1 //===-- X86Disassembler.cpp - Disassembler for x86 and x86_64 -------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file is part of the X86 Disassembler.
11 // It contains code to translate the data produced by the decoder into
12 // MCInsts.
13 // Documentation for the disassembler can be found in X86Disassembler.h.
14 //
15 //===----------------------------------------------------------------------===//
16
17 #include "X86Disassembler.h"
18 #include "X86DisassemblerDecoder.h"
19
20 #include "llvm/MC/EDInstInfo.h"
21 #include "llvm/MC/MCExpr.h"
22 #include "llvm/MC/MCContext.h"
23 #include "llvm/MC/MCDisassembler.h"
24 #include "llvm/MC/MCInst.h"
25 #include "llvm/MC/MCInstrInfo.h"
26 #include "llvm/MC/MCSubtargetInfo.h"
27 #include "llvm/Support/Debug.h"
28 #include "llvm/Support/MemoryObject.h"
29 #include "llvm/Support/TargetRegistry.h"
30 #include "llvm/Support/raw_ostream.h"
31
32 #define GET_REGINFO_ENUM
33 #include "X86GenRegisterInfo.inc"
34 #define GET_INSTRINFO_ENUM
35 #include "X86GenInstrInfo.inc"
36 #include "X86GenEDInfo.inc"
37
38 using namespace llvm;
39 using namespace llvm::X86Disassembler;
40
x86DisassemblerDebug(const char * file,unsigned line,const char * s)41 void x86DisassemblerDebug(const char *file,
42 unsigned line,
43 const char *s) {
44 dbgs() << file << ":" << line << ": " << s;
45 }
46
x86DisassemblerGetInstrName(unsigned Opcode,const void * mii)47 const char *x86DisassemblerGetInstrName(unsigned Opcode, const void *mii) {
48 const MCInstrInfo *MII = static_cast<const MCInstrInfo *>(mii);
49 return MII->getName(Opcode);
50 }
51
52 #define debug(s) DEBUG(x86DisassemblerDebug(__FILE__, __LINE__, s));
53
54 namespace llvm {
55
56 // Fill-ins to make the compiler happy. These constants are never actually
57 // assigned; they are just filler to make an automatically-generated switch
58 // statement work.
59 namespace X86 {
60 enum {
61 BX_SI = 500,
62 BX_DI = 501,
63 BP_SI = 502,
64 BP_DI = 503,
65 sib = 504,
66 sib64 = 505
67 };
68 }
69
70 extern Target TheX86_32Target, TheX86_64Target;
71
72 }
73
74 static bool translateInstruction(MCInst &target,
75 InternalInstruction &source,
76 const MCDisassembler *Dis);
77
X86GenericDisassembler(const MCSubtargetInfo & STI,DisassemblerMode mode,const MCInstrInfo * MII)78 X86GenericDisassembler::X86GenericDisassembler(const MCSubtargetInfo &STI,
79 DisassemblerMode mode,
80 const MCInstrInfo *MII)
81 : MCDisassembler(STI), MII(MII), fMode(mode) {}
82
~X86GenericDisassembler()83 X86GenericDisassembler::~X86GenericDisassembler() {
84 delete MII;
85 }
86
getEDInfo() const87 const EDInstInfo *X86GenericDisassembler::getEDInfo() const {
88 return instInfoX86;
89 }
90
91 /// regionReader - a callback function that wraps the readByte method from
92 /// MemoryObject.
93 ///
94 /// @param arg - The generic callback parameter. In this case, this should
95 /// be a pointer to a MemoryObject.
96 /// @param byte - A pointer to the byte to be read.
97 /// @param address - The address to be read.
regionReader(const void * arg,uint8_t * byte,uint64_t address)98 static int regionReader(const void* arg, uint8_t* byte, uint64_t address) {
99 const MemoryObject* region = static_cast<const MemoryObject*>(arg);
100 return region->readByte(address, byte);
101 }
102
103 /// logger - a callback function that wraps the operator<< method from
104 /// raw_ostream.
105 ///
106 /// @param arg - The generic callback parameter. This should be a pointe
107 /// to a raw_ostream.
108 /// @param log - A string to be logged. logger() adds a newline.
logger(void * arg,const char * log)109 static void logger(void* arg, const char* log) {
110 if (!arg)
111 return;
112
113 raw_ostream &vStream = *(static_cast<raw_ostream*>(arg));
114 vStream << log << "\n";
115 }
116
117 //
118 // Public interface for the disassembler
119 //
120
121 MCDisassembler::DecodeStatus
getInstruction(MCInst & instr,uint64_t & size,const MemoryObject & region,uint64_t address,raw_ostream & vStream,raw_ostream & cStream) const122 X86GenericDisassembler::getInstruction(MCInst &instr,
123 uint64_t &size,
124 const MemoryObject ®ion,
125 uint64_t address,
126 raw_ostream &vStream,
127 raw_ostream &cStream) const {
128 CommentStream = &cStream;
129
130 InternalInstruction internalInstr;
131
132 dlog_t loggerFn = logger;
133 if (&vStream == &nulls())
134 loggerFn = 0; // Disable logging completely if it's going to nulls().
135
136 int ret = decodeInstruction(&internalInstr,
137 regionReader,
138 (const void*)®ion,
139 loggerFn,
140 (void*)&vStream,
141 (const void*)MII,
142 address,
143 fMode);
144
145 if (ret) {
146 size = internalInstr.readerCursor - address;
147 return Fail;
148 }
149 else {
150 size = internalInstr.length;
151 return (!translateInstruction(instr, internalInstr, this)) ?
152 Success : Fail;
153 }
154 }
155
156 //
157 // Private code that translates from struct InternalInstructions to MCInsts.
158 //
159
160 /// translateRegister - Translates an internal register to the appropriate LLVM
161 /// register, and appends it as an operand to an MCInst.
162 ///
163 /// @param mcInst - The MCInst to append to.
164 /// @param reg - The Reg to append.
translateRegister(MCInst & mcInst,Reg reg)165 static void translateRegister(MCInst &mcInst, Reg reg) {
166 #define ENTRY(x) X86::x,
167 uint8_t llvmRegnums[] = {
168 ALL_REGS
169 0
170 };
171 #undef ENTRY
172
173 uint8_t llvmRegnum = llvmRegnums[reg];
174 mcInst.addOperand(MCOperand::CreateReg(llvmRegnum));
175 }
176
177 /// tryAddingSymbolicOperand - trys to add a symbolic operand in place of the
178 /// immediate Value in the MCInst.
179 ///
180 /// @param Value - The immediate Value, has had any PC adjustment made by
181 /// the caller.
182 /// @param isBranch - If the instruction is a branch instruction
183 /// @param Address - The starting address of the instruction
184 /// @param Offset - The byte offset to this immediate in the instruction
185 /// @param Width - The byte width of this immediate in the instruction
186 ///
187 /// If the getOpInfo() function was set when setupForSymbolicDisassembly() was
188 /// called then that function is called to get any symbolic information for the
189 /// immediate in the instruction using the Address, Offset and Width. If that
190 /// returns non-zero then the symbolic information it returns is used to create
191 /// an MCExpr and that is added as an operand to the MCInst. If getOpInfo()
192 /// returns zero and isBranch is true then a symbol look up for immediate Value
193 /// is done and if a symbol is found an MCExpr is created with that, else
194 /// an MCExpr with the immediate Value is created. This function returns true
195 /// if it adds an operand to the MCInst and false otherwise.
tryAddingSymbolicOperand(int64_t Value,bool isBranch,uint64_t Address,uint64_t Offset,uint64_t Width,MCInst & MI,const MCDisassembler * Dis)196 static bool tryAddingSymbolicOperand(int64_t Value, bool isBranch,
197 uint64_t Address, uint64_t Offset,
198 uint64_t Width, MCInst &MI,
199 const MCDisassembler *Dis) {
200 LLVMOpInfoCallback getOpInfo = Dis->getLLVMOpInfoCallback();
201 struct LLVMOpInfo1 SymbolicOp;
202 memset(&SymbolicOp, '\0', sizeof(struct LLVMOpInfo1));
203 SymbolicOp.Value = Value;
204 void *DisInfo = Dis->getDisInfoBlock();
205
206 if (!getOpInfo ||
207 !getOpInfo(DisInfo, Address, Offset, Width, 1, &SymbolicOp)) {
208 // Clear SymbolicOp.Value from above and also all other fields.
209 memset(&SymbolicOp, '\0', sizeof(struct LLVMOpInfo1));
210 LLVMSymbolLookupCallback SymbolLookUp = Dis->getLLVMSymbolLookupCallback();
211 if (!SymbolLookUp)
212 return false;
213 uint64_t ReferenceType;
214 if (isBranch)
215 ReferenceType = LLVMDisassembler_ReferenceType_In_Branch;
216 else
217 ReferenceType = LLVMDisassembler_ReferenceType_InOut_None;
218 const char *ReferenceName;
219 const char *Name = SymbolLookUp(DisInfo, Value, &ReferenceType, Address,
220 &ReferenceName);
221 if (Name) {
222 SymbolicOp.AddSymbol.Name = Name;
223 SymbolicOp.AddSymbol.Present = true;
224 }
225 // For branches always create an MCExpr so it gets printed as hex address.
226 else if (isBranch) {
227 SymbolicOp.Value = Value;
228 }
229 if(ReferenceType == LLVMDisassembler_ReferenceType_Out_SymbolStub)
230 (*Dis->CommentStream) << "symbol stub for: " << ReferenceName;
231 if (!Name && !isBranch)
232 return false;
233 }
234
235 MCContext *Ctx = Dis->getMCContext();
236 const MCExpr *Add = NULL;
237 if (SymbolicOp.AddSymbol.Present) {
238 if (SymbolicOp.AddSymbol.Name) {
239 StringRef Name(SymbolicOp.AddSymbol.Name);
240 MCSymbol *Sym = Ctx->GetOrCreateSymbol(Name);
241 Add = MCSymbolRefExpr::Create(Sym, *Ctx);
242 } else {
243 Add = MCConstantExpr::Create((int)SymbolicOp.AddSymbol.Value, *Ctx);
244 }
245 }
246
247 const MCExpr *Sub = NULL;
248 if (SymbolicOp.SubtractSymbol.Present) {
249 if (SymbolicOp.SubtractSymbol.Name) {
250 StringRef Name(SymbolicOp.SubtractSymbol.Name);
251 MCSymbol *Sym = Ctx->GetOrCreateSymbol(Name);
252 Sub = MCSymbolRefExpr::Create(Sym, *Ctx);
253 } else {
254 Sub = MCConstantExpr::Create((int)SymbolicOp.SubtractSymbol.Value, *Ctx);
255 }
256 }
257
258 const MCExpr *Off = NULL;
259 if (SymbolicOp.Value != 0)
260 Off = MCConstantExpr::Create(SymbolicOp.Value, *Ctx);
261
262 const MCExpr *Expr;
263 if (Sub) {
264 const MCExpr *LHS;
265 if (Add)
266 LHS = MCBinaryExpr::CreateSub(Add, Sub, *Ctx);
267 else
268 LHS = MCUnaryExpr::CreateMinus(Sub, *Ctx);
269 if (Off != 0)
270 Expr = MCBinaryExpr::CreateAdd(LHS, Off, *Ctx);
271 else
272 Expr = LHS;
273 } else if (Add) {
274 if (Off != 0)
275 Expr = MCBinaryExpr::CreateAdd(Add, Off, *Ctx);
276 else
277 Expr = Add;
278 } else {
279 if (Off != 0)
280 Expr = Off;
281 else
282 Expr = MCConstantExpr::Create(0, *Ctx);
283 }
284
285 MI.addOperand(MCOperand::CreateExpr(Expr));
286
287 return true;
288 }
289
290 /// tryAddingPcLoadReferenceComment - trys to add a comment as to what is being
291 /// referenced by a load instruction with the base register that is the rip.
292 /// These can often be addresses in a literal pool. The Address of the
293 /// instruction and its immediate Value are used to determine the address
294 /// being referenced in the literal pool entry. The SymbolLookUp call back will
295 /// return a pointer to a literal 'C' string if the referenced address is an
296 /// address into a section with 'C' string literals.
tryAddingPcLoadReferenceComment(uint64_t Address,uint64_t Value,const void * Decoder)297 static void tryAddingPcLoadReferenceComment(uint64_t Address, uint64_t Value,
298 const void *Decoder) {
299 const MCDisassembler *Dis = static_cast<const MCDisassembler*>(Decoder);
300 LLVMSymbolLookupCallback SymbolLookUp = Dis->getLLVMSymbolLookupCallback();
301 if (SymbolLookUp) {
302 void *DisInfo = Dis->getDisInfoBlock();
303 uint64_t ReferenceType = LLVMDisassembler_ReferenceType_In_PCrel_Load;
304 const char *ReferenceName;
305 (void)SymbolLookUp(DisInfo, Value, &ReferenceType, Address, &ReferenceName);
306 if(ReferenceType == LLVMDisassembler_ReferenceType_Out_LitPool_CstrAddr)
307 (*Dis->CommentStream) << "literal pool for: " << ReferenceName;
308 }
309 }
310
311 /// translateImmediate - Appends an immediate operand to an MCInst.
312 ///
313 /// @param mcInst - The MCInst to append to.
314 /// @param immediate - The immediate value to append.
315 /// @param operand - The operand, as stored in the descriptor table.
316 /// @param insn - The internal instruction.
translateImmediate(MCInst & mcInst,uint64_t immediate,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)317 static void translateImmediate(MCInst &mcInst, uint64_t immediate,
318 const OperandSpecifier &operand,
319 InternalInstruction &insn,
320 const MCDisassembler *Dis) {
321 // Sign-extend the immediate if necessary.
322
323 OperandType type = (OperandType)operand.type;
324
325 bool isBranch = false;
326 uint64_t pcrel = 0;
327 if (type == TYPE_RELv) {
328 isBranch = true;
329 pcrel = insn.startLocation +
330 insn.immediateOffset + insn.immediateSize;
331 switch (insn.displacementSize) {
332 default:
333 break;
334 case 1:
335 type = TYPE_MOFFS8;
336 break;
337 case 2:
338 type = TYPE_MOFFS16;
339 break;
340 case 4:
341 type = TYPE_MOFFS32;
342 break;
343 case 8:
344 type = TYPE_MOFFS64;
345 break;
346 }
347 }
348 // By default sign-extend all X86 immediates based on their encoding.
349 else if (type == TYPE_IMM8 || type == TYPE_IMM16 || type == TYPE_IMM32 ||
350 type == TYPE_IMM64) {
351 uint32_t Opcode = mcInst.getOpcode();
352 switch (operand.encoding) {
353 default:
354 break;
355 case ENCODING_IB:
356 // Special case those X86 instructions that use the imm8 as a set of
357 // bits, bit count, etc. and are not sign-extend.
358 if (Opcode != X86::BLENDPSrri && Opcode != X86::BLENDPDrri &&
359 Opcode != X86::PBLENDWrri && Opcode != X86::MPSADBWrri &&
360 Opcode != X86::DPPSrri && Opcode != X86::DPPDrri &&
361 Opcode != X86::INSERTPSrr && Opcode != X86::VBLENDPSYrri &&
362 Opcode != X86::VBLENDPSYrmi && Opcode != X86::VBLENDPDYrri &&
363 Opcode != X86::VBLENDPDYrmi && Opcode != X86::VPBLENDWrri &&
364 Opcode != X86::VMPSADBWrri && Opcode != X86::VDPPSYrri &&
365 Opcode != X86::VDPPSYrmi && Opcode != X86::VDPPDrri &&
366 Opcode != X86::VINSERTPSrr)
367 type = TYPE_MOFFS8;
368 break;
369 case ENCODING_IW:
370 type = TYPE_MOFFS16;
371 break;
372 case ENCODING_ID:
373 type = TYPE_MOFFS32;
374 break;
375 case ENCODING_IO:
376 type = TYPE_MOFFS64;
377 break;
378 }
379 }
380
381 switch (type) {
382 case TYPE_XMM32:
383 case TYPE_XMM64:
384 case TYPE_XMM128:
385 mcInst.addOperand(MCOperand::CreateReg(X86::XMM0 + (immediate >> 4)));
386 return;
387 case TYPE_XMM256:
388 mcInst.addOperand(MCOperand::CreateReg(X86::YMM0 + (immediate >> 4)));
389 return;
390 case TYPE_REL8:
391 isBranch = true;
392 pcrel = insn.startLocation + insn.immediateOffset + insn.immediateSize;
393 // fall through to sign extend the immediate if needed.
394 case TYPE_MOFFS8:
395 if(immediate & 0x80)
396 immediate |= ~(0xffull);
397 break;
398 case TYPE_MOFFS16:
399 if(immediate & 0x8000)
400 immediate |= ~(0xffffull);
401 break;
402 case TYPE_REL32:
403 case TYPE_REL64:
404 isBranch = true;
405 pcrel = insn.startLocation + insn.immediateOffset + insn.immediateSize;
406 // fall through to sign extend the immediate if needed.
407 case TYPE_MOFFS32:
408 if(immediate & 0x80000000)
409 immediate |= ~(0xffffffffull);
410 break;
411 case TYPE_MOFFS64:
412 default:
413 // operand is 64 bits wide. Do nothing.
414 break;
415 }
416
417 if(!tryAddingSymbolicOperand(immediate + pcrel, isBranch, insn.startLocation,
418 insn.immediateOffset, insn.immediateSize,
419 mcInst, Dis))
420 mcInst.addOperand(MCOperand::CreateImm(immediate));
421 }
422
423 /// translateRMRegister - Translates a register stored in the R/M field of the
424 /// ModR/M byte to its LLVM equivalent and appends it to an MCInst.
425 /// @param mcInst - The MCInst to append to.
426 /// @param insn - The internal instruction to extract the R/M field
427 /// from.
428 /// @return - 0 on success; -1 otherwise
translateRMRegister(MCInst & mcInst,InternalInstruction & insn)429 static bool translateRMRegister(MCInst &mcInst,
430 InternalInstruction &insn) {
431 if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
432 debug("A R/M register operand may not have a SIB byte");
433 return true;
434 }
435
436 switch (insn.eaBase) {
437 default:
438 debug("Unexpected EA base register");
439 return true;
440 case EA_BASE_NONE:
441 debug("EA_BASE_NONE for ModR/M base");
442 return true;
443 #define ENTRY(x) case EA_BASE_##x:
444 ALL_EA_BASES
445 #undef ENTRY
446 debug("A R/M register operand may not have a base; "
447 "the operand must be a register.");
448 return true;
449 #define ENTRY(x) \
450 case EA_REG_##x: \
451 mcInst.addOperand(MCOperand::CreateReg(X86::x)); break;
452 ALL_REGS
453 #undef ENTRY
454 }
455
456 return false;
457 }
458
459 /// translateRMMemory - Translates a memory operand stored in the Mod and R/M
460 /// fields of an internal instruction (and possibly its SIB byte) to a memory
461 /// operand in LLVM's format, and appends it to an MCInst.
462 ///
463 /// @param mcInst - The MCInst to append to.
464 /// @param insn - The instruction to extract Mod, R/M, and SIB fields
465 /// from.
466 /// @return - 0 on success; nonzero otherwise
translateRMMemory(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis)467 static bool translateRMMemory(MCInst &mcInst, InternalInstruction &insn,
468 const MCDisassembler *Dis) {
469 // Addresses in an MCInst are represented as five operands:
470 // 1. basereg (register) The R/M base, or (if there is a SIB) the
471 // SIB base
472 // 2. scaleamount (immediate) 1, or (if there is a SIB) the specified
473 // scale amount
474 // 3. indexreg (register) x86_registerNONE, or (if there is a SIB)
475 // the index (which is multiplied by the
476 // scale amount)
477 // 4. displacement (immediate) 0, or the displacement if there is one
478 // 5. segmentreg (register) x86_registerNONE for now, but could be set
479 // if we have segment overrides
480
481 MCOperand baseReg;
482 MCOperand scaleAmount;
483 MCOperand indexReg;
484 MCOperand displacement;
485 MCOperand segmentReg;
486 uint64_t pcrel = 0;
487
488 if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
489 if (insn.sibBase != SIB_BASE_NONE) {
490 switch (insn.sibBase) {
491 default:
492 debug("Unexpected sibBase");
493 return true;
494 #define ENTRY(x) \
495 case SIB_BASE_##x: \
496 baseReg = MCOperand::CreateReg(X86::x); break;
497 ALL_SIB_BASES
498 #undef ENTRY
499 }
500 } else {
501 baseReg = MCOperand::CreateReg(0);
502 }
503
504 // Check whether we are handling VSIB addressing mode for GATHER.
505 // If sibIndex was set to SIB_INDEX_NONE, index offset is 4 and
506 // we should use SIB_INDEX_XMM4|YMM4 for VSIB.
507 // I don't see a way to get the correct IndexReg in readSIB:
508 // We can tell whether it is VSIB or SIB after instruction ID is decoded,
509 // but instruction ID may not be decoded yet when calling readSIB.
510 uint32_t Opcode = mcInst.getOpcode();
511 bool IndexIs128 = (Opcode == X86::VGATHERDPDrm ||
512 Opcode == X86::VGATHERDPDYrm ||
513 Opcode == X86::VGATHERQPDrm ||
514 Opcode == X86::VGATHERDPSrm ||
515 Opcode == X86::VGATHERQPSrm ||
516 Opcode == X86::VPGATHERDQrm ||
517 Opcode == X86::VPGATHERDQYrm ||
518 Opcode == X86::VPGATHERQQrm ||
519 Opcode == X86::VPGATHERDDrm ||
520 Opcode == X86::VPGATHERQDrm);
521 bool IndexIs256 = (Opcode == X86::VGATHERQPDYrm ||
522 Opcode == X86::VGATHERDPSYrm ||
523 Opcode == X86::VGATHERQPSYrm ||
524 Opcode == X86::VPGATHERQQYrm ||
525 Opcode == X86::VPGATHERDDYrm ||
526 Opcode == X86::VPGATHERQDYrm);
527 if (IndexIs128 || IndexIs256) {
528 unsigned IndexOffset = insn.sibIndex -
529 (insn.addressSize == 8 ? SIB_INDEX_RAX:SIB_INDEX_EAX);
530 SIBIndex IndexBase = IndexIs256 ? SIB_INDEX_YMM0 : SIB_INDEX_XMM0;
531 insn.sibIndex = (SIBIndex)(IndexBase +
532 (insn.sibIndex == SIB_INDEX_NONE ? 4 : IndexOffset));
533 }
534
535 if (insn.sibIndex != SIB_INDEX_NONE) {
536 switch (insn.sibIndex) {
537 default:
538 debug("Unexpected sibIndex");
539 return true;
540 #define ENTRY(x) \
541 case SIB_INDEX_##x: \
542 indexReg = MCOperand::CreateReg(X86::x); break;
543 EA_BASES_32BIT
544 EA_BASES_64BIT
545 REGS_XMM
546 REGS_YMM
547 #undef ENTRY
548 }
549 } else {
550 indexReg = MCOperand::CreateReg(0);
551 }
552
553 scaleAmount = MCOperand::CreateImm(insn.sibScale);
554 } else {
555 switch (insn.eaBase) {
556 case EA_BASE_NONE:
557 if (insn.eaDisplacement == EA_DISP_NONE) {
558 debug("EA_BASE_NONE and EA_DISP_NONE for ModR/M base");
559 return true;
560 }
561 if (insn.mode == MODE_64BIT){
562 pcrel = insn.startLocation +
563 insn.displacementOffset + insn.displacementSize;
564 tryAddingPcLoadReferenceComment(insn.startLocation +
565 insn.displacementOffset,
566 insn.displacement + pcrel, Dis);
567 baseReg = MCOperand::CreateReg(X86::RIP); // Section 2.2.1.6
568 }
569 else
570 baseReg = MCOperand::CreateReg(0);
571
572 indexReg = MCOperand::CreateReg(0);
573 break;
574 case EA_BASE_BX_SI:
575 baseReg = MCOperand::CreateReg(X86::BX);
576 indexReg = MCOperand::CreateReg(X86::SI);
577 break;
578 case EA_BASE_BX_DI:
579 baseReg = MCOperand::CreateReg(X86::BX);
580 indexReg = MCOperand::CreateReg(X86::DI);
581 break;
582 case EA_BASE_BP_SI:
583 baseReg = MCOperand::CreateReg(X86::BP);
584 indexReg = MCOperand::CreateReg(X86::SI);
585 break;
586 case EA_BASE_BP_DI:
587 baseReg = MCOperand::CreateReg(X86::BP);
588 indexReg = MCOperand::CreateReg(X86::DI);
589 break;
590 default:
591 indexReg = MCOperand::CreateReg(0);
592 switch (insn.eaBase) {
593 default:
594 debug("Unexpected eaBase");
595 return true;
596 // Here, we will use the fill-ins defined above. However,
597 // BX_SI, BX_DI, BP_SI, and BP_DI are all handled above and
598 // sib and sib64 were handled in the top-level if, so they're only
599 // placeholders to keep the compiler happy.
600 #define ENTRY(x) \
601 case EA_BASE_##x: \
602 baseReg = MCOperand::CreateReg(X86::x); break;
603 ALL_EA_BASES
604 #undef ENTRY
605 #define ENTRY(x) case EA_REG_##x:
606 ALL_REGS
607 #undef ENTRY
608 debug("A R/M memory operand may not be a register; "
609 "the base field must be a base.");
610 return true;
611 }
612 }
613
614 scaleAmount = MCOperand::CreateImm(1);
615 }
616
617 displacement = MCOperand::CreateImm(insn.displacement);
618
619 static const uint8_t segmentRegnums[SEG_OVERRIDE_max] = {
620 0, // SEG_OVERRIDE_NONE
621 X86::CS,
622 X86::SS,
623 X86::DS,
624 X86::ES,
625 X86::FS,
626 X86::GS
627 };
628
629 segmentReg = MCOperand::CreateReg(segmentRegnums[insn.segmentOverride]);
630
631 mcInst.addOperand(baseReg);
632 mcInst.addOperand(scaleAmount);
633 mcInst.addOperand(indexReg);
634 if(!tryAddingSymbolicOperand(insn.displacement + pcrel, false,
635 insn.startLocation, insn.displacementOffset,
636 insn.displacementSize, mcInst, Dis))
637 mcInst.addOperand(displacement);
638 mcInst.addOperand(segmentReg);
639 return false;
640 }
641
642 /// translateRM - Translates an operand stored in the R/M (and possibly SIB)
643 /// byte of an instruction to LLVM form, and appends it to an MCInst.
644 ///
645 /// @param mcInst - The MCInst to append to.
646 /// @param operand - The operand, as stored in the descriptor table.
647 /// @param insn - The instruction to extract Mod, R/M, and SIB fields
648 /// from.
649 /// @return - 0 on success; nonzero otherwise
translateRM(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)650 static bool translateRM(MCInst &mcInst, const OperandSpecifier &operand,
651 InternalInstruction &insn, const MCDisassembler *Dis) {
652 switch (operand.type) {
653 default:
654 debug("Unexpected type for a R/M operand");
655 return true;
656 case TYPE_R8:
657 case TYPE_R16:
658 case TYPE_R32:
659 case TYPE_R64:
660 case TYPE_Rv:
661 case TYPE_MM:
662 case TYPE_MM32:
663 case TYPE_MM64:
664 case TYPE_XMM:
665 case TYPE_XMM32:
666 case TYPE_XMM64:
667 case TYPE_XMM128:
668 case TYPE_XMM256:
669 case TYPE_DEBUGREG:
670 case TYPE_CONTROLREG:
671 return translateRMRegister(mcInst, insn);
672 case TYPE_M:
673 case TYPE_M8:
674 case TYPE_M16:
675 case TYPE_M32:
676 case TYPE_M64:
677 case TYPE_M128:
678 case TYPE_M256:
679 case TYPE_M512:
680 case TYPE_Mv:
681 case TYPE_M32FP:
682 case TYPE_M64FP:
683 case TYPE_M80FP:
684 case TYPE_M16INT:
685 case TYPE_M32INT:
686 case TYPE_M64INT:
687 case TYPE_M1616:
688 case TYPE_M1632:
689 case TYPE_M1664:
690 case TYPE_LEA:
691 return translateRMMemory(mcInst, insn, Dis);
692 }
693 }
694
695 /// translateFPRegister - Translates a stack position on the FPU stack to its
696 /// LLVM form, and appends it to an MCInst.
697 ///
698 /// @param mcInst - The MCInst to append to.
699 /// @param stackPos - The stack position to translate.
700 /// @return - 0 on success; nonzero otherwise.
translateFPRegister(MCInst & mcInst,uint8_t stackPos)701 static bool translateFPRegister(MCInst &mcInst,
702 uint8_t stackPos) {
703 if (stackPos >= 8) {
704 debug("Invalid FP stack position");
705 return true;
706 }
707
708 mcInst.addOperand(MCOperand::CreateReg(X86::ST0 + stackPos));
709
710 return false;
711 }
712
713 /// translateOperand - Translates an operand stored in an internal instruction
714 /// to LLVM's format and appends it to an MCInst.
715 ///
716 /// @param mcInst - The MCInst to append to.
717 /// @param operand - The operand, as stored in the descriptor table.
718 /// @param insn - The internal instruction.
719 /// @return - false on success; true otherwise.
translateOperand(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)720 static bool translateOperand(MCInst &mcInst, const OperandSpecifier &operand,
721 InternalInstruction &insn,
722 const MCDisassembler *Dis) {
723 switch (operand.encoding) {
724 default:
725 debug("Unhandled operand encoding during translation");
726 return true;
727 case ENCODING_REG:
728 translateRegister(mcInst, insn.reg);
729 return false;
730 case ENCODING_RM:
731 return translateRM(mcInst, operand, insn, Dis);
732 case ENCODING_CB:
733 case ENCODING_CW:
734 case ENCODING_CD:
735 case ENCODING_CP:
736 case ENCODING_CO:
737 case ENCODING_CT:
738 debug("Translation of code offsets isn't supported.");
739 return true;
740 case ENCODING_IB:
741 case ENCODING_IW:
742 case ENCODING_ID:
743 case ENCODING_IO:
744 case ENCODING_Iv:
745 case ENCODING_Ia:
746 translateImmediate(mcInst,
747 insn.immediates[insn.numImmediatesTranslated++],
748 operand,
749 insn,
750 Dis);
751 return false;
752 case ENCODING_RB:
753 case ENCODING_RW:
754 case ENCODING_RD:
755 case ENCODING_RO:
756 translateRegister(mcInst, insn.opcodeRegister);
757 return false;
758 case ENCODING_I:
759 return translateFPRegister(mcInst, insn.opcodeModifier);
760 case ENCODING_Rv:
761 translateRegister(mcInst, insn.opcodeRegister);
762 return false;
763 case ENCODING_VVVV:
764 translateRegister(mcInst, insn.vvvv);
765 return false;
766 case ENCODING_DUP:
767 return translateOperand(mcInst, insn.operands[operand.type - TYPE_DUP0],
768 insn, Dis);
769 }
770 }
771
772 /// translateInstruction - Translates an internal instruction and all its
773 /// operands to an MCInst.
774 ///
775 /// @param mcInst - The MCInst to populate with the instruction's data.
776 /// @param insn - The internal instruction.
777 /// @return - false on success; true otherwise.
translateInstruction(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis)778 static bool translateInstruction(MCInst &mcInst,
779 InternalInstruction &insn,
780 const MCDisassembler *Dis) {
781 if (!insn.spec) {
782 debug("Instruction has no specification");
783 return true;
784 }
785
786 mcInst.setOpcode(insn.instructionID);
787
788 int index;
789
790 insn.numImmediatesTranslated = 0;
791
792 for (index = 0; index < X86_MAX_OPERANDS; ++index) {
793 if (insn.operands[index].encoding != ENCODING_NONE) {
794 if (translateOperand(mcInst, insn.operands[index], insn, Dis)) {
795 return true;
796 }
797 }
798 }
799
800 return false;
801 }
802
createX86_32Disassembler(const Target & T,const MCSubtargetInfo & STI)803 static MCDisassembler *createX86_32Disassembler(const Target &T,
804 const MCSubtargetInfo &STI) {
805 return new X86Disassembler::X86GenericDisassembler(STI, MODE_32BIT,
806 T.createMCInstrInfo());
807 }
808
createX86_64Disassembler(const Target & T,const MCSubtargetInfo & STI)809 static MCDisassembler *createX86_64Disassembler(const Target &T,
810 const MCSubtargetInfo &STI) {
811 return new X86Disassembler::X86GenericDisassembler(STI, MODE_64BIT,
812 T.createMCInstrInfo());
813 }
814
LLVMInitializeX86Disassembler()815 extern "C" void LLVMInitializeX86Disassembler() {
816 // Register the disassembler.
817 TargetRegistry::RegisterMCDisassembler(TheX86_32Target,
818 createX86_32Disassembler);
819 TargetRegistry::RegisterMCDisassembler(TheX86_64Target,
820 createX86_64Disassembler);
821 }
822