seg_override; 5129 op->val = 0; 5130 op->count = 1; 5131 break; 5132 case OpXLat: 5133 op->type = OP_MEM; 5134 op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; 5135 op->addr.mem.ea = 5136 address_mask(ctxt, 5137 reg_read(ctxt, VCPU_REGS_RBX) + 5138 (reg_read(ctxt, VCPU_REGS_RAX) & 0xff)); 5139 op->addr.mem.seg = ctxt->seg_override; 5140 op->val = 0; 5141 break; 5142 case OpImmFAddr: 5143 op->type = OP_IMM; 5144 op->addr.mem.ea = ctxt->_eip; 5145 op->bytes = ctxt->op_bytes + 2; 5146 insn_fetch_arr(op->valptr, op->bytes, ctxt); 5147 break; 5148 case OpMemFAddr: 5149 ctxt->memop.bytes = ctxt->op_bytes + 2; 5150 goto mem_common; 5151 case OpES: 5152 op->type = OP_IMM; 5153 op->val = VCPU_SREG_ES; 5154 break; 5155 case OpCS: 5156 op->type = OP_IMM; 5157 op->val = VCPU_SREG_CS; 5158 break; 5159 case OpSS: 5160 op->type = OP_IMM; 5161 op->val = VCPU_SREG_SS; 5162 break; 5163 case OpDS: 5164 op->type = OP_IMM; 5165 op->val = VCPU_SREG_DS; 5166 break; 5167 case OpFS: 5168 op->type = OP_IMM; 5169 op->val = VCPU_SREG_FS; 5170 break; 5171 case OpGS: 5172 op->type = OP_IMM; 5173 op->val = VCPU_SREG_GS; 5174 break; 5175 case OpImplicit: 5176 /* Special instructions do their own operand decoding. */ 5177 default: 5178 op->type = OP_NONE; /* Disable writeback. */ 5179 break; 5180 } 5181 5182 done: 5183 return rc; 5184 } 5185 5186 int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len) 5187 { 5188 int rc = X86EMUL_CONTINUE; 5189 int mode = ctxt->mode; 5190 int def_op_bytes, def_ad_bytes, goffset, simd_prefix; 5191 bool op_prefix = false; 5192 bool has_seg_override = false; 5193 struct opcode opcode; 5194 u16 dummy; 5195 struct desc_struct desc; 5196 5197 ctxt->memop.type = OP_NONE; 5198 ctxt->memopp = NULL; 5199 ctxt->_eip = ctxt->eip; 5200 ctxt->fetch.ptr = ctxt->fetch.data; 5201 ctxt->fetch.end = ctxt->fetch.data + insn_len; 5202 ctxt->opcode_len = 1; 5203 ctxt->intercept = x86_intercept_none; 5204 if (insn_len > 0) 5205 memcpy(ctxt->fetch.data, insn, insn_len); 5206 else { 5207 rc = __do_insn_fetch_bytes(ctxt, 1); 5208 if (rc != X86EMUL_CONTINUE) 5209 goto done; 5210 } 5211 5212 switch (mode) { 5213 case X86EMUL_MODE_REAL: 5214 case X86EMUL_MODE_VM86: 5215 def_op_bytes = def_ad_bytes = 2; 5216 ctxt->ops->get_segment(ctxt, &dummy, &desc, NULL, VCPU_SREG_CS); 5217 if (desc.d) 5218 def_op_bytes = def_ad_bytes = 4; 5219 break; 5220 case X86EMUL_MODE_PROT16: 5221 def_op_bytes = def_ad_bytes = 2; 5222 break; 5223 case X86EMUL_MODE_PROT32: 5224 def_op_bytes = def_ad_bytes = 4; 5225 break; 5226 #ifdef CONFIG_X86_64 5227 case X86EMUL_MODE_PROT64: 5228 def_op_bytes = 4; 5229 def_ad_bytes = 8; 5230 break; 5231 #endif 5232 default: 5233 return EMULATION_FAILED; 5234 } 5235 5236 ctxt->op_bytes = def_op_bytes; 5237 ctxt->ad_bytes = def_ad_bytes; 5238 5239 /* Legacy prefixes. */ 5240 for (;;) { 5241 switch (ctxt->b = insn_fetch(u8, ctxt)) { 5242 case 0x66: /* operand-size override */ 5243 op_prefix = true; 5244 /* switch between 2/4 bytes */ 5245 ctxt->op_bytes = def_op_bytes ^ 6; 5246 break; 5247 case 0x67: /* address-size override */ 5248 if (mode == X86EMUL_MODE_PROT64) 5249 /* switch between 4/8 bytes */ 5250 ctxt->ad_bytes = def_ad_bytes ^ 12; 5251 else 5252 /* switch between 2/4 bytes */ 5253 ctxt->ad_bytes = def_ad_bytes ^ 6; 5254 break; 5255 case 0x26: /* ES override */ 5256 has_seg_override = true; 5257 ctxt->seg_override = VCPU_SREG_ES; 5258 break; 5259 case 0x2e: /* CS override */ 5260 has_seg_override = true; 5261 ctxt->seg_override = VCPU_SREG_CS; 5262 break; 5263 case 0x36: /* SS override */ 5264 has_seg_override = true; 5265 ctxt->seg_override = VCPU_SREG_SS; 5266 break; 5267 case 0x3e: /* DS override */ 5268 has_seg_override = true; 5269 ctxt->seg_override = VCPU_SREG_DS; 5270 break; 5271 case 0x64: /* FS override */ 5272 has_seg_override = true; 5273 ctxt->seg_override = VCPU_SREG_FS; 5274 break; 5275 case 0x65: /* GS override */ 5276 has_seg_override = true; 5277 ctxt->seg_override = VCPU_SREG_GS; 5278 break; 5279 case 0x40 ... 0x4f: /* REX */ 5280 if (mode != X86EMUL_MODE_PROT64) 5281 goto done_prefixes; 5282 ctxt->rex_prefix = ctxt->b; 5283 continue; 5284 case 0xf0: /* LOCK */ 5285 ctxt->lock_prefix = 1; 5286 break; 5287 case 0xf2: /* REPNE/REPNZ */ 5288 case 0xf3: /* REP/REPE/REPZ */ 5289 ctxt->rep_prefix = ctxt->b; 5290 break; 5291 default: 5292 goto done_prefixes; 5293 } 5294 5295 /* Any legacy prefix after a REX prefix nullifies its effect. */ 5296 5297 ctxt->rex_prefix = 0; 5298 } 5299 5300 done_prefixes: 5301 5302 /* REX prefix. */ 5303 if (ctxt->rex_prefix & 8) 5304 ctxt->op_bytes = 8; /* REX.W */ 5305 5306 /* Opcode byte(s). */ 5307 opcode = opcode_table[ctxt->b]; 5308 /* Two-byte opcode? */ 5309 if (ctxt->b == 0x0f) { 5310 ctxt->opcode_len = 2; 5311 ctxt->b = insn_fetch(u8, ctxt); 5312 opcode = twobyte_table[ctxt->b]; 5313 5314 /* 0F_38 opcode map */ 5315 if (ctxt->b == 0x38) { 5316 ctxt->opcode_len = 3; 5317 ctxt->b = insn_fetch(u8, ctxt); 5318 opcode = opcode_map_0f_38[ctxt->b]; 5319 } 5320 } 5321 ctxt->d = opcode.flags; 5322 5323 if (ctxt->d & ModRM) 5324 ctxt->modrm = insn_fetch(u8, ctxt); 5325 5326 /* vex-prefix instructions are not implemented */ 5327 if (ctxt->opcode_len == 1 && (ctxt->b == 0xc5 || ctxt->b == 0xc4) && 5328 (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0)) { 5329 ctxt->d = NotImpl; 5330 } 5331 5332 while (ctxt->d & GroupMask) { 5333 switch (ctxt->d & GroupMask) { 5334 case Group: 5335 goffset = (ctxt->modrm >> 3) & 7; 5336 opcode = opcode.u.group[goffset]; 5337 break; 5338 case GroupDual: 5339 goffset = (ctxt->modrm >> 3) & 7; 5340 if ((ctxt->modrm >> 6) == 3) 5341 opcode = opcode.u.gdual->mod3[goffset]; 5342 else 5343 opcode = opcode.u.gdual->mod012[goffset]; 5344 break; 5345 case RMExt: 5346 goffset = ctxt->modrm & 7; 5347 opcode = opcode.u.group[goffset]; 5348 break; 5349 case Prefix: 5350 if (ctxt->rep_prefix && op_prefix) 5351 return EMULATION_FAILED; 5352 simd_prefix = op_prefix ? 0x66 : ctxt->rep_prefix; 5353 switch (simd_prefix) { 5354 case 0x00: opcode = opcode.u.gprefix->pfx_no; break; 5355 case 0x66: opcode = opcode.u.gprefix->pfx_66; break; 5356 case 0xf2: opcode = opcode.u.gprefix->pfx_f2; break; 5357 case 0xf3: opcode = opcode.u.gprefix->pfx_f3; break; 5358 } 5359 break; 5360 case Escape: 5361 if (ctxt->modrm > 0xbf) { 5362 size_t size = ARRAY_SIZE(opcode.u.esc->high); 5363 u32 index = array_index_nospec( 5364 ctxt->modrm - 0xc0, size); 5365 5366 opcode = opcode.u.esc->high[index]; 5367 } else { 5368 opcode = opcode.u.esc->op[(ctxt->modrm >> 3) & 7]; 5369 } 5370 break; 5371 case InstrDual: 5372 if ((ctxt->modrm >> 6) == 3) 5373 opcode = opcode.u.idual->mod3; 5374 else 5375 opcode = opcode.u.idual->mod012; 5376 break; 5377 case ModeDual: 5378 if (ctxt->mode == X86EMUL_MODE_PROT64) 5379 opcode = opcode.u.mdual->mode64; 5380 else 5381 opcode = opcode.u.mdual->mode32; 5382 break; 5383 default: 5384 return EMULATION_FAILED; 5385 } 5386 5387 ctxt->d &= ~(u64)GroupMask; 5388 ctxt->d |= opcode.flags; 5389 } 5390 5391 /* Unrecognised? */ 5392 if (ctxt->d == 0) 5393 return EMULATION_FAILED; 5394 5395 ctxt->execute = opcode.u.execute; 5396 5397 if (unlikely(ctxt->ud) && likely(!(ctxt->d & EmulateOnUD))) 5398 return EMULATION_FAILED; 5399 5400 if (unlikely(ctxt->d & 5401 (NotImpl|Stack|Op3264|Sse|Mmx|Intercept|CheckPerm|NearBranch| 5402 No16))) { 5403 /* 5404 * These are copied unconditionally here, and checked unconditionally 5405 * in x86_emulate_insn. 5406 */ 5407 ctxt->check_perm = opcode.check_perm; 5408 ctxt->intercept = opcode.intercept; 5409 5410 if (ctxt->d & NotImpl) 5411 return EMULATION_FAILED; 5412 5413 if (mode == X86EMUL_MODE_PROT64) { 5414 if (ctxt->op_bytes == 4 && (ctxt->d & Stack)) 5415 ctxt->op_bytes = 8; 5416 else if (ctxt->d & NearBranch) 5417 ctxt->op_bytes = 8; 5418 } 5419 5420 if (ctxt->d & Op3264) { 5421 if (mode == X86EMUL_MODE_PROT64) 5422 ctxt->op_bytes = 8; 5423 else 5424 ctxt->op_bytes = 4; 5425 } 5426 5427 if ((ctxt->d & No16) && ctxt->op_bytes == 2) 5428 ctxt->op_bytes = 4; 5429 5430 if (ctxt->d & Sse) 5431 ctxt->op_bytes = 16; 5432 else if (ctxt->d & Mmx) 5433 ctxt->op_bytes = 8; 5434 } 5435 5436 /* ModRM and SIB bytes. */ 5437 if (ctxt->d & ModRM) { 5438 rc = decode_modrm(ctxt, &ctxt->memop); 5439 if (!has_seg_override) { 5440 has_seg_override = true; 5441 ctxt->seg_override = ctxt->modrm_seg; 5442 } 5443 } else if (ctxt->d & MemAbs) 5444 rc = decode_abs(ctxt, &ctxt->memop); 5445 if (rc != X86EMUL_CONTINUE) 5446 goto done; 5447 5448 if (!has_seg_override) 5449 ctxt->seg_override = VCPU_SREG_DS; 5450 5451 ctxt->memop.addr.mem.seg = ctxt->seg_override; 5452 5453 /* 5454 * Decode and fetch the source operand: register, memory 5455 * or immediate. 5456 */ 5457 rc = decode_operand(ctxt, &ctxt->src, (ctxt->d >> SrcShift) & OpMask); 5458 if (rc != X86EMUL_CONTINUE) 5459 goto done; 5460 5461 /* 5462 * Decode and fetch the second source operand: register, memory 5463 * or immediate. 5464 */ 5465 rc = decode_operand(ctxt, &ctxt->src2, (ctxt->d >> Src2Shift) & OpMask); 5466 if (rc != X86EMUL_CONTINUE) 5467 goto done; 5468 5469 /* Decode and fetch the destination operand: register or memory. */ 5470 rc = decode_operand(ctxt, &ctxt->dst, (ctxt->d >> DstShift) & OpMask); 5471 5472 if (ctxt->rip_relative && likely(ctxt->memopp)) 5473 ctxt->memopp->addr.mem.ea = address_mask(ctxt, 5474 ctxt->memopp->addr.mem.ea + ctxt->_eip); 5475 5476 done: 5477 if (rc == X86EMUL_PROPAGATE_FAULT) 5478 ctxt->have_exception = true; 5479 return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK; 5480 } 5481 5482 bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt) 5483 { 5484 return ctxt->d & PageTable; 5485 } 5486 5487 static bool string_insn_completed(struct x86_emulate_ctxt *ctxt) 5488 { 5489 /* The second termination condition only applies for REPE 5490 * and REPNE. Test if the repeat string operation prefix is 5491 * REPE/REPZ or REPNE/REPNZ and if it's the case it tests the 5492 * corresponding termination condition according to: 5493 * - if REPE/REPZ and ZF = 0 then done 5494 * - if REPNE/REPNZ and ZF = 1 then done 5495 */ 5496 if (((ctxt->b == 0xa6) || (ctxt->b == 0xa7) || 5497 (ctxt->b == 0xae) || (ctxt->b == 0xaf)) 5498 && (((ctxt->rep_prefix == REPE_PREFIX) && 5499 ((ctxt->eflags & X86_EFLAGS_ZF) == 0)) 5500 || ((ctxt->rep_prefix == REPNE_PREFIX) && 5501 ((ctxt->eflags & X86_EFLAGS_ZF) == X86_EFLAGS_ZF)))) 5502 return true; 5503 5504 return false; 5505 } 5506 5507 static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt) 5508 { 5509 int rc; 5510 5511 emulator_get_fpu(); 5512 rc = asm_safe("fwait"); 5513 emulator_put_fpu(); 5514 5515 if (unlikely(rc != X86EMUL_CONTINUE)) 5516 return emulate_exception(ctxt, MF_VECTOR, 0, false); 5517 5518 return X86EMUL_CONTINUE; 5519 } 5520 5521 static void fetch_possible_mmx_operand(struct operand *op) 5522 { 5523 if (op->type == OP_MM) 5524 read_mmx_reg(&op->mm_val, op->addr.mm); 5525 } 5526 5527 static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop) 5528 { 5529 ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF; 5530 5531 if (!(ctxt->d & ByteOp)) 5532 fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE; 5533 5534 asm("push %[flags]; popf; " CALL_NOSPEC " ; pushf; pop %[flags]\n" 5535 : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags), 5536 [thunk_target]"+S"(fop), ASM_CALL_CONSTRAINT 5537 : "c"(ctxt->src2.val)); 5538 5539 ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK); 5540 if (!fop) /* exception is returned in fop variable */ 5541 return emulate_de(ctxt); 5542 return X86EMUL_CONTINUE; 5543 } 5544 5545 void init_decode_cache(struct x86_emulate_ctxt *ctxt) 5546 { 5547 memset(&ctxt->rip_relative, 0, 5548 (void *)&ctxt->modrm - (void *)&ctxt->rip_relative); 5549 5550 ctxt->io_read.pos = 0; 5551 ctxt->io_read.end = 0; 5552 ctxt->mem_read.end = 0; 5553 } 5554 5555 int x86_emulate_insn(struct x86_emulate_ctxt *ctxt) 5556 { 5557 const struct x86_emulate_ops *ops = ctxt->ops; 5558 int rc = X86EMUL_CONTINUE; 5559 int saved_dst_type = ctxt->dst.type; 5560 unsigned emul_flags; 5561 5562 ctxt->mem_read.pos = 0; 5563 5564 /* LOCK prefix is allowed only with some instructions */ 5565 if (ctxt->lock_prefix && (!(ctxt->d & Lock) || ctxt->dst.type != OP_MEM)) { 5566 rc = emulate_ud(ctxt); 5567 goto done; 5568 } 5569 5570 if ((ctxt->d & SrcMask) == SrcMemFAddr && ctxt->src.type != OP_MEM) { 5571 rc = emulate_ud(ctxt); 5572 goto done; 5573 } 5574 5575 emul_flags = ctxt->ops->get_hflags(ctxt); 5576 if (unlikely(ctxt->d & 5577 (No64|Undefined|Sse|Mmx|Intercept|CheckPerm|Priv|Prot|String))) { 5578 if ((ctxt->mode == X86EMUL_MODE_PROT64 && (ctxt->d & No64)) || 5579 (ctxt->d & Undefined)) { 5580 rc = emulate_ud(ctxt); 5581 goto done; 5582 } 5583 5584 if (((ctxt->d & (Sse|Mmx)) && ((ops->get_cr(ctxt, 0) & X86_CR0_EM))) 5585 || ((ctxt->d & Sse) && !(ops->get_cr(ctxt, 4) & X86_CR4_OSFXSR))) { 5586 rc = emulate_ud(ctxt); 5587 goto done; 5588 } 5589 5590 if ((ctxt->d & (Sse|Mmx)) && (ops->get_cr(ctxt, 0) & X86_CR0_TS)) { 5591 rc = emulate_nm(ctxt); 5592 goto done; 5593 } 5594 5595 if (ctxt->d & Mmx) { 5596 rc = flush_pending_x87_faults(ctxt); 5597 if (rc != X86EMUL_CONTINUE) 5598 goto done; 5599 /* 5600 * Now that we know the fpu is exception safe, we can fetch 5601 * operands from it. 5602 */ 5603 fetch_possible_mmx_operand(&ctxt->src); 5604 fetch_possible_mmx_operand(&ctxt->src2); 5605 if (!(ctxt->d & Mov)) 5606 fetch_possible_mmx_operand(&ctxt->dst); 5607 } 5608 5609 if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && ctxt->intercept) { 5610 rc = emulator_check_intercept(ctxt, ctxt->intercept, 5611 X86_ICPT_PRE_EXCEPT); 5612 if (rc != X86EMUL_CONTINUE) 5613 goto done; 5614 } 5615 5616 /* Instruction can only be executed in protected mode */ 5617 if ((ctxt->d & Prot) && ctxt->mode < X86EMUL_MODE_PROT16) { 5618 rc = emulate_ud(ctxt); 5619 goto done; 5620 } 5621 5622 /* Privileged instruction can be executed only in CPL=0 */ 5623 if ((ctxt->d & Priv) && ops->cpl(ctxt)) { 5624 if (ctxt->d & PrivUD) 5625 rc = emulate_ud(ctxt); 5626 else 5627 rc = emulate_gp(ctxt, 0); 5628 goto done; 5629 } 5630 5631 /* Do instruction specific permission checks */ 5632 if (ctxt->d & CheckPerm) { 5633 rc = ctxt->check_perm(ctxt); 5634 if (rc != X86EMUL_CONTINUE) 5635 goto done; 5636 } 5637 5638 if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) { 5639 rc = emulator_check_intercept(ctxt, ctxt->intercept, 5640 X86_ICPT_POST_EXCEPT); 5641 if (rc != X86EMUL_CONTINUE) 5642 goto done; 5643 } 5644 5645 if (ctxt->rep_prefix && (ctxt->d & String)) { 5646 /* All REP prefixes have the same first termination condition */ 5647 if (address_mask(ctxt, reg_read(ctxt, VCPU_REGS_RCX)) == 0) { 5648 string_registers_quirk(ctxt); 5649 ctxt->eip = ctxt->_eip; 5650 ctxt->eflags &= ~X86_EFLAGS_RF; 5651 goto done; 5652 } 5653 } 5654 } 5655 5656 if ((ctxt->src.type == OP_MEM) && !(ctxt->d & NoAccess)) { 5657 rc = segmented_read(ctxt, ctxt->src.addr.mem, 5658 ctxt->src.valptr, ctxt->src.bytes); 5659 if (rc != X86EMUL_CONTINUE) 5660 goto done; 5661 ctxt->src.orig_val64 = ctxt->src.val64; 5662 } 5663 5664 if (ctxt->src2.type == OP_MEM) { 5665 rc = segmented_read(ctxt, ctxt->src2.addr.mem, 5666 &ctxt->src2.val, ctxt->src2.bytes); 5667 if (rc != X86EMUL_CONTINUE) 5668 goto done; 5669 } 5670 5671 if ((ctxt->d & DstMask) == ImplicitOps) 5672 goto special_insn; 5673 5674 5675 if ((ctxt->dst.type == OP_MEM) && !(ctxt->d & Mov)) { 5676 /* optimisation - avoid slow emulated read if Mov */ 5677 rc = segmented_read(ctxt, ctxt->dst.addr.mem, 5678 &ctxt->dst.val, ctxt->dst.bytes); 5679 if (rc != X86EMUL_CONTINUE) { 5680 if (!(ctxt->d & NoWrite) && 5681 rc == X86EMUL_PROPAGATE_FAULT && 5682 ctxt->exception.vector == PF_VECTOR) 5683 ctxt->exception.error_code |= PFERR_WRITE_MASK; 5684 goto done; 5685 } 5686 } 5687 /* Copy full 64-bit value for CMPXCHG8B. */ 5688 ctxt->dst.orig_val64 = ctxt->dst.val64; 5689 5690 special_insn: 5691 5692 if (unlikely(emul_flags & X86EMUL_GUEST_MASK) && (ctxt->d & Intercept)) { 5693 rc = emulator_check_intercept(ctxt, ctxt->intercept, 5694 X86_ICPT_POST_MEMACCESS); 5695 if (rc != X86EMUL_CONTINUE) 5696 goto done; 5697 } 5698 5699 if (ctxt->rep_prefix && (ctxt->d & String)) 5700 ctxt->eflags |= X86_EFLAGS_RF; 5701 else 5702 ctxt->eflags &= ~X86_EFLAGS_RF; 5703 5704 if (ctxt->execute) { 5705 if (ctxt->d & Fastop) 5706 rc = fastop(ctxt, ctxt->fop); 5707 else 5708 rc = ctxt->execute(ctxt); 5709 if (rc != X86EMUL_CONTINUE) 5710 goto done; 5711 goto writeback; 5712 } 5713 5714 if (ctxt->opcode_len == 2) 5715 goto twobyte_insn; 5716 else if (ctxt->opcode_len == 3) 5717 goto threebyte_insn; 5718 5719 switch (ctxt->b) { 5720 case 0x70 ... 0x7f: /* jcc (short) */ 5721 if (test_cc(ctxt->b, ctxt->eflags)) 5722 rc = jmp_rel(ctxt, ctxt->src.val); 5723 break; 5724 case 0x8d: /* lea r16/r32, m */ 5725 ctxt->dst.val = ctxt->src.addr.mem.ea; 5726 break; 5727 case 0x90 ... 0x97: /* nop / xchg reg, rax */ 5728 if (ctxt->dst.addr.reg == reg_rmw(ctxt, VCPU_REGS_RAX)) 5729 ctxt->dst.type = OP_NONE; 5730 else 5731 rc = em_xchg(ctxt); 5732 break; 5733 case 0x98: /* cbw/cwde/cdqe */ 5734 switch (ctxt->op_bytes) { 5735 case 2: ctxt->dst.val = (s8)ctxt->dst.val; break; 5736 case 4: ctxt->dst.val = (s16)ctxt->dst.val; break; 5737 case 8: ctxt->dst.val = (s32)ctxt->dst.val; break; 5738 } 5739 break; 5740 case 0xcc: /* int3 */ 5741 rc = emulate_int(ctxt, 3); 5742 break; 5743 case 0xcd: /* int n */ 5744 rc = emulate_int(ctxt, ctxt->src.val); 5745 break; 5746 case 0xce: /* into */ 5747 if (ctxt->eflags & X86_EFLAGS_OF) 5748 rc = emulate_int(ctxt, 4); 5749 break; 5750 case 0xe9: /* jmp rel */ 5751 case 0xeb: /* jmp rel short */ 5752 rc = jmp_rel(ctxt, ctxt->src.val); 5753 ctxt->dst.type = OP_NONE; /* Disable writeback. */ 5754 break; 5755 case 0xf4: /* hlt */ 5756 ctxt->ops->halt(ctxt); 5757 break; 5758 case 0xf5: /* cmc */ 5759 /* complement carry flag from eflags reg */ 5760 ctxt->eflags ^= X86_EFLAGS_CF; 5761 break; 5762 case 0xf8: /* clc */ 5763 ctxt->eflags &= ~X86_EFLAGS_CF; 5764 break; 5765 case 0xf9: /* stc */ 5766 ctxt->eflags |= X86_EFLAGS_CF; 5767 break; 5768 case 0xfc: /* cld */ 5769 ctxt->eflags &= ~X86_EFLAGS_DF; 5770 break; 5771 case 0xfd: /* std */ 5772 ctxt->eflags |= X86_EFLAGS_DF; 5773 break; 5774 default: 5775 goto cannot_emulate; 5776 } 5777 5778 if (rc != X86EMUL_CONTINUE) 5779 goto done; 5780 5781 writeback: 5782 if (ctxt->d & SrcWrite) { 5783 BUG_ON(ctxt->src.type == OP_MEM || ctxt->src.type == OP_MEM_STR); 5784 rc = writeback(ctxt, &ctxt->src); 5785 if (rc != X86EMUL_CONTINUE) 5786 goto done; 5787 } 5788 if (!(ctxt->d & NoWrite)) { 5789 rc = writeback(ctxt, &ctxt->dst); 5790 if (rc != X86EMUL_CONTINUE) 5791 goto done; 5792 } 5793 5794 /* 5795 * restore dst type in case the decoding will be reused 5796 * (happens for string instruction ) 5797 */ 5798 ctxt->dst.type = saved_dst_type; 5799 5800 if ((ctxt->d & SrcMask) == SrcSI) 5801 string_addr_inc(ctxt, VCPU_REGS_RSI, &ctxt->src); 5802 5803 if ((ctxt->d & DstMask) == DstDI) 5804 string_addr_inc(ctxt, VCPU_REGS_RDI, &ctxt->dst); 5805 5806 if (ctxt->rep_prefix && (ctxt->d & String)) { 5807 unsigned int count; 5808 struct read_cache *r = &ctxt->io_read; 5809 if ((ctxt->d & SrcMask) == SrcSI) 5810 count = ctxt->src.count; 5811 else 5812 count = ctxt->dst.count; 5813 register_address_increment(ctxt, VCPU_REGS_RCX, -count); 5814 5815 if (!string_insn_completed(ctxt)) { 5816 /* 5817 * Re-enter guest when pio read ahead buffer is empty 5818 * or, if it is not used, after each 1024 iteration. 5819 */ 5820 if ((r->end != 0 || reg_read(ctxt, VCPU_REGS_RCX) & 0x3ff) && 5821 (r->end == 0 || r->end != r->pos)) { 5822 /* 5823 * Reset read cache. Usually happens before 5824 * decode, but since instruction is restarted 5825 * we have to do it here. 5826 */ 5827 ctxt->mem_read.end = 0; 5828 writeback_registers(ctxt); 5829 return EMULATION_RESTART; 5830 } 5831 goto done; /* skip rip writeback */ 5832 } 5833 ctxt->eflags &= ~X86_EFLAGS_RF; 5834 } 5835 5836 ctxt->eip = ctxt->_eip; 5837 if (ctxt->mode != X86EMUL_MODE_PROT64) 5838 ctxt->eip = (u32)ctxt->_eip; 5839 5840 done: 5841 if (rc == X86EMUL_PROPAGATE_FAULT) { 5842 WARN_ON(ctxt->exception.vector > 0x1f); 5843 ctxt->have_exception = true; 5844 } 5845 if (rc == X86EMUL_INTERCEPTED) 5846 return EMULATION_INTERCEPTED; 5847 5848 if (rc == X86EMUL_CONTINUE) 5849 writeback_registers(ctxt); 5850 5851 return (rc == X86EMUL_UNHANDLEABLE) ? EMULATION_FAILED : EMULATION_OK; 5852 5853 twobyte_insn: 5854 switch (ctxt->b) { 5855 case 0x09: /* wbinvd */ 5856 (ctxt->ops->wbinvd)(ctxt); 5857 break; 5858 case 0x08: /* invd */ 5859 case 0x0d: /* GrpP (prefetch) */ 5860 case 0x18: /* Grp16 (prefetch/nop) */ 5861 case 0x1f: /* nop */ 5862 break; 5863 case 0x20: /* mov cr, reg */ 5864 ctxt->dst.val = ops->get_cr(ctxt, ctxt->modrm_reg); 5865 break; 5866 case 0x21: /* mov from dr to reg */ 5867 ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val); 5868 break; 5869 case 0x40 ... 0x4f: /* cmov */ 5870 if (test_cc(ctxt->b, ctxt->eflags)) 5871 ctxt->dst.val = ctxt->src.val; 5872 else if (ctxt->op_bytes != 4) 5873 ctxt->dst.type = OP_NONE; /* no writeback */ 5874 break; 5875 case 0x80 ... 0x8f: /* jnz rel, etc*/ 5876 if (test_cc(ctxt->b, ctxt->eflags)) 5877 rc = jmp_rel(ctxt, ctxt->src.val); 5878 break; 5879 case 0x90 ... 0x9f: /* setcc r/m8 */ 5880 ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags); 5881 break; 5882 case 0xb6 ... 0xb7: /* movzx */ 5883 ctxt->dst.bytes = ctxt->op_bytes; 5884 ctxt->dst.val = (ctxt->src.bytes == 1) ? (u8) ctxt->src.val 5885 : (u16) ctxt->src.val; 5886 break; 5887 case 0xbe ... 0xbf: /* movsx */ 5888 ctxt->dst.bytes = ctxt->op_bytes; 5889 ctxt->dst.val = (ctxt->src.bytes == 1) ? (s8) ctxt->src.val : 5890 (s16) ctxt->src.val; 5891 break; 5892 default: 5893 goto cannot_emulate; 5894 } 5895 5896 threebyte_insn: 5897 5898 if (rc != X86EMUL_CONTINUE) 5899 goto done; 5900 5901 goto writeback; 5902 5903 cannot_emulate: 5904 return EMULATION_FAILED; 5905 } 5906 5907 void emulator_invalidate_register_cache(struct x86_emulate_ctxt *ctxt) 5908 { 5909 invalidate_registers(ctxt); 5910 } 5911 5912 void emulator_writeback_register_cache(struct x86_emulate_ctxt *ctxt) 5913 { 5914 writeback_registers(ctxt); 5915 } 5916 5917 bool emulator_can_use_gpa(struct x86_emulate_ctxt *ctxt) 5918 { 5919 if (ctxt->rep_prefix && (ctxt->d & String)) 5920 return false; 5921 5922 if (ctxt->d & TwoMemOp) 5923 return false; 5924 5925 return true; 5926 } 5927