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 //
14 //
15 // The X86 disassembler is a table-driven disassembler for the 16-, 32-, and
16 // 64-bit X86 instruction sets.  The main decode sequence for an assembly
17 // instruction in this disassembler is:
18 //
19 // 1. Read the prefix bytes and determine the attributes of the instruction.
20 //    These attributes, recorded in enum attributeBits
21 //    (X86DisassemblerDecoderCommon.h), form a bitmask.  The table CONTEXTS_SYM
22 //    provides a mapping from bitmasks to contexts, which are represented by
23 //    enum InstructionContext (ibid.).
24 //
25 // 2. Read the opcode, and determine what kind of opcode it is.  The
26 //    disassembler distinguishes four kinds of opcodes, which are enumerated in
27 //    OpcodeType (X86DisassemblerDecoderCommon.h): one-byte (0xnn), two-byte
28 //    (0x0f 0xnn), three-byte-38 (0x0f 0x38 0xnn), or three-byte-3a
29 //    (0x0f 0x3a 0xnn).  Mandatory prefixes are treated as part of the context.
30 //
31 // 3. Depending on the opcode type, look in one of four ClassDecision structures
32 //    (X86DisassemblerDecoderCommon.h).  Use the opcode class to determine which
33 //    OpcodeDecision (ibid.) to look the opcode in.  Look up the opcode, to get
34 //    a ModRMDecision (ibid.).
35 //
36 // 4. Some instructions, such as escape opcodes or extended opcodes, or even
37 //    instructions that have ModRM*Reg / ModRM*Mem forms in LLVM, need the
38 //    ModR/M byte to complete decode.  The ModRMDecision's type is an entry from
39 //    ModRMDecisionType (X86DisassemblerDecoderCommon.h) that indicates if the
40 //    ModR/M byte is required and how to interpret it.
41 //
42 // 5. After resolving the ModRMDecision, the disassembler has a unique ID
43 //    of type InstrUID (X86DisassemblerDecoderCommon.h).  Looking this ID up in
44 //    INSTRUCTIONS_SYM yields the name of the instruction and the encodings and
45 //    meanings of its operands.
46 //
47 // 6. For each operand, its encoding is an entry from OperandEncoding
48 //    (X86DisassemblerDecoderCommon.h) and its type is an entry from
49 //    OperandType (ibid.).  The encoding indicates how to read it from the
50 //    instruction; the type indicates how to interpret the value once it has
51 //    been read.  For example, a register operand could be stored in the R/M
52 //    field of the ModR/M byte, the REG field of the ModR/M byte, or added to
53 //    the main opcode.  This is orthogonal from its meaning (an GPR or an XMM
54 //    register, for instance).  Given this information, the operands can be
55 //    extracted and interpreted.
56 //
57 // 7. As the last step, the disassembler translates the instruction information
58 //    and operands into a format understandable by the client - in this case, an
59 //    MCInst for use by the MC infrastructure.
60 //
61 // The disassembler is broken broadly into two parts: the table emitter that
62 // emits the instruction decode tables discussed above during compilation, and
63 // the disassembler itself.  The table emitter is documented in more detail in
64 // utils/TableGen/X86DisassemblerEmitter.h.
65 //
66 // X86Disassembler.cpp contains the code responsible for step 7, and for
67 //   invoking the decoder to execute steps 1-6.
68 // X86DisassemblerDecoderCommon.h contains the definitions needed by both the
69 //   table emitter and the disassembler.
70 // X86DisassemblerDecoder.h contains the public interface of the decoder,
71 //   factored out into C for possible use by other projects.
72 // X86DisassemblerDecoder.c contains the source code of the decoder, which is
73 //   responsible for steps 1-6.
74 //
75 //===----------------------------------------------------------------------===//
76 
77 #include "MCTargetDesc/X86BaseInfo.h"
78 #include "MCTargetDesc/X86MCTargetDesc.h"
79 #include "X86DisassemblerDecoder.h"
80 #include "llvm/MC/MCContext.h"
81 #include "llvm/MC/MCDisassembler/MCDisassembler.h"
82 #include "llvm/MC/MCExpr.h"
83 #include "llvm/MC/MCInst.h"
84 #include "llvm/MC/MCInstrInfo.h"
85 #include "llvm/MC/MCSubtargetInfo.h"
86 #include "llvm/Support/Debug.h"
87 #include "llvm/Support/TargetRegistry.h"
88 #include "llvm/Support/raw_ostream.h"
89 
90 using namespace llvm;
91 using namespace llvm::X86Disassembler;
92 
93 #define DEBUG_TYPE "x86-disassembler"
94 
Debug(const char * file,unsigned line,const char * s)95 void llvm::X86Disassembler::Debug(const char *file, unsigned line,
96                                   const char *s) {
97   dbgs() << file << ":" << line << ": " << s;
98 }
99 
GetInstrName(unsigned Opcode,const void * mii)100 StringRef llvm::X86Disassembler::GetInstrName(unsigned Opcode,
101                                                 const void *mii) {
102   const MCInstrInfo *MII = static_cast<const MCInstrInfo *>(mii);
103   return MII->getName(Opcode);
104 }
105 
106 #define debug(s) LLVM_DEBUG(Debug(__FILE__, __LINE__, s));
107 
108 namespace llvm {
109 
110 // Fill-ins to make the compiler happy.  These constants are never actually
111 //   assigned; they are just filler to make an automatically-generated switch
112 //   statement work.
113 namespace X86 {
114   enum {
115     BX_SI = 500,
116     BX_DI = 501,
117     BP_SI = 502,
118     BP_DI = 503,
119     sib   = 504,
120     sib64 = 505
121   };
122 }
123 
124 }
125 
126 static bool translateInstruction(MCInst &target,
127                                 InternalInstruction &source,
128                                 const MCDisassembler *Dis);
129 
130 namespace {
131 
132 /// Generic disassembler for all X86 platforms. All each platform class should
133 /// have to do is subclass the constructor, and provide a different
134 /// disassemblerMode value.
135 class X86GenericDisassembler : public MCDisassembler {
136   std::unique_ptr<const MCInstrInfo> MII;
137 public:
138   X86GenericDisassembler(const MCSubtargetInfo &STI, MCContext &Ctx,
139                          std::unique_ptr<const MCInstrInfo> MII);
140 public:
141   DecodeStatus getInstruction(MCInst &instr, uint64_t &size,
142                               ArrayRef<uint8_t> Bytes, uint64_t Address,
143                               raw_ostream &vStream,
144                               raw_ostream &cStream) const override;
145 
146 private:
147   DisassemblerMode              fMode;
148 };
149 
150 }
151 
X86GenericDisassembler(const MCSubtargetInfo & STI,MCContext & Ctx,std::unique_ptr<const MCInstrInfo> MII)152 X86GenericDisassembler::X86GenericDisassembler(
153                                          const MCSubtargetInfo &STI,
154                                          MCContext &Ctx,
155                                          std::unique_ptr<const MCInstrInfo> MII)
156   : MCDisassembler(STI, Ctx), MII(std::move(MII)) {
157   const FeatureBitset &FB = STI.getFeatureBits();
158   if (FB[X86::Mode16Bit]) {
159     fMode = MODE_16BIT;
160     return;
161   } else if (FB[X86::Mode32Bit]) {
162     fMode = MODE_32BIT;
163     return;
164   } else if (FB[X86::Mode64Bit]) {
165     fMode = MODE_64BIT;
166     return;
167   }
168 
169   llvm_unreachable("Invalid CPU mode");
170 }
171 
172 namespace {
173 struct Region {
174   ArrayRef<uint8_t> Bytes;
175   uint64_t Base;
Region__anona4268e100311::Region176   Region(ArrayRef<uint8_t> Bytes, uint64_t Base) : Bytes(Bytes), Base(Base) {}
177 };
178 } // end anonymous namespace
179 
180 /// A callback function that wraps the readByte method from Region.
181 ///
182 /// @param Arg      - The generic callback parameter.  In this case, this should
183 ///                   be a pointer to a Region.
184 /// @param Byte     - A pointer to the byte to be read.
185 /// @param Address  - The address to be read.
regionReader(const void * Arg,uint8_t * Byte,uint64_t Address)186 static int regionReader(const void *Arg, uint8_t *Byte, uint64_t Address) {
187   auto *R = static_cast<const Region *>(Arg);
188   ArrayRef<uint8_t> Bytes = R->Bytes;
189   unsigned Index = Address - R->Base;
190   if (Bytes.size() <= Index)
191     return -1;
192   *Byte = Bytes[Index];
193   return 0;
194 }
195 
196 /// logger - a callback function that wraps the operator<< method from
197 ///   raw_ostream.
198 ///
199 /// @param arg      - The generic callback parameter.  This should be a pointe
200 ///                   to a raw_ostream.
201 /// @param log      - A string to be logged.  logger() adds a newline.
logger(void * arg,const char * log)202 static void logger(void* arg, const char* log) {
203   if (!arg)
204     return;
205 
206   raw_ostream &vStream = *(static_cast<raw_ostream*>(arg));
207   vStream << log << "\n";
208 }
209 
210 //
211 // Public interface for the disassembler
212 //
213 
getInstruction(MCInst & Instr,uint64_t & Size,ArrayRef<uint8_t> Bytes,uint64_t Address,raw_ostream & VStream,raw_ostream & CStream) const214 MCDisassembler::DecodeStatus X86GenericDisassembler::getInstruction(
215     MCInst &Instr, uint64_t &Size, ArrayRef<uint8_t> Bytes, uint64_t Address,
216     raw_ostream &VStream, raw_ostream &CStream) const {
217   CommentStream = &CStream;
218 
219   InternalInstruction InternalInstr;
220 
221   dlog_t LoggerFn = logger;
222   if (&VStream == &nulls())
223     LoggerFn = nullptr; // Disable logging completely if it's going to nulls().
224 
225   Region R(Bytes, Address);
226 
227   int Ret = decodeInstruction(&InternalInstr, regionReader, (const void *)&R,
228                               LoggerFn, (void *)&VStream,
229                               (const void *)MII.get(), Address, fMode);
230 
231   if (Ret) {
232     Size = InternalInstr.readerCursor - Address;
233     return Fail;
234   } else {
235     Size = InternalInstr.length;
236     bool Ret = translateInstruction(Instr, InternalInstr, this);
237     if (!Ret) {
238       unsigned Flags = X86::IP_NO_PREFIX;
239       if (InternalInstr.hasAdSize)
240         Flags |= X86::IP_HAS_AD_SIZE;
241       if (!InternalInstr.mandatoryPrefix) {
242         if (InternalInstr.hasOpSize)
243           Flags |= X86::IP_HAS_OP_SIZE;
244         if (InternalInstr.repeatPrefix == 0xf2)
245           Flags |= X86::IP_HAS_REPEAT_NE;
246         else if (InternalInstr.repeatPrefix == 0xf3 &&
247                  // It should not be 'pause' f3 90
248                  InternalInstr.opcode != 0x90)
249           Flags |= X86::IP_HAS_REPEAT;
250         if (InternalInstr.hasLockPrefix)
251           Flags |= X86::IP_HAS_LOCK;
252       }
253       Instr.setFlags(Flags);
254     }
255     return (!Ret) ? Success : Fail;
256   }
257 }
258 
259 //
260 // Private code that translates from struct InternalInstructions to MCInsts.
261 //
262 
263 /// translateRegister - Translates an internal register to the appropriate LLVM
264 ///   register, and appends it as an operand to an MCInst.
265 ///
266 /// @param mcInst     - The MCInst to append to.
267 /// @param reg        - The Reg to append.
translateRegister(MCInst & mcInst,Reg reg)268 static void translateRegister(MCInst &mcInst, Reg reg) {
269 #define ENTRY(x) X86::x,
270   static constexpr MCPhysReg llvmRegnums[] = {ALL_REGS};
271 #undef ENTRY
272 
273   MCPhysReg llvmRegnum = llvmRegnums[reg];
274   mcInst.addOperand(MCOperand::createReg(llvmRegnum));
275 }
276 
277 /// tryAddingSymbolicOperand - trys to add a symbolic operand in place of the
278 /// immediate Value in the MCInst.
279 ///
280 /// @param Value      - The immediate Value, has had any PC adjustment made by
281 ///                     the caller.
282 /// @param isBranch   - If the instruction is a branch instruction
283 /// @param Address    - The starting address of the instruction
284 /// @param Offset     - The byte offset to this immediate in the instruction
285 /// @param Width      - The byte width of this immediate in the instruction
286 ///
287 /// If the getOpInfo() function was set when setupForSymbolicDisassembly() was
288 /// called then that function is called to get any symbolic information for the
289 /// immediate in the instruction using the Address, Offset and Width.  If that
290 /// returns non-zero then the symbolic information it returns is used to create
291 /// an MCExpr and that is added as an operand to the MCInst.  If getOpInfo()
292 /// returns zero and isBranch is true then a symbol look up for immediate Value
293 /// is done and if a symbol is found an MCExpr is created with that, else
294 /// an MCExpr with the immediate Value is created.  This function returns true
295 /// 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)296 static bool tryAddingSymbolicOperand(int64_t Value, bool isBranch,
297                                      uint64_t Address, uint64_t Offset,
298                                      uint64_t Width, MCInst &MI,
299                                      const MCDisassembler *Dis) {
300   return Dis->tryAddingSymbolicOperand(MI, Value, Address, isBranch,
301                                        Offset, Width);
302 }
303 
304 /// tryAddingPcLoadReferenceComment - trys to add a comment as to what is being
305 /// referenced by a load instruction with the base register that is the rip.
306 /// These can often be addresses in a literal pool.  The Address of the
307 /// instruction and its immediate Value are used to determine the address
308 /// being referenced in the literal pool entry.  The SymbolLookUp call back will
309 /// return a pointer to a literal 'C' string if the referenced address is an
310 /// address into a section with 'C' string literals.
tryAddingPcLoadReferenceComment(uint64_t Address,uint64_t Value,const void * Decoder)311 static void tryAddingPcLoadReferenceComment(uint64_t Address, uint64_t Value,
312                                             const void *Decoder) {
313   const MCDisassembler *Dis = static_cast<const MCDisassembler*>(Decoder);
314   Dis->tryAddingPcLoadReferenceComment(Value, Address);
315 }
316 
317 static const uint8_t segmentRegnums[SEG_OVERRIDE_max] = {
318   0,        // SEG_OVERRIDE_NONE
319   X86::CS,
320   X86::SS,
321   X86::DS,
322   X86::ES,
323   X86::FS,
324   X86::GS
325 };
326 
327 /// translateSrcIndex   - Appends a source index operand to an MCInst.
328 ///
329 /// @param mcInst       - The MCInst to append to.
330 /// @param insn         - The internal instruction.
translateSrcIndex(MCInst & mcInst,InternalInstruction & insn)331 static bool translateSrcIndex(MCInst &mcInst, InternalInstruction &insn) {
332   unsigned baseRegNo;
333 
334   if (insn.mode == MODE_64BIT)
335     baseRegNo = insn.hasAdSize ? X86::ESI : X86::RSI;
336   else if (insn.mode == MODE_32BIT)
337     baseRegNo = insn.hasAdSize ? X86::SI : X86::ESI;
338   else {
339     assert(insn.mode == MODE_16BIT);
340     baseRegNo = insn.hasAdSize ? X86::ESI : X86::SI;
341   }
342   MCOperand baseReg = MCOperand::createReg(baseRegNo);
343   mcInst.addOperand(baseReg);
344 
345   MCOperand segmentReg;
346   segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
347   mcInst.addOperand(segmentReg);
348   return false;
349 }
350 
351 /// translateDstIndex   - Appends a destination index operand to an MCInst.
352 ///
353 /// @param mcInst       - The MCInst to append to.
354 /// @param insn         - The internal instruction.
355 
translateDstIndex(MCInst & mcInst,InternalInstruction & insn)356 static bool translateDstIndex(MCInst &mcInst, InternalInstruction &insn) {
357   unsigned baseRegNo;
358 
359   if (insn.mode == MODE_64BIT)
360     baseRegNo = insn.hasAdSize ? X86::EDI : X86::RDI;
361   else if (insn.mode == MODE_32BIT)
362     baseRegNo = insn.hasAdSize ? X86::DI : X86::EDI;
363   else {
364     assert(insn.mode == MODE_16BIT);
365     baseRegNo = insn.hasAdSize ? X86::EDI : X86::DI;
366   }
367   MCOperand baseReg = MCOperand::createReg(baseRegNo);
368   mcInst.addOperand(baseReg);
369   return false;
370 }
371 
372 /// translateImmediate  - Appends an immediate operand to an MCInst.
373 ///
374 /// @param mcInst       - The MCInst to append to.
375 /// @param immediate    - The immediate value to append.
376 /// @param operand      - The operand, as stored in the descriptor table.
377 /// @param insn         - The internal instruction.
translateImmediate(MCInst & mcInst,uint64_t immediate,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)378 static void translateImmediate(MCInst &mcInst, uint64_t immediate,
379                                const OperandSpecifier &operand,
380                                InternalInstruction &insn,
381                                const MCDisassembler *Dis) {
382   // Sign-extend the immediate if necessary.
383 
384   OperandType type = (OperandType)operand.type;
385 
386   bool isBranch = false;
387   uint64_t pcrel = 0;
388   if (type == TYPE_REL) {
389     isBranch = true;
390     pcrel = insn.startLocation +
391             insn.immediateOffset + insn.immediateSize;
392     switch (operand.encoding) {
393     default:
394       break;
395     case ENCODING_Iv:
396       switch (insn.displacementSize) {
397       default:
398         break;
399       case 1:
400         if(immediate & 0x80)
401           immediate |= ~(0xffull);
402         break;
403       case 2:
404         if(immediate & 0x8000)
405           immediate |= ~(0xffffull);
406         break;
407       case 4:
408         if(immediate & 0x80000000)
409           immediate |= ~(0xffffffffull);
410         break;
411       case 8:
412         break;
413       }
414       break;
415     case ENCODING_IB:
416       if(immediate & 0x80)
417         immediate |= ~(0xffull);
418       break;
419     case ENCODING_IW:
420       if(immediate & 0x8000)
421         immediate |= ~(0xffffull);
422       break;
423     case ENCODING_ID:
424       if(immediate & 0x80000000)
425         immediate |= ~(0xffffffffull);
426       break;
427     }
428   }
429   // By default sign-extend all X86 immediates based on their encoding.
430   else if (type == TYPE_IMM) {
431     switch (operand.encoding) {
432     default:
433       break;
434     case ENCODING_IB:
435       if(immediate & 0x80)
436         immediate |= ~(0xffull);
437       break;
438     case ENCODING_IW:
439       if(immediate & 0x8000)
440         immediate |= ~(0xffffull);
441       break;
442     case ENCODING_ID:
443       if(immediate & 0x80000000)
444         immediate |= ~(0xffffffffull);
445       break;
446     case ENCODING_IO:
447       break;
448     }
449   } else if (type == TYPE_IMM3) {
450     // Check for immediates that printSSECC can't handle.
451     if (immediate >= 8) {
452       unsigned NewOpc;
453       switch (mcInst.getOpcode()) {
454       default: llvm_unreachable("unexpected opcode");
455       case X86::CMPPDrmi:  NewOpc = X86::CMPPDrmi_alt;  break;
456       case X86::CMPPDrri:  NewOpc = X86::CMPPDrri_alt;  break;
457       case X86::CMPPSrmi:  NewOpc = X86::CMPPSrmi_alt;  break;
458       case X86::CMPPSrri:  NewOpc = X86::CMPPSrri_alt;  break;
459       case X86::CMPSDrm:   NewOpc = X86::CMPSDrm_alt;   break;
460       case X86::CMPSDrr:   NewOpc = X86::CMPSDrr_alt;   break;
461       case X86::CMPSSrm:   NewOpc = X86::CMPSSrm_alt;   break;
462       case X86::CMPSSrr:   NewOpc = X86::CMPSSrr_alt;   break;
463       case X86::VPCOMBri:  NewOpc = X86::VPCOMBri_alt;  break;
464       case X86::VPCOMBmi:  NewOpc = X86::VPCOMBmi_alt;  break;
465       case X86::VPCOMWri:  NewOpc = X86::VPCOMWri_alt;  break;
466       case X86::VPCOMWmi:  NewOpc = X86::VPCOMWmi_alt;  break;
467       case X86::VPCOMDri:  NewOpc = X86::VPCOMDri_alt;  break;
468       case X86::VPCOMDmi:  NewOpc = X86::VPCOMDmi_alt;  break;
469       case X86::VPCOMQri:  NewOpc = X86::VPCOMQri_alt;  break;
470       case X86::VPCOMQmi:  NewOpc = X86::VPCOMQmi_alt;  break;
471       case X86::VPCOMUBri: NewOpc = X86::VPCOMUBri_alt; break;
472       case X86::VPCOMUBmi: NewOpc = X86::VPCOMUBmi_alt; break;
473       case X86::VPCOMUWri: NewOpc = X86::VPCOMUWri_alt; break;
474       case X86::VPCOMUWmi: NewOpc = X86::VPCOMUWmi_alt; break;
475       case X86::VPCOMUDri: NewOpc = X86::VPCOMUDri_alt; break;
476       case X86::VPCOMUDmi: NewOpc = X86::VPCOMUDmi_alt; break;
477       case X86::VPCOMUQri: NewOpc = X86::VPCOMUQri_alt; break;
478       case X86::VPCOMUQmi: NewOpc = X86::VPCOMUQmi_alt; break;
479       }
480       // Switch opcode to the one that doesn't get special printing.
481       mcInst.setOpcode(NewOpc);
482     }
483   } else if (type == TYPE_IMM5) {
484     // Check for immediates that printAVXCC can't handle.
485     if (immediate >= 32) {
486       unsigned NewOpc;
487       switch (mcInst.getOpcode()) {
488       default: llvm_unreachable("unexpected opcode");
489       case X86::VCMPPDrmi:   NewOpc = X86::VCMPPDrmi_alt;   break;
490       case X86::VCMPPDrri:   NewOpc = X86::VCMPPDrri_alt;   break;
491       case X86::VCMPPSrmi:   NewOpc = X86::VCMPPSrmi_alt;   break;
492       case X86::VCMPPSrri:   NewOpc = X86::VCMPPSrri_alt;   break;
493       case X86::VCMPSDrm:    NewOpc = X86::VCMPSDrm_alt;    break;
494       case X86::VCMPSDrr:    NewOpc = X86::VCMPSDrr_alt;    break;
495       case X86::VCMPSSrm:    NewOpc = X86::VCMPSSrm_alt;    break;
496       case X86::VCMPSSrr:    NewOpc = X86::VCMPSSrr_alt;    break;
497       case X86::VCMPPDYrmi:  NewOpc = X86::VCMPPDYrmi_alt;  break;
498       case X86::VCMPPDYrri:  NewOpc = X86::VCMPPDYrri_alt;  break;
499       case X86::VCMPPSYrmi:  NewOpc = X86::VCMPPSYrmi_alt;  break;
500       case X86::VCMPPSYrri:  NewOpc = X86::VCMPPSYrri_alt;  break;
501       case X86::VCMPPDZrmi:  NewOpc = X86::VCMPPDZrmi_alt;  break;
502       case X86::VCMPPDZrri:  NewOpc = X86::VCMPPDZrri_alt;  break;
503       case X86::VCMPPDZrrib: NewOpc = X86::VCMPPDZrrib_alt; break;
504       case X86::VCMPPSZrmi:  NewOpc = X86::VCMPPSZrmi_alt;  break;
505       case X86::VCMPPSZrri:  NewOpc = X86::VCMPPSZrri_alt;  break;
506       case X86::VCMPPSZrrib: NewOpc = X86::VCMPPSZrrib_alt; break;
507       case X86::VCMPPDZ128rmi:  NewOpc = X86::VCMPPDZ128rmi_alt;  break;
508       case X86::VCMPPDZ128rri:  NewOpc = X86::VCMPPDZ128rri_alt;  break;
509       case X86::VCMPPSZ128rmi:  NewOpc = X86::VCMPPSZ128rmi_alt;  break;
510       case X86::VCMPPSZ128rri:  NewOpc = X86::VCMPPSZ128rri_alt;  break;
511       case X86::VCMPPDZ256rmi:  NewOpc = X86::VCMPPDZ256rmi_alt;  break;
512       case X86::VCMPPDZ256rri:  NewOpc = X86::VCMPPDZ256rri_alt;  break;
513       case X86::VCMPPSZ256rmi:  NewOpc = X86::VCMPPSZ256rmi_alt;  break;
514       case X86::VCMPPSZ256rri:  NewOpc = X86::VCMPPSZ256rri_alt;  break;
515       case X86::VCMPSDZrm_Int:  NewOpc = X86::VCMPSDZrmi_alt;  break;
516       case X86::VCMPSDZrr_Int:  NewOpc = X86::VCMPSDZrri_alt;  break;
517       case X86::VCMPSDZrrb_Int: NewOpc = X86::VCMPSDZrrb_alt;  break;
518       case X86::VCMPSSZrm_Int:  NewOpc = X86::VCMPSSZrmi_alt;  break;
519       case X86::VCMPSSZrr_Int:  NewOpc = X86::VCMPSSZrri_alt;  break;
520       case X86::VCMPSSZrrb_Int: NewOpc = X86::VCMPSSZrrb_alt;  break;
521       }
522       // Switch opcode to the one that doesn't get special printing.
523       mcInst.setOpcode(NewOpc);
524     }
525   } else if (type == TYPE_AVX512ICC) {
526     if (immediate >= 8 || ((immediate & 0x3) == 3)) {
527       unsigned NewOpc;
528       switch (mcInst.getOpcode()) {
529       default: llvm_unreachable("unexpected opcode");
530       case X86::VPCMPBZ128rmi:    NewOpc = X86::VPCMPBZ128rmi_alt;    break;
531       case X86::VPCMPBZ128rmik:   NewOpc = X86::VPCMPBZ128rmik_alt;   break;
532       case X86::VPCMPBZ128rri:    NewOpc = X86::VPCMPBZ128rri_alt;    break;
533       case X86::VPCMPBZ128rrik:   NewOpc = X86::VPCMPBZ128rrik_alt;   break;
534       case X86::VPCMPBZ256rmi:    NewOpc = X86::VPCMPBZ256rmi_alt;    break;
535       case X86::VPCMPBZ256rmik:   NewOpc = X86::VPCMPBZ256rmik_alt;   break;
536       case X86::VPCMPBZ256rri:    NewOpc = X86::VPCMPBZ256rri_alt;    break;
537       case X86::VPCMPBZ256rrik:   NewOpc = X86::VPCMPBZ256rrik_alt;   break;
538       case X86::VPCMPBZrmi:       NewOpc = X86::VPCMPBZrmi_alt;       break;
539       case X86::VPCMPBZrmik:      NewOpc = X86::VPCMPBZrmik_alt;      break;
540       case X86::VPCMPBZrri:       NewOpc = X86::VPCMPBZrri_alt;       break;
541       case X86::VPCMPBZrrik:      NewOpc = X86::VPCMPBZrrik_alt;      break;
542       case X86::VPCMPDZ128rmi:    NewOpc = X86::VPCMPDZ128rmi_alt;    break;
543       case X86::VPCMPDZ128rmib:   NewOpc = X86::VPCMPDZ128rmib_alt;   break;
544       case X86::VPCMPDZ128rmibk:  NewOpc = X86::VPCMPDZ128rmibk_alt;  break;
545       case X86::VPCMPDZ128rmik:   NewOpc = X86::VPCMPDZ128rmik_alt;   break;
546       case X86::VPCMPDZ128rri:    NewOpc = X86::VPCMPDZ128rri_alt;    break;
547       case X86::VPCMPDZ128rrik:   NewOpc = X86::VPCMPDZ128rrik_alt;   break;
548       case X86::VPCMPDZ256rmi:    NewOpc = X86::VPCMPDZ256rmi_alt;    break;
549       case X86::VPCMPDZ256rmib:   NewOpc = X86::VPCMPDZ256rmib_alt;   break;
550       case X86::VPCMPDZ256rmibk:  NewOpc = X86::VPCMPDZ256rmibk_alt;  break;
551       case X86::VPCMPDZ256rmik:   NewOpc = X86::VPCMPDZ256rmik_alt;   break;
552       case X86::VPCMPDZ256rri:    NewOpc = X86::VPCMPDZ256rri_alt;    break;
553       case X86::VPCMPDZ256rrik:   NewOpc = X86::VPCMPDZ256rrik_alt;   break;
554       case X86::VPCMPDZrmi:       NewOpc = X86::VPCMPDZrmi_alt;       break;
555       case X86::VPCMPDZrmib:      NewOpc = X86::VPCMPDZrmib_alt;      break;
556       case X86::VPCMPDZrmibk:     NewOpc = X86::VPCMPDZrmibk_alt;     break;
557       case X86::VPCMPDZrmik:      NewOpc = X86::VPCMPDZrmik_alt;      break;
558       case X86::VPCMPDZrri:       NewOpc = X86::VPCMPDZrri_alt;       break;
559       case X86::VPCMPDZrrik:      NewOpc = X86::VPCMPDZrrik_alt;      break;
560       case X86::VPCMPQZ128rmi:    NewOpc = X86::VPCMPQZ128rmi_alt;    break;
561       case X86::VPCMPQZ128rmib:   NewOpc = X86::VPCMPQZ128rmib_alt;   break;
562       case X86::VPCMPQZ128rmibk:  NewOpc = X86::VPCMPQZ128rmibk_alt;  break;
563       case X86::VPCMPQZ128rmik:   NewOpc = X86::VPCMPQZ128rmik_alt;   break;
564       case X86::VPCMPQZ128rri:    NewOpc = X86::VPCMPQZ128rri_alt;    break;
565       case X86::VPCMPQZ128rrik:   NewOpc = X86::VPCMPQZ128rrik_alt;   break;
566       case X86::VPCMPQZ256rmi:    NewOpc = X86::VPCMPQZ256rmi_alt;    break;
567       case X86::VPCMPQZ256rmib:   NewOpc = X86::VPCMPQZ256rmib_alt;   break;
568       case X86::VPCMPQZ256rmibk:  NewOpc = X86::VPCMPQZ256rmibk_alt;  break;
569       case X86::VPCMPQZ256rmik:   NewOpc = X86::VPCMPQZ256rmik_alt;   break;
570       case X86::VPCMPQZ256rri:    NewOpc = X86::VPCMPQZ256rri_alt;    break;
571       case X86::VPCMPQZ256rrik:   NewOpc = X86::VPCMPQZ256rrik_alt;   break;
572       case X86::VPCMPQZrmi:       NewOpc = X86::VPCMPQZrmi_alt;       break;
573       case X86::VPCMPQZrmib:      NewOpc = X86::VPCMPQZrmib_alt;      break;
574       case X86::VPCMPQZrmibk:     NewOpc = X86::VPCMPQZrmibk_alt;     break;
575       case X86::VPCMPQZrmik:      NewOpc = X86::VPCMPQZrmik_alt;      break;
576       case X86::VPCMPQZrri:       NewOpc = X86::VPCMPQZrri_alt;       break;
577       case X86::VPCMPQZrrik:      NewOpc = X86::VPCMPQZrrik_alt;      break;
578       case X86::VPCMPUBZ128rmi:   NewOpc = X86::VPCMPUBZ128rmi_alt;   break;
579       case X86::VPCMPUBZ128rmik:  NewOpc = X86::VPCMPUBZ128rmik_alt;  break;
580       case X86::VPCMPUBZ128rri:   NewOpc = X86::VPCMPUBZ128rri_alt;   break;
581       case X86::VPCMPUBZ128rrik:  NewOpc = X86::VPCMPUBZ128rrik_alt;  break;
582       case X86::VPCMPUBZ256rmi:   NewOpc = X86::VPCMPUBZ256rmi_alt;   break;
583       case X86::VPCMPUBZ256rmik:  NewOpc = X86::VPCMPUBZ256rmik_alt;  break;
584       case X86::VPCMPUBZ256rri:   NewOpc = X86::VPCMPUBZ256rri_alt;   break;
585       case X86::VPCMPUBZ256rrik:  NewOpc = X86::VPCMPUBZ256rrik_alt;  break;
586       case X86::VPCMPUBZrmi:      NewOpc = X86::VPCMPUBZrmi_alt;      break;
587       case X86::VPCMPUBZrmik:     NewOpc = X86::VPCMPUBZrmik_alt;     break;
588       case X86::VPCMPUBZrri:      NewOpc = X86::VPCMPUBZrri_alt;      break;
589       case X86::VPCMPUBZrrik:     NewOpc = X86::VPCMPUBZrrik_alt;     break;
590       case X86::VPCMPUDZ128rmi:   NewOpc = X86::VPCMPUDZ128rmi_alt;   break;
591       case X86::VPCMPUDZ128rmib:  NewOpc = X86::VPCMPUDZ128rmib_alt;  break;
592       case X86::VPCMPUDZ128rmibk: NewOpc = X86::VPCMPUDZ128rmibk_alt; break;
593       case X86::VPCMPUDZ128rmik:  NewOpc = X86::VPCMPUDZ128rmik_alt;  break;
594       case X86::VPCMPUDZ128rri:   NewOpc = X86::VPCMPUDZ128rri_alt;   break;
595       case X86::VPCMPUDZ128rrik:  NewOpc = X86::VPCMPUDZ128rrik_alt;  break;
596       case X86::VPCMPUDZ256rmi:   NewOpc = X86::VPCMPUDZ256rmi_alt;   break;
597       case X86::VPCMPUDZ256rmib:  NewOpc = X86::VPCMPUDZ256rmib_alt;  break;
598       case X86::VPCMPUDZ256rmibk: NewOpc = X86::VPCMPUDZ256rmibk_alt; break;
599       case X86::VPCMPUDZ256rmik:  NewOpc = X86::VPCMPUDZ256rmik_alt;  break;
600       case X86::VPCMPUDZ256rri:   NewOpc = X86::VPCMPUDZ256rri_alt;   break;
601       case X86::VPCMPUDZ256rrik:  NewOpc = X86::VPCMPUDZ256rrik_alt;  break;
602       case X86::VPCMPUDZrmi:      NewOpc = X86::VPCMPUDZrmi_alt;      break;
603       case X86::VPCMPUDZrmib:     NewOpc = X86::VPCMPUDZrmib_alt;     break;
604       case X86::VPCMPUDZrmibk:    NewOpc = X86::VPCMPUDZrmibk_alt;    break;
605       case X86::VPCMPUDZrmik:     NewOpc = X86::VPCMPUDZrmik_alt;     break;
606       case X86::VPCMPUDZrri:      NewOpc = X86::VPCMPUDZrri_alt;      break;
607       case X86::VPCMPUDZrrik:     NewOpc = X86::VPCMPUDZrrik_alt;     break;
608       case X86::VPCMPUQZ128rmi:   NewOpc = X86::VPCMPUQZ128rmi_alt;   break;
609       case X86::VPCMPUQZ128rmib:  NewOpc = X86::VPCMPUQZ128rmib_alt;  break;
610       case X86::VPCMPUQZ128rmibk: NewOpc = X86::VPCMPUQZ128rmibk_alt; break;
611       case X86::VPCMPUQZ128rmik:  NewOpc = X86::VPCMPUQZ128rmik_alt;  break;
612       case X86::VPCMPUQZ128rri:   NewOpc = X86::VPCMPUQZ128rri_alt;   break;
613       case X86::VPCMPUQZ128rrik:  NewOpc = X86::VPCMPUQZ128rrik_alt;  break;
614       case X86::VPCMPUQZ256rmi:   NewOpc = X86::VPCMPUQZ256rmi_alt;   break;
615       case X86::VPCMPUQZ256rmib:  NewOpc = X86::VPCMPUQZ256rmib_alt;  break;
616       case X86::VPCMPUQZ256rmibk: NewOpc = X86::VPCMPUQZ256rmibk_alt; break;
617       case X86::VPCMPUQZ256rmik:  NewOpc = X86::VPCMPUQZ256rmik_alt;  break;
618       case X86::VPCMPUQZ256rri:   NewOpc = X86::VPCMPUQZ256rri_alt;   break;
619       case X86::VPCMPUQZ256rrik:  NewOpc = X86::VPCMPUQZ256rrik_alt;  break;
620       case X86::VPCMPUQZrmi:      NewOpc = X86::VPCMPUQZrmi_alt;      break;
621       case X86::VPCMPUQZrmib:     NewOpc = X86::VPCMPUQZrmib_alt;     break;
622       case X86::VPCMPUQZrmibk:    NewOpc = X86::VPCMPUQZrmibk_alt;    break;
623       case X86::VPCMPUQZrmik:     NewOpc = X86::VPCMPUQZrmik_alt;     break;
624       case X86::VPCMPUQZrri:      NewOpc = X86::VPCMPUQZrri_alt;      break;
625       case X86::VPCMPUQZrrik:     NewOpc = X86::VPCMPUQZrrik_alt;     break;
626       case X86::VPCMPUWZ128rmi:   NewOpc = X86::VPCMPUWZ128rmi_alt;   break;
627       case X86::VPCMPUWZ128rmik:  NewOpc = X86::VPCMPUWZ128rmik_alt;  break;
628       case X86::VPCMPUWZ128rri:   NewOpc = X86::VPCMPUWZ128rri_alt;   break;
629       case X86::VPCMPUWZ128rrik:  NewOpc = X86::VPCMPUWZ128rrik_alt;  break;
630       case X86::VPCMPUWZ256rmi:   NewOpc = X86::VPCMPUWZ256rmi_alt;   break;
631       case X86::VPCMPUWZ256rmik:  NewOpc = X86::VPCMPUWZ256rmik_alt;  break;
632       case X86::VPCMPUWZ256rri:   NewOpc = X86::VPCMPUWZ256rri_alt;   break;
633       case X86::VPCMPUWZ256rrik:  NewOpc = X86::VPCMPUWZ256rrik_alt;  break;
634       case X86::VPCMPUWZrmi:      NewOpc = X86::VPCMPUWZrmi_alt;      break;
635       case X86::VPCMPUWZrmik:     NewOpc = X86::VPCMPUWZrmik_alt;     break;
636       case X86::VPCMPUWZrri:      NewOpc = X86::VPCMPUWZrri_alt;      break;
637       case X86::VPCMPUWZrrik:     NewOpc = X86::VPCMPUWZrrik_alt;     break;
638       case X86::VPCMPWZ128rmi:    NewOpc = X86::VPCMPWZ128rmi_alt;    break;
639       case X86::VPCMPWZ128rmik:   NewOpc = X86::VPCMPWZ128rmik_alt;   break;
640       case X86::VPCMPWZ128rri:    NewOpc = X86::VPCMPWZ128rri_alt;    break;
641       case X86::VPCMPWZ128rrik:   NewOpc = X86::VPCMPWZ128rrik_alt;   break;
642       case X86::VPCMPWZ256rmi:    NewOpc = X86::VPCMPWZ256rmi_alt;    break;
643       case X86::VPCMPWZ256rmik:   NewOpc = X86::VPCMPWZ256rmik_alt;   break;
644       case X86::VPCMPWZ256rri:    NewOpc = X86::VPCMPWZ256rri_alt;    break;
645       case X86::VPCMPWZ256rrik:   NewOpc = X86::VPCMPWZ256rrik_alt;   break;
646       case X86::VPCMPWZrmi:       NewOpc = X86::VPCMPWZrmi_alt;       break;
647       case X86::VPCMPWZrmik:      NewOpc = X86::VPCMPWZrmik_alt;      break;
648       case X86::VPCMPWZrri:       NewOpc = X86::VPCMPWZrri_alt;       break;
649       case X86::VPCMPWZrrik:      NewOpc = X86::VPCMPWZrrik_alt;      break;
650       }
651       // Switch opcode to the one that doesn't get special printing.
652       mcInst.setOpcode(NewOpc);
653     }
654   }
655 
656   switch (type) {
657   case TYPE_XMM:
658     mcInst.addOperand(MCOperand::createReg(X86::XMM0 + (immediate >> 4)));
659     return;
660   case TYPE_YMM:
661     mcInst.addOperand(MCOperand::createReg(X86::YMM0 + (immediate >> 4)));
662     return;
663   case TYPE_ZMM:
664     mcInst.addOperand(MCOperand::createReg(X86::ZMM0 + (immediate >> 4)));
665     return;
666   default:
667     // operand is 64 bits wide.  Do nothing.
668     break;
669   }
670 
671   if(!tryAddingSymbolicOperand(immediate + pcrel, isBranch, insn.startLocation,
672                                insn.immediateOffset, insn.immediateSize,
673                                mcInst, Dis))
674     mcInst.addOperand(MCOperand::createImm(immediate));
675 
676   if (type == TYPE_MOFFS) {
677     MCOperand segmentReg;
678     segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
679     mcInst.addOperand(segmentReg);
680   }
681 }
682 
683 /// translateRMRegister - Translates a register stored in the R/M field of the
684 ///   ModR/M byte to its LLVM equivalent and appends it to an MCInst.
685 /// @param mcInst       - The MCInst to append to.
686 /// @param insn         - The internal instruction to extract the R/M field
687 ///                       from.
688 /// @return             - 0 on success; -1 otherwise
translateRMRegister(MCInst & mcInst,InternalInstruction & insn)689 static bool translateRMRegister(MCInst &mcInst,
690                                 InternalInstruction &insn) {
691   if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
692     debug("A R/M register operand may not have a SIB byte");
693     return true;
694   }
695 
696   switch (insn.eaBase) {
697   default:
698     debug("Unexpected EA base register");
699     return true;
700   case EA_BASE_NONE:
701     debug("EA_BASE_NONE for ModR/M base");
702     return true;
703 #define ENTRY(x) case EA_BASE_##x:
704   ALL_EA_BASES
705 #undef ENTRY
706     debug("A R/M register operand may not have a base; "
707           "the operand must be a register.");
708     return true;
709 #define ENTRY(x)                                                      \
710   case EA_REG_##x:                                                    \
711     mcInst.addOperand(MCOperand::createReg(X86::x)); break;
712   ALL_REGS
713 #undef ENTRY
714   }
715 
716   return false;
717 }
718 
719 /// translateRMMemory - Translates a memory operand stored in the Mod and R/M
720 ///   fields of an internal instruction (and possibly its SIB byte) to a memory
721 ///   operand in LLVM's format, and appends it to an MCInst.
722 ///
723 /// @param mcInst       - The MCInst to append to.
724 /// @param insn         - The instruction to extract Mod, R/M, and SIB fields
725 ///                       from.
726 /// @return             - 0 on success; nonzero otherwise
translateRMMemory(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis)727 static bool translateRMMemory(MCInst &mcInst, InternalInstruction &insn,
728                               const MCDisassembler *Dis) {
729   // Addresses in an MCInst are represented as five operands:
730   //   1. basereg       (register)  The R/M base, or (if there is a SIB) the
731   //                                SIB base
732   //   2. scaleamount   (immediate) 1, or (if there is a SIB) the specified
733   //                                scale amount
734   //   3. indexreg      (register)  x86_registerNONE, or (if there is a SIB)
735   //                                the index (which is multiplied by the
736   //                                scale amount)
737   //   4. displacement  (immediate) 0, or the displacement if there is one
738   //   5. segmentreg    (register)  x86_registerNONE for now, but could be set
739   //                                if we have segment overrides
740 
741   MCOperand baseReg;
742   MCOperand scaleAmount;
743   MCOperand indexReg;
744   MCOperand displacement;
745   MCOperand segmentReg;
746   uint64_t pcrel = 0;
747 
748   if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
749     if (insn.sibBase != SIB_BASE_NONE) {
750       switch (insn.sibBase) {
751       default:
752         debug("Unexpected sibBase");
753         return true;
754 #define ENTRY(x)                                          \
755       case SIB_BASE_##x:                                  \
756         baseReg = MCOperand::createReg(X86::x); break;
757       ALL_SIB_BASES
758 #undef ENTRY
759       }
760     } else {
761       baseReg = MCOperand::createReg(X86::NoRegister);
762     }
763 
764     if (insn.sibIndex != SIB_INDEX_NONE) {
765       switch (insn.sibIndex) {
766       default:
767         debug("Unexpected sibIndex");
768         return true;
769 #define ENTRY(x)                                          \
770       case SIB_INDEX_##x:                                 \
771         indexReg = MCOperand::createReg(X86::x); break;
772       EA_BASES_32BIT
773       EA_BASES_64BIT
774       REGS_XMM
775       REGS_YMM
776       REGS_ZMM
777 #undef ENTRY
778       }
779     } else {
780       // Use EIZ/RIZ for a few ambiguous cases where the SIB byte is present,
781       // but no index is used and modrm alone should have been enough.
782       // -No base register in 32-bit mode. In 64-bit mode this is used to
783       //  avoid rip-relative addressing.
784       // -Any base register used other than ESP/RSP/R12D/R12. Using these as a
785       //  base always requires a SIB byte.
786       // -A scale other than 1 is used.
787       if (insn.sibScale != 1 ||
788           (insn.sibBase == SIB_BASE_NONE && insn.mode != MODE_64BIT) ||
789           (insn.sibBase != SIB_BASE_NONE &&
790            insn.sibBase != SIB_BASE_ESP && insn.sibBase != SIB_BASE_RSP &&
791            insn.sibBase != SIB_BASE_R12D && insn.sibBase != SIB_BASE_R12)) {
792         indexReg = MCOperand::createReg(insn.addressSize == 4 ? X86::EIZ :
793                                                                 X86::RIZ);
794       } else
795         indexReg = MCOperand::createReg(X86::NoRegister);
796     }
797 
798     scaleAmount = MCOperand::createImm(insn.sibScale);
799   } else {
800     switch (insn.eaBase) {
801     case EA_BASE_NONE:
802       if (insn.eaDisplacement == EA_DISP_NONE) {
803         debug("EA_BASE_NONE and EA_DISP_NONE for ModR/M base");
804         return true;
805       }
806       if (insn.mode == MODE_64BIT){
807         pcrel = insn.startLocation +
808                 insn.displacementOffset + insn.displacementSize;
809         tryAddingPcLoadReferenceComment(insn.startLocation +
810                                         insn.displacementOffset,
811                                         insn.displacement + pcrel, Dis);
812         // Section 2.2.1.6
813         baseReg = MCOperand::createReg(insn.addressSize == 4 ? X86::EIP :
814                                                                X86::RIP);
815       }
816       else
817         baseReg = MCOperand::createReg(X86::NoRegister);
818 
819       indexReg = MCOperand::createReg(X86::NoRegister);
820       break;
821     case EA_BASE_BX_SI:
822       baseReg = MCOperand::createReg(X86::BX);
823       indexReg = MCOperand::createReg(X86::SI);
824       break;
825     case EA_BASE_BX_DI:
826       baseReg = MCOperand::createReg(X86::BX);
827       indexReg = MCOperand::createReg(X86::DI);
828       break;
829     case EA_BASE_BP_SI:
830       baseReg = MCOperand::createReg(X86::BP);
831       indexReg = MCOperand::createReg(X86::SI);
832       break;
833     case EA_BASE_BP_DI:
834       baseReg = MCOperand::createReg(X86::BP);
835       indexReg = MCOperand::createReg(X86::DI);
836       break;
837     default:
838       indexReg = MCOperand::createReg(X86::NoRegister);
839       switch (insn.eaBase) {
840       default:
841         debug("Unexpected eaBase");
842         return true;
843         // Here, we will use the fill-ins defined above.  However,
844         //   BX_SI, BX_DI, BP_SI, and BP_DI are all handled above and
845         //   sib and sib64 were handled in the top-level if, so they're only
846         //   placeholders to keep the compiler happy.
847 #define ENTRY(x)                                        \
848       case EA_BASE_##x:                                 \
849         baseReg = MCOperand::createReg(X86::x); break;
850       ALL_EA_BASES
851 #undef ENTRY
852 #define ENTRY(x) case EA_REG_##x:
853       ALL_REGS
854 #undef ENTRY
855         debug("A R/M memory operand may not be a register; "
856               "the base field must be a base.");
857         return true;
858       }
859     }
860 
861     scaleAmount = MCOperand::createImm(1);
862   }
863 
864   displacement = MCOperand::createImm(insn.displacement);
865 
866   segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
867 
868   mcInst.addOperand(baseReg);
869   mcInst.addOperand(scaleAmount);
870   mcInst.addOperand(indexReg);
871   if(!tryAddingSymbolicOperand(insn.displacement + pcrel, false,
872                                insn.startLocation, insn.displacementOffset,
873                                insn.displacementSize, mcInst, Dis))
874     mcInst.addOperand(displacement);
875   mcInst.addOperand(segmentReg);
876   return false;
877 }
878 
879 /// translateRM - Translates an operand stored in the R/M (and possibly SIB)
880 ///   byte of an instruction to LLVM form, and appends it to an MCInst.
881 ///
882 /// @param mcInst       - The MCInst to append to.
883 /// @param operand      - The operand, as stored in the descriptor table.
884 /// @param insn         - The instruction to extract Mod, R/M, and SIB fields
885 ///                       from.
886 /// @return             - 0 on success; nonzero otherwise
translateRM(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)887 static bool translateRM(MCInst &mcInst, const OperandSpecifier &operand,
888                         InternalInstruction &insn, const MCDisassembler *Dis) {
889   switch (operand.type) {
890   default:
891     debug("Unexpected type for a R/M operand");
892     return true;
893   case TYPE_R8:
894   case TYPE_R16:
895   case TYPE_R32:
896   case TYPE_R64:
897   case TYPE_Rv:
898   case TYPE_MM64:
899   case TYPE_XMM:
900   case TYPE_YMM:
901   case TYPE_ZMM:
902   case TYPE_VK:
903   case TYPE_DEBUGREG:
904   case TYPE_CONTROLREG:
905   case TYPE_BNDR:
906     return translateRMRegister(mcInst, insn);
907   case TYPE_M:
908   case TYPE_MVSIBX:
909   case TYPE_MVSIBY:
910   case TYPE_MVSIBZ:
911     return translateRMMemory(mcInst, insn, Dis);
912   }
913 }
914 
915 /// translateFPRegister - Translates a stack position on the FPU stack to its
916 ///   LLVM form, and appends it to an MCInst.
917 ///
918 /// @param mcInst       - The MCInst to append to.
919 /// @param stackPos     - The stack position to translate.
translateFPRegister(MCInst & mcInst,uint8_t stackPos)920 static void translateFPRegister(MCInst &mcInst,
921                                 uint8_t stackPos) {
922   mcInst.addOperand(MCOperand::createReg(X86::ST0 + stackPos));
923 }
924 
925 /// translateMaskRegister - Translates a 3-bit mask register number to
926 ///   LLVM form, and appends it to an MCInst.
927 ///
928 /// @param mcInst       - The MCInst to append to.
929 /// @param maskRegNum   - Number of mask register from 0 to 7.
930 /// @return             - false on success; true otherwise.
translateMaskRegister(MCInst & mcInst,uint8_t maskRegNum)931 static bool translateMaskRegister(MCInst &mcInst,
932                                 uint8_t maskRegNum) {
933   if (maskRegNum >= 8) {
934     debug("Invalid mask register number");
935     return true;
936   }
937 
938   mcInst.addOperand(MCOperand::createReg(X86::K0 + maskRegNum));
939   return false;
940 }
941 
942 /// translateOperand - Translates an operand stored in an internal instruction
943 ///   to LLVM's format and appends it to an MCInst.
944 ///
945 /// @param mcInst       - The MCInst to append to.
946 /// @param operand      - The operand, as stored in the descriptor table.
947 /// @param insn         - The internal instruction.
948 /// @return             - false on success; true otherwise.
translateOperand(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)949 static bool translateOperand(MCInst &mcInst, const OperandSpecifier &operand,
950                              InternalInstruction &insn,
951                              const MCDisassembler *Dis) {
952   switch (operand.encoding) {
953   default:
954     debug("Unhandled operand encoding during translation");
955     return true;
956   case ENCODING_REG:
957     translateRegister(mcInst, insn.reg);
958     return false;
959   case ENCODING_WRITEMASK:
960     return translateMaskRegister(mcInst, insn.writemask);
961   CASE_ENCODING_RM:
962   CASE_ENCODING_VSIB:
963     return translateRM(mcInst, operand, insn, Dis);
964   case ENCODING_IB:
965   case ENCODING_IW:
966   case ENCODING_ID:
967   case ENCODING_IO:
968   case ENCODING_Iv:
969   case ENCODING_Ia:
970     translateImmediate(mcInst,
971                        insn.immediates[insn.numImmediatesTranslated++],
972                        operand,
973                        insn,
974                        Dis);
975     return false;
976   case ENCODING_IRC:
977     mcInst.addOperand(MCOperand::createImm(insn.RC));
978     return false;
979   case ENCODING_SI:
980     return translateSrcIndex(mcInst, insn);
981   case ENCODING_DI:
982     return translateDstIndex(mcInst, insn);
983   case ENCODING_RB:
984   case ENCODING_RW:
985   case ENCODING_RD:
986   case ENCODING_RO:
987   case ENCODING_Rv:
988     translateRegister(mcInst, insn.opcodeRegister);
989     return false;
990   case ENCODING_FP:
991     translateFPRegister(mcInst, insn.modRM & 7);
992     return false;
993   case ENCODING_VVVV:
994     translateRegister(mcInst, insn.vvvv);
995     return false;
996   case ENCODING_DUP:
997     return translateOperand(mcInst, insn.operands[operand.type - TYPE_DUP0],
998                             insn, Dis);
999   }
1000 }
1001 
1002 /// translateInstruction - Translates an internal instruction and all its
1003 ///   operands to an MCInst.
1004 ///
1005 /// @param mcInst       - The MCInst to populate with the instruction's data.
1006 /// @param insn         - The internal instruction.
1007 /// @return             - false on success; true otherwise.
translateInstruction(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis)1008 static bool translateInstruction(MCInst &mcInst,
1009                                 InternalInstruction &insn,
1010                                 const MCDisassembler *Dis) {
1011   if (!insn.spec) {
1012     debug("Instruction has no specification");
1013     return true;
1014   }
1015 
1016   mcInst.clear();
1017   mcInst.setOpcode(insn.instructionID);
1018   // If when reading the prefix bytes we determined the overlapping 0xf2 or 0xf3
1019   // prefix bytes should be disassembled as xrelease and xacquire then set the
1020   // opcode to those instead of the rep and repne opcodes.
1021   if (insn.xAcquireRelease) {
1022     if(mcInst.getOpcode() == X86::REP_PREFIX)
1023       mcInst.setOpcode(X86::XRELEASE_PREFIX);
1024     else if(mcInst.getOpcode() == X86::REPNE_PREFIX)
1025       mcInst.setOpcode(X86::XACQUIRE_PREFIX);
1026   }
1027 
1028   insn.numImmediatesTranslated = 0;
1029 
1030   for (const auto &Op : insn.operands) {
1031     if (Op.encoding != ENCODING_NONE) {
1032       if (translateOperand(mcInst, Op, insn, Dis)) {
1033         return true;
1034       }
1035     }
1036   }
1037 
1038   return false;
1039 }
1040 
createX86Disassembler(const Target & T,const MCSubtargetInfo & STI,MCContext & Ctx)1041 static MCDisassembler *createX86Disassembler(const Target &T,
1042                                              const MCSubtargetInfo &STI,
1043                                              MCContext &Ctx) {
1044   std::unique_ptr<const MCInstrInfo> MII(T.createMCInstrInfo());
1045   return new X86GenericDisassembler(STI, Ctx, std::move(MII));
1046 }
1047 
LLVMInitializeX86Disassembler()1048 extern "C" void LLVMInitializeX86Disassembler() {
1049   // Register the disassembler.
1050   TargetRegistry::RegisterMCDisassembler(getTheX86_32Target(),
1051                                          createX86Disassembler);
1052   TargetRegistry::RegisterMCDisassembler(getTheX86_64Target(),
1053                                          createX86Disassembler);
1054 }
1055