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 #include "llvm/MC/MCContext.h"
20 #include "llvm/MC/MCDisassembler.h"
21 #include "llvm/MC/MCExpr.h"
22 #include "llvm/MC/MCInst.h"
23 #include "llvm/MC/MCInstrInfo.h"
24 #include "llvm/MC/MCSubtargetInfo.h"
25 #include "llvm/Support/Debug.h"
26 #include "llvm/Support/TargetRegistry.h"
27 #include "llvm/Support/raw_ostream.h"
28 
29 using namespace llvm;
30 using namespace llvm::X86Disassembler;
31 
32 #define DEBUG_TYPE "x86-disassembler"
33 
34 #define GET_REGINFO_ENUM
35 #include "X86GenRegisterInfo.inc"
36 #define GET_INSTRINFO_ENUM
37 #include "X86GenInstrInfo.inc"
38 #define GET_SUBTARGETINFO_ENUM
39 #include "X86GenSubtargetInfo.inc"
40 
Debug(const char * file,unsigned line,const char * s)41 void llvm::X86Disassembler::Debug(const char *file, unsigned line,
42                                   const char *s) {
43   dbgs() << file << ":" << line << ": " << s;
44 }
45 
GetInstrName(unsigned Opcode,const void * mii)46 const char *llvm::X86Disassembler::GetInstrName(unsigned Opcode,
47                                                 const void *mii) {
48   const MCInstrInfo *MII = static_cast<const MCInstrInfo *>(mii);
49   return MII->getName(Opcode);
50 }
51 
52 #define debug(s) DEBUG(Debug(__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,MCContext & Ctx,std::unique_ptr<const MCInstrInfo> MII)78 X86GenericDisassembler::X86GenericDisassembler(
79                                          const MCSubtargetInfo &STI,
80                                          MCContext &Ctx,
81                                          std::unique_ptr<const MCInstrInfo> MII)
82   : MCDisassembler(STI, Ctx), MII(std::move(MII)) {
83   const FeatureBitset &FB = STI.getFeatureBits();
84   if (FB[X86::Mode16Bit]) {
85     fMode = MODE_16BIT;
86     return;
87   } else if (FB[X86::Mode32Bit]) {
88     fMode = MODE_32BIT;
89     return;
90   } else if (FB[X86::Mode64Bit]) {
91     fMode = MODE_64BIT;
92     return;
93   }
94 
95   llvm_unreachable("Invalid CPU mode");
96 }
97 
98 namespace {
99 struct Region {
100   ArrayRef<uint8_t> Bytes;
101   uint64_t Base;
Region__anonb4f9d04e0211::Region102   Region(ArrayRef<uint8_t> Bytes, uint64_t Base) : Bytes(Bytes), Base(Base) {}
103 };
104 } // end anonymous namespace
105 
106 /// A callback function that wraps the readByte method from Region.
107 ///
108 /// @param Arg      - The generic callback parameter.  In this case, this should
109 ///                   be a pointer to a Region.
110 /// @param Byte     - A pointer to the byte to be read.
111 /// @param Address  - The address to be read.
regionReader(const void * Arg,uint8_t * Byte,uint64_t Address)112 static int regionReader(const void *Arg, uint8_t *Byte, uint64_t Address) {
113   auto *R = static_cast<const Region *>(Arg);
114   ArrayRef<uint8_t> Bytes = R->Bytes;
115   unsigned Index = Address - R->Base;
116   if (Bytes.size() <= Index)
117     return -1;
118   *Byte = Bytes[Index];
119   return 0;
120 }
121 
122 /// logger - a callback function that wraps the operator<< method from
123 ///   raw_ostream.
124 ///
125 /// @param arg      - The generic callback parameter.  This should be a pointe
126 ///                   to a raw_ostream.
127 /// @param log      - A string to be logged.  logger() adds a newline.
logger(void * arg,const char * log)128 static void logger(void* arg, const char* log) {
129   if (!arg)
130     return;
131 
132   raw_ostream &vStream = *(static_cast<raw_ostream*>(arg));
133   vStream << log << "\n";
134 }
135 
136 //
137 // Public interface for the disassembler
138 //
139 
getInstruction(MCInst & Instr,uint64_t & Size,ArrayRef<uint8_t> Bytes,uint64_t Address,raw_ostream & VStream,raw_ostream & CStream) const140 MCDisassembler::DecodeStatus X86GenericDisassembler::getInstruction(
141     MCInst &Instr, uint64_t &Size, ArrayRef<uint8_t> Bytes, uint64_t Address,
142     raw_ostream &VStream, raw_ostream &CStream) const {
143   CommentStream = &CStream;
144 
145   InternalInstruction InternalInstr;
146 
147   dlog_t LoggerFn = logger;
148   if (&VStream == &nulls())
149     LoggerFn = nullptr; // Disable logging completely if it's going to nulls().
150 
151   Region R(Bytes, Address);
152 
153   int Ret = decodeInstruction(&InternalInstr, regionReader, (const void *)&R,
154                               LoggerFn, (void *)&VStream,
155                               (const void *)MII.get(), Address, fMode);
156 
157   if (Ret) {
158     Size = InternalInstr.readerCursor - Address;
159     return Fail;
160   } else {
161     Size = InternalInstr.length;
162     return (!translateInstruction(Instr, InternalInstr, this)) ? Success : Fail;
163   }
164 }
165 
166 //
167 // Private code that translates from struct InternalInstructions to MCInsts.
168 //
169 
170 /// translateRegister - Translates an internal register to the appropriate LLVM
171 ///   register, and appends it as an operand to an MCInst.
172 ///
173 /// @param mcInst     - The MCInst to append to.
174 /// @param reg        - The Reg to append.
translateRegister(MCInst & mcInst,Reg reg)175 static void translateRegister(MCInst &mcInst, Reg reg) {
176 #define ENTRY(x) X86::x,
177   uint8_t llvmRegnums[] = {
178     ALL_REGS
179     0
180   };
181 #undef ENTRY
182 
183   uint8_t llvmRegnum = llvmRegnums[reg];
184   mcInst.addOperand(MCOperand::createReg(llvmRegnum));
185 }
186 
187 /// tryAddingSymbolicOperand - trys to add a symbolic operand in place of the
188 /// immediate Value in the MCInst.
189 ///
190 /// @param Value      - The immediate Value, has had any PC adjustment made by
191 ///                     the caller.
192 /// @param isBranch   - If the instruction is a branch instruction
193 /// @param Address    - The starting address of the instruction
194 /// @param Offset     - The byte offset to this immediate in the instruction
195 /// @param Width      - The byte width of this immediate in the instruction
196 ///
197 /// If the getOpInfo() function was set when setupForSymbolicDisassembly() was
198 /// called then that function is called to get any symbolic information for the
199 /// immediate in the instruction using the Address, Offset and Width.  If that
200 /// returns non-zero then the symbolic information it returns is used to create
201 /// an MCExpr and that is added as an operand to the MCInst.  If getOpInfo()
202 /// returns zero and isBranch is true then a symbol look up for immediate Value
203 /// is done and if a symbol is found an MCExpr is created with that, else
204 /// an MCExpr with the immediate Value is created.  This function returns true
205 /// 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)206 static bool tryAddingSymbolicOperand(int64_t Value, bool isBranch,
207                                      uint64_t Address, uint64_t Offset,
208                                      uint64_t Width, MCInst &MI,
209                                      const MCDisassembler *Dis) {
210   return Dis->tryAddingSymbolicOperand(MI, Value, Address, isBranch,
211                                        Offset, Width);
212 }
213 
214 /// tryAddingPcLoadReferenceComment - trys to add a comment as to what is being
215 /// referenced by a load instruction with the base register that is the rip.
216 /// These can often be addresses in a literal pool.  The Address of the
217 /// instruction and its immediate Value are used to determine the address
218 /// being referenced in the literal pool entry.  The SymbolLookUp call back will
219 /// return a pointer to a literal 'C' string if the referenced address is an
220 /// address into a section with 'C' string literals.
tryAddingPcLoadReferenceComment(uint64_t Address,uint64_t Value,const void * Decoder)221 static void tryAddingPcLoadReferenceComment(uint64_t Address, uint64_t Value,
222                                             const void *Decoder) {
223   const MCDisassembler *Dis = static_cast<const MCDisassembler*>(Decoder);
224   Dis->tryAddingPcLoadReferenceComment(Value, Address);
225 }
226 
227 static const uint8_t segmentRegnums[SEG_OVERRIDE_max] = {
228   0,        // SEG_OVERRIDE_NONE
229   X86::CS,
230   X86::SS,
231   X86::DS,
232   X86::ES,
233   X86::FS,
234   X86::GS
235 };
236 
237 /// translateSrcIndex   - Appends a source index operand to an MCInst.
238 ///
239 /// @param mcInst       - The MCInst to append to.
240 /// @param insn         - The internal instruction.
translateSrcIndex(MCInst & mcInst,InternalInstruction & insn)241 static bool translateSrcIndex(MCInst &mcInst, InternalInstruction &insn) {
242   unsigned baseRegNo;
243 
244   if (insn.mode == MODE_64BIT)
245     baseRegNo = insn.prefixPresent[0x67] ? X86::ESI : X86::RSI;
246   else if (insn.mode == MODE_32BIT)
247     baseRegNo = insn.prefixPresent[0x67] ? X86::SI : X86::ESI;
248   else {
249     assert(insn.mode == MODE_16BIT);
250     baseRegNo = insn.prefixPresent[0x67] ? X86::ESI : X86::SI;
251   }
252   MCOperand baseReg = MCOperand::createReg(baseRegNo);
253   mcInst.addOperand(baseReg);
254 
255   MCOperand segmentReg;
256   segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
257   mcInst.addOperand(segmentReg);
258   return false;
259 }
260 
261 /// translateDstIndex   - Appends a destination index operand to an MCInst.
262 ///
263 /// @param mcInst       - The MCInst to append to.
264 /// @param insn         - The internal instruction.
265 
translateDstIndex(MCInst & mcInst,InternalInstruction & insn)266 static bool translateDstIndex(MCInst &mcInst, InternalInstruction &insn) {
267   unsigned baseRegNo;
268 
269   if (insn.mode == MODE_64BIT)
270     baseRegNo = insn.prefixPresent[0x67] ? X86::EDI : X86::RDI;
271   else if (insn.mode == MODE_32BIT)
272     baseRegNo = insn.prefixPresent[0x67] ? X86::DI : X86::EDI;
273   else {
274     assert(insn.mode == MODE_16BIT);
275     baseRegNo = insn.prefixPresent[0x67] ? X86::EDI : X86::DI;
276   }
277   MCOperand baseReg = MCOperand::createReg(baseRegNo);
278   mcInst.addOperand(baseReg);
279   return false;
280 }
281 
282 /// translateImmediate  - Appends an immediate operand to an MCInst.
283 ///
284 /// @param mcInst       - The MCInst to append to.
285 /// @param immediate    - The immediate value to append.
286 /// @param operand      - The operand, as stored in the descriptor table.
287 /// @param insn         - The internal instruction.
translateImmediate(MCInst & mcInst,uint64_t immediate,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)288 static void translateImmediate(MCInst &mcInst, uint64_t immediate,
289                                const OperandSpecifier &operand,
290                                InternalInstruction &insn,
291                                const MCDisassembler *Dis) {
292   // Sign-extend the immediate if necessary.
293 
294   OperandType type = (OperandType)operand.type;
295 
296   bool isBranch = false;
297   uint64_t pcrel = 0;
298   if (type == TYPE_RELv) {
299     isBranch = true;
300     pcrel = insn.startLocation +
301             insn.immediateOffset + insn.immediateSize;
302     switch (insn.displacementSize) {
303     default:
304       break;
305     case 1:
306       if(immediate & 0x80)
307         immediate |= ~(0xffull);
308       break;
309     case 2:
310       if(immediate & 0x8000)
311         immediate |= ~(0xffffull);
312       break;
313     case 4:
314       if(immediate & 0x80000000)
315         immediate |= ~(0xffffffffull);
316       break;
317     case 8:
318       break;
319     }
320   }
321   // By default sign-extend all X86 immediates based on their encoding.
322   else if (type == TYPE_IMM8 || type == TYPE_IMM16 || type == TYPE_IMM32 ||
323            type == TYPE_IMM64 || type == TYPE_IMMv) {
324     switch (operand.encoding) {
325     default:
326       break;
327     case ENCODING_IB:
328       if(immediate & 0x80)
329         immediate |= ~(0xffull);
330       break;
331     case ENCODING_IW:
332       if(immediate & 0x8000)
333         immediate |= ~(0xffffull);
334       break;
335     case ENCODING_ID:
336       if(immediate & 0x80000000)
337         immediate |= ~(0xffffffffull);
338       break;
339     case ENCODING_IO:
340       break;
341     }
342   } else if (type == TYPE_IMM3) {
343     // Check for immediates that printSSECC can't handle.
344     if (immediate >= 8) {
345       unsigned NewOpc;
346       switch (mcInst.getOpcode()) {
347       default: llvm_unreachable("unexpected opcode");
348       case X86::CMPPDrmi:  NewOpc = X86::CMPPDrmi_alt;  break;
349       case X86::CMPPDrri:  NewOpc = X86::CMPPDrri_alt;  break;
350       case X86::CMPPSrmi:  NewOpc = X86::CMPPSrmi_alt;  break;
351       case X86::CMPPSrri:  NewOpc = X86::CMPPSrri_alt;  break;
352       case X86::CMPSDrm:   NewOpc = X86::CMPSDrm_alt;   break;
353       case X86::CMPSDrr:   NewOpc = X86::CMPSDrr_alt;   break;
354       case X86::CMPSSrm:   NewOpc = X86::CMPSSrm_alt;   break;
355       case X86::CMPSSrr:   NewOpc = X86::CMPSSrr_alt;   break;
356       case X86::VPCOMBri:  NewOpc = X86::VPCOMBri_alt;  break;
357       case X86::VPCOMBmi:  NewOpc = X86::VPCOMBmi_alt;  break;
358       case X86::VPCOMWri:  NewOpc = X86::VPCOMWri_alt;  break;
359       case X86::VPCOMWmi:  NewOpc = X86::VPCOMWmi_alt;  break;
360       case X86::VPCOMDri:  NewOpc = X86::VPCOMDri_alt;  break;
361       case X86::VPCOMDmi:  NewOpc = X86::VPCOMDmi_alt;  break;
362       case X86::VPCOMQri:  NewOpc = X86::VPCOMQri_alt;  break;
363       case X86::VPCOMQmi:  NewOpc = X86::VPCOMQmi_alt;  break;
364       case X86::VPCOMUBri: NewOpc = X86::VPCOMUBri_alt; break;
365       case X86::VPCOMUBmi: NewOpc = X86::VPCOMUBmi_alt; break;
366       case X86::VPCOMUWri: NewOpc = X86::VPCOMUWri_alt; break;
367       case X86::VPCOMUWmi: NewOpc = X86::VPCOMUWmi_alt; break;
368       case X86::VPCOMUDri: NewOpc = X86::VPCOMUDri_alt; break;
369       case X86::VPCOMUDmi: NewOpc = X86::VPCOMUDmi_alt; break;
370       case X86::VPCOMUQri: NewOpc = X86::VPCOMUQri_alt; break;
371       case X86::VPCOMUQmi: NewOpc = X86::VPCOMUQmi_alt; break;
372       }
373       // Switch opcode to the one that doesn't get special printing.
374       mcInst.setOpcode(NewOpc);
375     }
376   } else if (type == TYPE_IMM5) {
377     // Check for immediates that printAVXCC can't handle.
378     if (immediate >= 32) {
379       unsigned NewOpc;
380       switch (mcInst.getOpcode()) {
381       default: llvm_unreachable("unexpected opcode");
382       case X86::VCMPPDrmi:   NewOpc = X86::VCMPPDrmi_alt;   break;
383       case X86::VCMPPDrri:   NewOpc = X86::VCMPPDrri_alt;   break;
384       case X86::VCMPPSrmi:   NewOpc = X86::VCMPPSrmi_alt;   break;
385       case X86::VCMPPSrri:   NewOpc = X86::VCMPPSrri_alt;   break;
386       case X86::VCMPSDrm:    NewOpc = X86::VCMPSDrm_alt;    break;
387       case X86::VCMPSDrr:    NewOpc = X86::VCMPSDrr_alt;    break;
388       case X86::VCMPSSrm:    NewOpc = X86::VCMPSSrm_alt;    break;
389       case X86::VCMPSSrr:    NewOpc = X86::VCMPSSrr_alt;    break;
390       case X86::VCMPPDYrmi:  NewOpc = X86::VCMPPDYrmi_alt;  break;
391       case X86::VCMPPDYrri:  NewOpc = X86::VCMPPDYrri_alt;  break;
392       case X86::VCMPPSYrmi:  NewOpc = X86::VCMPPSYrmi_alt;  break;
393       case X86::VCMPPSYrri:  NewOpc = X86::VCMPPSYrri_alt;  break;
394       case X86::VCMPPDZrmi:  NewOpc = X86::VCMPPDZrmi_alt;  break;
395       case X86::VCMPPDZrri:  NewOpc = X86::VCMPPDZrri_alt;  break;
396       case X86::VCMPPDZrrib: NewOpc = X86::VCMPPDZrrib_alt; break;
397       case X86::VCMPPSZrmi:  NewOpc = X86::VCMPPSZrmi_alt;  break;
398       case X86::VCMPPSZrri:  NewOpc = X86::VCMPPSZrri_alt;  break;
399       case X86::VCMPPSZrrib: NewOpc = X86::VCMPPSZrrib_alt; break;
400       case X86::VCMPSDZrm:   NewOpc = X86::VCMPSDZrmi_alt;  break;
401       case X86::VCMPSDZrr:   NewOpc = X86::VCMPSDZrri_alt;  break;
402       case X86::VCMPSSZrm:   NewOpc = X86::VCMPSSZrmi_alt;  break;
403       case X86::VCMPSSZrr:   NewOpc = X86::VCMPSSZrri_alt;  break;
404       }
405       // Switch opcode to the one that doesn't get special printing.
406       mcInst.setOpcode(NewOpc);
407     }
408   } else if (type == TYPE_AVX512ICC) {
409     if (immediate >= 8 || ((immediate & 0x3) == 3)) {
410       unsigned NewOpc;
411       switch (mcInst.getOpcode()) {
412       default: llvm_unreachable("unexpected opcode");
413       case X86::VPCMPBZ128rmi:    NewOpc = X86::VPCMPBZ128rmi_alt;    break;
414       case X86::VPCMPBZ128rmik:   NewOpc = X86::VPCMPBZ128rmik_alt;   break;
415       case X86::VPCMPBZ128rri:    NewOpc = X86::VPCMPBZ128rri_alt;    break;
416       case X86::VPCMPBZ128rrik:   NewOpc = X86::VPCMPBZ128rrik_alt;   break;
417       case X86::VPCMPBZ256rmi:    NewOpc = X86::VPCMPBZ256rmi_alt;    break;
418       case X86::VPCMPBZ256rmik:   NewOpc = X86::VPCMPBZ256rmik_alt;   break;
419       case X86::VPCMPBZ256rri:    NewOpc = X86::VPCMPBZ256rri_alt;    break;
420       case X86::VPCMPBZ256rrik:   NewOpc = X86::VPCMPBZ256rrik_alt;   break;
421       case X86::VPCMPBZrmi:       NewOpc = X86::VPCMPBZrmi_alt;       break;
422       case X86::VPCMPBZrmik:      NewOpc = X86::VPCMPBZrmik_alt;      break;
423       case X86::VPCMPBZrri:       NewOpc = X86::VPCMPBZrri_alt;       break;
424       case X86::VPCMPBZrrik:      NewOpc = X86::VPCMPBZrrik_alt;      break;
425       case X86::VPCMPDZ128rmi:    NewOpc = X86::VPCMPDZ128rmi_alt;    break;
426       case X86::VPCMPDZ128rmib:   NewOpc = X86::VPCMPDZ128rmib_alt;   break;
427       case X86::VPCMPDZ128rmibk:  NewOpc = X86::VPCMPDZ128rmibk_alt;  break;
428       case X86::VPCMPDZ128rmik:   NewOpc = X86::VPCMPDZ128rmik_alt;   break;
429       case X86::VPCMPDZ128rri:    NewOpc = X86::VPCMPDZ128rri_alt;    break;
430       case X86::VPCMPDZ128rrik:   NewOpc = X86::VPCMPDZ128rrik_alt;   break;
431       case X86::VPCMPDZ256rmi:    NewOpc = X86::VPCMPDZ256rmi_alt;    break;
432       case X86::VPCMPDZ256rmib:   NewOpc = X86::VPCMPDZ256rmib_alt;   break;
433       case X86::VPCMPDZ256rmibk:  NewOpc = X86::VPCMPDZ256rmibk_alt;  break;
434       case X86::VPCMPDZ256rmik:   NewOpc = X86::VPCMPDZ256rmik_alt;   break;
435       case X86::VPCMPDZ256rri:    NewOpc = X86::VPCMPDZ256rri_alt;    break;
436       case X86::VPCMPDZ256rrik:   NewOpc = X86::VPCMPDZ256rrik_alt;   break;
437       case X86::VPCMPDZrmi:       NewOpc = X86::VPCMPDZrmi_alt;       break;
438       case X86::VPCMPDZrmib:      NewOpc = X86::VPCMPDZrmib_alt;      break;
439       case X86::VPCMPDZrmibk:     NewOpc = X86::VPCMPDZrmibk_alt;     break;
440       case X86::VPCMPDZrmik:      NewOpc = X86::VPCMPDZrmik_alt;      break;
441       case X86::VPCMPDZrri:       NewOpc = X86::VPCMPDZrri_alt;       break;
442       case X86::VPCMPDZrrik:      NewOpc = X86::VPCMPDZrrik_alt;      break;
443       case X86::VPCMPQZ128rmi:    NewOpc = X86::VPCMPQZ128rmi_alt;    break;
444       case X86::VPCMPQZ128rmib:   NewOpc = X86::VPCMPQZ128rmib_alt;   break;
445       case X86::VPCMPQZ128rmibk:  NewOpc = X86::VPCMPQZ128rmibk_alt;  break;
446       case X86::VPCMPQZ128rmik:   NewOpc = X86::VPCMPQZ128rmik_alt;   break;
447       case X86::VPCMPQZ128rri:    NewOpc = X86::VPCMPQZ128rri_alt;    break;
448       case X86::VPCMPQZ128rrik:   NewOpc = X86::VPCMPQZ128rrik_alt;   break;
449       case X86::VPCMPQZ256rmi:    NewOpc = X86::VPCMPQZ256rmi_alt;    break;
450       case X86::VPCMPQZ256rmib:   NewOpc = X86::VPCMPQZ256rmib_alt;   break;
451       case X86::VPCMPQZ256rmibk:  NewOpc = X86::VPCMPQZ256rmibk_alt;  break;
452       case X86::VPCMPQZ256rmik:   NewOpc = X86::VPCMPQZ256rmik_alt;   break;
453       case X86::VPCMPQZ256rri:    NewOpc = X86::VPCMPQZ256rri_alt;    break;
454       case X86::VPCMPQZ256rrik:   NewOpc = X86::VPCMPQZ256rrik_alt;   break;
455       case X86::VPCMPQZrmi:       NewOpc = X86::VPCMPQZrmi_alt;       break;
456       case X86::VPCMPQZrmib:      NewOpc = X86::VPCMPQZrmib_alt;      break;
457       case X86::VPCMPQZrmibk:     NewOpc = X86::VPCMPQZrmibk_alt;     break;
458       case X86::VPCMPQZrmik:      NewOpc = X86::VPCMPQZrmik_alt;      break;
459       case X86::VPCMPQZrri:       NewOpc = X86::VPCMPQZrri_alt;       break;
460       case X86::VPCMPQZrrik:      NewOpc = X86::VPCMPQZrrik_alt;      break;
461       case X86::VPCMPUBZ128rmi:   NewOpc = X86::VPCMPUBZ128rmi_alt;   break;
462       case X86::VPCMPUBZ128rmik:  NewOpc = X86::VPCMPUBZ128rmik_alt;  break;
463       case X86::VPCMPUBZ128rri:   NewOpc = X86::VPCMPUBZ128rri_alt;   break;
464       case X86::VPCMPUBZ128rrik:  NewOpc = X86::VPCMPUBZ128rrik_alt;  break;
465       case X86::VPCMPUBZ256rmi:   NewOpc = X86::VPCMPUBZ256rmi_alt;   break;
466       case X86::VPCMPUBZ256rmik:  NewOpc = X86::VPCMPUBZ256rmik_alt;  break;
467       case X86::VPCMPUBZ256rri:   NewOpc = X86::VPCMPUBZ256rri_alt;   break;
468       case X86::VPCMPUBZ256rrik:  NewOpc = X86::VPCMPUBZ256rrik_alt;  break;
469       case X86::VPCMPUBZrmi:      NewOpc = X86::VPCMPUBZrmi_alt;      break;
470       case X86::VPCMPUBZrmik:     NewOpc = X86::VPCMPUBZrmik_alt;     break;
471       case X86::VPCMPUBZrri:      NewOpc = X86::VPCMPUBZrri_alt;      break;
472       case X86::VPCMPUBZrrik:     NewOpc = X86::VPCMPUBZrrik_alt;     break;
473       case X86::VPCMPUDZ128rmi:   NewOpc = X86::VPCMPUDZ128rmi_alt;   break;
474       case X86::VPCMPUDZ128rmib:  NewOpc = X86::VPCMPUDZ128rmib_alt;  break;
475       case X86::VPCMPUDZ128rmibk: NewOpc = X86::VPCMPUDZ128rmibk_alt; break;
476       case X86::VPCMPUDZ128rmik:  NewOpc = X86::VPCMPUDZ128rmik_alt;  break;
477       case X86::VPCMPUDZ128rri:   NewOpc = X86::VPCMPUDZ128rri_alt;   break;
478       case X86::VPCMPUDZ128rrik:  NewOpc = X86::VPCMPUDZ128rrik_alt;  break;
479       case X86::VPCMPUDZ256rmi:   NewOpc = X86::VPCMPUDZ256rmi_alt;   break;
480       case X86::VPCMPUDZ256rmib:  NewOpc = X86::VPCMPUDZ256rmib_alt;  break;
481       case X86::VPCMPUDZ256rmibk: NewOpc = X86::VPCMPUDZ256rmibk_alt; break;
482       case X86::VPCMPUDZ256rmik:  NewOpc = X86::VPCMPUDZ256rmik_alt;  break;
483       case X86::VPCMPUDZ256rri:   NewOpc = X86::VPCMPUDZ256rri_alt;   break;
484       case X86::VPCMPUDZ256rrik:  NewOpc = X86::VPCMPUDZ256rrik_alt;  break;
485       case X86::VPCMPUDZrmi:      NewOpc = X86::VPCMPUDZrmi_alt;      break;
486       case X86::VPCMPUDZrmib:     NewOpc = X86::VPCMPUDZrmib_alt;     break;
487       case X86::VPCMPUDZrmibk:    NewOpc = X86::VPCMPUDZrmibk_alt;    break;
488       case X86::VPCMPUDZrmik:     NewOpc = X86::VPCMPUDZrmik_alt;     break;
489       case X86::VPCMPUDZrri:      NewOpc = X86::VPCMPUDZrri_alt;      break;
490       case X86::VPCMPUDZrrik:     NewOpc = X86::VPCMPUDZrrik_alt;     break;
491       case X86::VPCMPUQZ128rmi:   NewOpc = X86::VPCMPUQZ128rmi_alt;   break;
492       case X86::VPCMPUQZ128rmib:  NewOpc = X86::VPCMPUQZ128rmib_alt;  break;
493       case X86::VPCMPUQZ128rmibk: NewOpc = X86::VPCMPUQZ128rmibk_alt; break;
494       case X86::VPCMPUQZ128rmik:  NewOpc = X86::VPCMPUQZ128rmik_alt;  break;
495       case X86::VPCMPUQZ128rri:   NewOpc = X86::VPCMPUQZ128rri_alt;   break;
496       case X86::VPCMPUQZ128rrik:  NewOpc = X86::VPCMPUQZ128rrik_alt;  break;
497       case X86::VPCMPUQZ256rmi:   NewOpc = X86::VPCMPUQZ256rmi_alt;   break;
498       case X86::VPCMPUQZ256rmib:  NewOpc = X86::VPCMPUQZ256rmib_alt;  break;
499       case X86::VPCMPUQZ256rmibk: NewOpc = X86::VPCMPUQZ256rmibk_alt; break;
500       case X86::VPCMPUQZ256rmik:  NewOpc = X86::VPCMPUQZ256rmik_alt;  break;
501       case X86::VPCMPUQZ256rri:   NewOpc = X86::VPCMPUQZ256rri_alt;   break;
502       case X86::VPCMPUQZ256rrik:  NewOpc = X86::VPCMPUQZ256rrik_alt;  break;
503       case X86::VPCMPUQZrmi:      NewOpc = X86::VPCMPUQZrmi_alt;      break;
504       case X86::VPCMPUQZrmib:     NewOpc = X86::VPCMPUQZrmib_alt;     break;
505       case X86::VPCMPUQZrmibk:    NewOpc = X86::VPCMPUQZrmibk_alt;    break;
506       case X86::VPCMPUQZrmik:     NewOpc = X86::VPCMPUQZrmik_alt;     break;
507       case X86::VPCMPUQZrri:      NewOpc = X86::VPCMPUQZrri_alt;      break;
508       case X86::VPCMPUQZrrik:     NewOpc = X86::VPCMPUQZrrik_alt;     break;
509       case X86::VPCMPUWZ128rmi:   NewOpc = X86::VPCMPUWZ128rmi_alt;   break;
510       case X86::VPCMPUWZ128rmik:  NewOpc = X86::VPCMPUWZ128rmik_alt;  break;
511       case X86::VPCMPUWZ128rri:   NewOpc = X86::VPCMPUWZ128rri_alt;   break;
512       case X86::VPCMPUWZ128rrik:  NewOpc = X86::VPCMPUWZ128rrik_alt;  break;
513       case X86::VPCMPUWZ256rmi:   NewOpc = X86::VPCMPUWZ256rmi_alt;   break;
514       case X86::VPCMPUWZ256rmik:  NewOpc = X86::VPCMPUWZ256rmik_alt;  break;
515       case X86::VPCMPUWZ256rri:   NewOpc = X86::VPCMPUWZ256rri_alt;   break;
516       case X86::VPCMPUWZ256rrik:  NewOpc = X86::VPCMPUWZ256rrik_alt;  break;
517       case X86::VPCMPUWZrmi:      NewOpc = X86::VPCMPUWZrmi_alt;      break;
518       case X86::VPCMPUWZrmik:     NewOpc = X86::VPCMPUWZrmik_alt;     break;
519       case X86::VPCMPUWZrri:      NewOpc = X86::VPCMPUWZrri_alt;      break;
520       case X86::VPCMPUWZrrik:     NewOpc = X86::VPCMPUWZrrik_alt;     break;
521       case X86::VPCMPWZ128rmi:    NewOpc = X86::VPCMPWZ128rmi_alt;    break;
522       case X86::VPCMPWZ128rmik:   NewOpc = X86::VPCMPWZ128rmik_alt;   break;
523       case X86::VPCMPWZ128rri:    NewOpc = X86::VPCMPWZ128rri_alt;    break;
524       case X86::VPCMPWZ128rrik:   NewOpc = X86::VPCMPWZ128rrik_alt;   break;
525       case X86::VPCMPWZ256rmi:    NewOpc = X86::VPCMPWZ256rmi_alt;    break;
526       case X86::VPCMPWZ256rmik:   NewOpc = X86::VPCMPWZ256rmik_alt;   break;
527       case X86::VPCMPWZ256rri:    NewOpc = X86::VPCMPWZ256rri_alt;    break;
528       case X86::VPCMPWZ256rrik:   NewOpc = X86::VPCMPWZ256rrik_alt;   break;
529       case X86::VPCMPWZrmi:       NewOpc = X86::VPCMPWZrmi_alt;       break;
530       case X86::VPCMPWZrmik:      NewOpc = X86::VPCMPWZrmik_alt;      break;
531       case X86::VPCMPWZrri:       NewOpc = X86::VPCMPWZrri_alt;       break;
532       case X86::VPCMPWZrrik:      NewOpc = X86::VPCMPWZrrik_alt;      break;
533       }
534       // Switch opcode to the one that doesn't get special printing.
535       mcInst.setOpcode(NewOpc);
536     }
537   }
538 
539   switch (type) {
540   case TYPE_XMM32:
541   case TYPE_XMM64:
542   case TYPE_XMM128:
543     mcInst.addOperand(MCOperand::createReg(X86::XMM0 + (immediate >> 4)));
544     return;
545   case TYPE_XMM256:
546     mcInst.addOperand(MCOperand::createReg(X86::YMM0 + (immediate >> 4)));
547     return;
548   case TYPE_XMM512:
549     mcInst.addOperand(MCOperand::createReg(X86::ZMM0 + (immediate >> 4)));
550     return;
551   case TYPE_BNDR:
552     mcInst.addOperand(MCOperand::createReg(X86::BND0 + (immediate >> 4)));
553   case TYPE_REL8:
554     isBranch = true;
555     pcrel = insn.startLocation + insn.immediateOffset + insn.immediateSize;
556     if (immediate & 0x80)
557       immediate |= ~(0xffull);
558     break;
559   case TYPE_REL16:
560     isBranch = true;
561     pcrel = insn.startLocation + insn.immediateOffset + insn.immediateSize;
562     if (immediate & 0x8000)
563       immediate |= ~(0xffffull);
564     break;
565   case TYPE_REL32:
566   case TYPE_REL64:
567     isBranch = true;
568     pcrel = insn.startLocation + insn.immediateOffset + insn.immediateSize;
569     if(immediate & 0x80000000)
570       immediate |= ~(0xffffffffull);
571     break;
572   default:
573     // operand is 64 bits wide.  Do nothing.
574     break;
575   }
576 
577   if(!tryAddingSymbolicOperand(immediate + pcrel, isBranch, insn.startLocation,
578                                insn.immediateOffset, insn.immediateSize,
579                                mcInst, Dis))
580     mcInst.addOperand(MCOperand::createImm(immediate));
581 
582   if (type == TYPE_MOFFS8 || type == TYPE_MOFFS16 ||
583       type == TYPE_MOFFS32 || type == TYPE_MOFFS64) {
584     MCOperand segmentReg;
585     segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
586     mcInst.addOperand(segmentReg);
587   }
588 }
589 
590 /// translateRMRegister - Translates a register stored in the R/M field of the
591 ///   ModR/M byte to its LLVM equivalent and appends it to an MCInst.
592 /// @param mcInst       - The MCInst to append to.
593 /// @param insn         - The internal instruction to extract the R/M field
594 ///                       from.
595 /// @return             - 0 on success; -1 otherwise
translateRMRegister(MCInst & mcInst,InternalInstruction & insn)596 static bool translateRMRegister(MCInst &mcInst,
597                                 InternalInstruction &insn) {
598   if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
599     debug("A R/M register operand may not have a SIB byte");
600     return true;
601   }
602 
603   switch (insn.eaBase) {
604   default:
605     debug("Unexpected EA base register");
606     return true;
607   case EA_BASE_NONE:
608     debug("EA_BASE_NONE for ModR/M base");
609     return true;
610 #define ENTRY(x) case EA_BASE_##x:
611   ALL_EA_BASES
612 #undef ENTRY
613     debug("A R/M register operand may not have a base; "
614           "the operand must be a register.");
615     return true;
616 #define ENTRY(x)                                                      \
617   case EA_REG_##x:                                                    \
618     mcInst.addOperand(MCOperand::createReg(X86::x)); break;
619   ALL_REGS
620 #undef ENTRY
621   }
622 
623   return false;
624 }
625 
626 /// translateRMMemory - Translates a memory operand stored in the Mod and R/M
627 ///   fields of an internal instruction (and possibly its SIB byte) to a memory
628 ///   operand in LLVM's format, and appends it to an MCInst.
629 ///
630 /// @param mcInst       - The MCInst to append to.
631 /// @param insn         - The instruction to extract Mod, R/M, and SIB fields
632 ///                       from.
633 /// @return             - 0 on success; nonzero otherwise
translateRMMemory(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis)634 static bool translateRMMemory(MCInst &mcInst, InternalInstruction &insn,
635                               const MCDisassembler *Dis) {
636   // Addresses in an MCInst are represented as five operands:
637   //   1. basereg       (register)  The R/M base, or (if there is a SIB) the
638   //                                SIB base
639   //   2. scaleamount   (immediate) 1, or (if there is a SIB) the specified
640   //                                scale amount
641   //   3. indexreg      (register)  x86_registerNONE, or (if there is a SIB)
642   //                                the index (which is multiplied by the
643   //                                scale amount)
644   //   4. displacement  (immediate) 0, or the displacement if there is one
645   //   5. segmentreg    (register)  x86_registerNONE for now, but could be set
646   //                                if we have segment overrides
647 
648   MCOperand baseReg;
649   MCOperand scaleAmount;
650   MCOperand indexReg;
651   MCOperand displacement;
652   MCOperand segmentReg;
653   uint64_t pcrel = 0;
654 
655   if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
656     if (insn.sibBase != SIB_BASE_NONE) {
657       switch (insn.sibBase) {
658       default:
659         debug("Unexpected sibBase");
660         return true;
661 #define ENTRY(x)                                          \
662       case SIB_BASE_##x:                                  \
663         baseReg = MCOperand::createReg(X86::x); break;
664       ALL_SIB_BASES
665 #undef ENTRY
666       }
667     } else {
668       baseReg = MCOperand::createReg(0);
669     }
670 
671     // Check whether we are handling VSIB addressing mode for GATHER.
672     // If sibIndex was set to SIB_INDEX_NONE, index offset is 4 and
673     // we should use SIB_INDEX_XMM4|YMM4 for VSIB.
674     // I don't see a way to get the correct IndexReg in readSIB:
675     //   We can tell whether it is VSIB or SIB after instruction ID is decoded,
676     //   but instruction ID may not be decoded yet when calling readSIB.
677     uint32_t Opcode = mcInst.getOpcode();
678     bool IndexIs128 = (Opcode == X86::VGATHERDPDrm ||
679                        Opcode == X86::VGATHERDPDYrm ||
680                        Opcode == X86::VGATHERQPDrm ||
681                        Opcode == X86::VGATHERDPSrm ||
682                        Opcode == X86::VGATHERQPSrm ||
683                        Opcode == X86::VPGATHERDQrm ||
684                        Opcode == X86::VPGATHERDQYrm ||
685                        Opcode == X86::VPGATHERQQrm ||
686                        Opcode == X86::VPGATHERDDrm ||
687                        Opcode == X86::VPGATHERQDrm);
688     bool IndexIs256 = (Opcode == X86::VGATHERQPDYrm ||
689                        Opcode == X86::VGATHERDPSYrm ||
690                        Opcode == X86::VGATHERQPSYrm ||
691                        Opcode == X86::VGATHERDPDZrm ||
692                        Opcode == X86::VPGATHERDQZrm ||
693                        Opcode == X86::VPGATHERQQYrm ||
694                        Opcode == X86::VPGATHERDDYrm ||
695                        Opcode == X86::VPGATHERQDYrm);
696     bool IndexIs512 = (Opcode == X86::VGATHERQPDZrm ||
697                        Opcode == X86::VGATHERDPSZrm ||
698                        Opcode == X86::VGATHERQPSZrm ||
699                        Opcode == X86::VPGATHERQQZrm ||
700                        Opcode == X86::VPGATHERDDZrm ||
701                        Opcode == X86::VPGATHERQDZrm);
702     if (IndexIs128 || IndexIs256 || IndexIs512) {
703       unsigned IndexOffset = insn.sibIndex -
704                          (insn.addressSize == 8 ? SIB_INDEX_RAX:SIB_INDEX_EAX);
705       SIBIndex IndexBase = IndexIs512 ? SIB_INDEX_ZMM0 :
706                            IndexIs256 ? SIB_INDEX_YMM0 : SIB_INDEX_XMM0;
707       insn.sibIndex = (SIBIndex)(IndexBase +
708                            (insn.sibIndex == SIB_INDEX_NONE ? 4 : IndexOffset));
709     }
710 
711     if (insn.sibIndex != SIB_INDEX_NONE) {
712       switch (insn.sibIndex) {
713       default:
714         debug("Unexpected sibIndex");
715         return true;
716 #define ENTRY(x)                                          \
717       case SIB_INDEX_##x:                                 \
718         indexReg = MCOperand::createReg(X86::x); break;
719       EA_BASES_32BIT
720       EA_BASES_64BIT
721       REGS_XMM
722       REGS_YMM
723       REGS_ZMM
724 #undef ENTRY
725       }
726     } else {
727       indexReg = MCOperand::createReg(0);
728     }
729 
730     scaleAmount = MCOperand::createImm(insn.sibScale);
731   } else {
732     switch (insn.eaBase) {
733     case EA_BASE_NONE:
734       if (insn.eaDisplacement == EA_DISP_NONE) {
735         debug("EA_BASE_NONE and EA_DISP_NONE for ModR/M base");
736         return true;
737       }
738       if (insn.mode == MODE_64BIT){
739         pcrel = insn.startLocation +
740                 insn.displacementOffset + insn.displacementSize;
741         tryAddingPcLoadReferenceComment(insn.startLocation +
742                                         insn.displacementOffset,
743                                         insn.displacement + pcrel, Dis);
744         baseReg = MCOperand::createReg(X86::RIP); // Section 2.2.1.6
745       }
746       else
747         baseReg = MCOperand::createReg(0);
748 
749       indexReg = MCOperand::createReg(0);
750       break;
751     case EA_BASE_BX_SI:
752       baseReg = MCOperand::createReg(X86::BX);
753       indexReg = MCOperand::createReg(X86::SI);
754       break;
755     case EA_BASE_BX_DI:
756       baseReg = MCOperand::createReg(X86::BX);
757       indexReg = MCOperand::createReg(X86::DI);
758       break;
759     case EA_BASE_BP_SI:
760       baseReg = MCOperand::createReg(X86::BP);
761       indexReg = MCOperand::createReg(X86::SI);
762       break;
763     case EA_BASE_BP_DI:
764       baseReg = MCOperand::createReg(X86::BP);
765       indexReg = MCOperand::createReg(X86::DI);
766       break;
767     default:
768       indexReg = MCOperand::createReg(0);
769       switch (insn.eaBase) {
770       default:
771         debug("Unexpected eaBase");
772         return true;
773         // Here, we will use the fill-ins defined above.  However,
774         //   BX_SI, BX_DI, BP_SI, and BP_DI are all handled above and
775         //   sib and sib64 were handled in the top-level if, so they're only
776         //   placeholders to keep the compiler happy.
777 #define ENTRY(x)                                        \
778       case EA_BASE_##x:                                 \
779         baseReg = MCOperand::createReg(X86::x); break;
780       ALL_EA_BASES
781 #undef ENTRY
782 #define ENTRY(x) case EA_REG_##x:
783       ALL_REGS
784 #undef ENTRY
785         debug("A R/M memory operand may not be a register; "
786               "the base field must be a base.");
787         return true;
788       }
789     }
790 
791     scaleAmount = MCOperand::createImm(1);
792   }
793 
794   displacement = MCOperand::createImm(insn.displacement);
795 
796   segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
797 
798   mcInst.addOperand(baseReg);
799   mcInst.addOperand(scaleAmount);
800   mcInst.addOperand(indexReg);
801   if(!tryAddingSymbolicOperand(insn.displacement + pcrel, false,
802                                insn.startLocation, insn.displacementOffset,
803                                insn.displacementSize, mcInst, Dis))
804     mcInst.addOperand(displacement);
805   mcInst.addOperand(segmentReg);
806   return false;
807 }
808 
809 /// translateRM - Translates an operand stored in the R/M (and possibly SIB)
810 ///   byte of an instruction to LLVM form, and appends it to an MCInst.
811 ///
812 /// @param mcInst       - The MCInst to append to.
813 /// @param operand      - The operand, as stored in the descriptor table.
814 /// @param insn         - The instruction to extract Mod, R/M, and SIB fields
815 ///                       from.
816 /// @return             - 0 on success; nonzero otherwise
translateRM(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)817 static bool translateRM(MCInst &mcInst, const OperandSpecifier &operand,
818                         InternalInstruction &insn, const MCDisassembler *Dis) {
819   switch (operand.type) {
820   default:
821     debug("Unexpected type for a R/M operand");
822     return true;
823   case TYPE_R8:
824   case TYPE_R16:
825   case TYPE_R32:
826   case TYPE_R64:
827   case TYPE_Rv:
828   case TYPE_MM64:
829   case TYPE_XMM:
830   case TYPE_XMM32:
831   case TYPE_XMM64:
832   case TYPE_XMM128:
833   case TYPE_XMM256:
834   case TYPE_XMM512:
835   case TYPE_VK1:
836   case TYPE_VK2:
837   case TYPE_VK4:
838   case TYPE_VK8:
839   case TYPE_VK16:
840   case TYPE_VK32:
841   case TYPE_VK64:
842   case TYPE_DEBUGREG:
843   case TYPE_CONTROLREG:
844   case TYPE_BNDR:
845     return translateRMRegister(mcInst, insn);
846   case TYPE_M:
847   case TYPE_M8:
848   case TYPE_M16:
849   case TYPE_M32:
850   case TYPE_M64:
851   case TYPE_M128:
852   case TYPE_M256:
853   case TYPE_M512:
854   case TYPE_Mv:
855   case TYPE_M32FP:
856   case TYPE_M64FP:
857   case TYPE_M80FP:
858   case TYPE_M1616:
859   case TYPE_M1632:
860   case TYPE_M1664:
861   case TYPE_LEA:
862     return translateRMMemory(mcInst, insn, Dis);
863   }
864 }
865 
866 /// translateFPRegister - Translates a stack position on the FPU stack to its
867 ///   LLVM form, and appends it to an MCInst.
868 ///
869 /// @param mcInst       - The MCInst to append to.
870 /// @param stackPos     - The stack position to translate.
translateFPRegister(MCInst & mcInst,uint8_t stackPos)871 static void translateFPRegister(MCInst &mcInst,
872                                 uint8_t stackPos) {
873   mcInst.addOperand(MCOperand::createReg(X86::ST0 + stackPos));
874 }
875 
876 /// translateMaskRegister - Translates a 3-bit mask register number to
877 ///   LLVM form, and appends it to an MCInst.
878 ///
879 /// @param mcInst       - The MCInst to append to.
880 /// @param maskRegNum   - Number of mask register from 0 to 7.
881 /// @return             - false on success; true otherwise.
translateMaskRegister(MCInst & mcInst,uint8_t maskRegNum)882 static bool translateMaskRegister(MCInst &mcInst,
883                                 uint8_t maskRegNum) {
884   if (maskRegNum >= 8) {
885     debug("Invalid mask register number");
886     return true;
887   }
888 
889   mcInst.addOperand(MCOperand::createReg(X86::K0 + maskRegNum));
890   return false;
891 }
892 
893 /// translateOperand - Translates an operand stored in an internal instruction
894 ///   to LLVM's format and appends it to an MCInst.
895 ///
896 /// @param mcInst       - The MCInst to append to.
897 /// @param operand      - The operand, as stored in the descriptor table.
898 /// @param insn         - The internal instruction.
899 /// @return             - false on success; true otherwise.
translateOperand(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)900 static bool translateOperand(MCInst &mcInst, const OperandSpecifier &operand,
901                              InternalInstruction &insn,
902                              const MCDisassembler *Dis) {
903   switch (operand.encoding) {
904   default:
905     debug("Unhandled operand encoding during translation");
906     return true;
907   case ENCODING_REG:
908     translateRegister(mcInst, insn.reg);
909     return false;
910   case ENCODING_WRITEMASK:
911     return translateMaskRegister(mcInst, insn.writemask);
912   CASE_ENCODING_RM:
913     return translateRM(mcInst, operand, insn, Dis);
914   case ENCODING_CB:
915   case ENCODING_CW:
916   case ENCODING_CD:
917   case ENCODING_CP:
918   case ENCODING_CO:
919   case ENCODING_CT:
920     debug("Translation of code offsets isn't supported.");
921     return true;
922   case ENCODING_IB:
923   case ENCODING_IW:
924   case ENCODING_ID:
925   case ENCODING_IO:
926   case ENCODING_Iv:
927   case ENCODING_Ia:
928     translateImmediate(mcInst,
929                        insn.immediates[insn.numImmediatesTranslated++],
930                        operand,
931                        insn,
932                        Dis);
933     return false;
934   case ENCODING_SI:
935     return translateSrcIndex(mcInst, insn);
936   case ENCODING_DI:
937     return translateDstIndex(mcInst, insn);
938   case ENCODING_RB:
939   case ENCODING_RW:
940   case ENCODING_RD:
941   case ENCODING_RO:
942   case ENCODING_Rv:
943     translateRegister(mcInst, insn.opcodeRegister);
944     return false;
945   case ENCODING_FP:
946     translateFPRegister(mcInst, insn.modRM & 7);
947     return false;
948   case ENCODING_VVVV:
949     translateRegister(mcInst, insn.vvvv);
950     return false;
951   case ENCODING_DUP:
952     return translateOperand(mcInst, insn.operands[operand.type - TYPE_DUP0],
953                             insn, Dis);
954   }
955 }
956 
957 /// translateInstruction - Translates an internal instruction and all its
958 ///   operands to an MCInst.
959 ///
960 /// @param mcInst       - The MCInst to populate with the instruction's data.
961 /// @param insn         - The internal instruction.
962 /// @return             - false on success; true otherwise.
translateInstruction(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis)963 static bool translateInstruction(MCInst &mcInst,
964                                 InternalInstruction &insn,
965                                 const MCDisassembler *Dis) {
966   if (!insn.spec) {
967     debug("Instruction has no specification");
968     return true;
969   }
970 
971   mcInst.clear();
972   mcInst.setOpcode(insn.instructionID);
973   // If when reading the prefix bytes we determined the overlapping 0xf2 or 0xf3
974   // prefix bytes should be disassembled as xrelease and xacquire then set the
975   // opcode to those instead of the rep and repne opcodes.
976   if (insn.xAcquireRelease) {
977     if(mcInst.getOpcode() == X86::REP_PREFIX)
978       mcInst.setOpcode(X86::XRELEASE_PREFIX);
979     else if(mcInst.getOpcode() == X86::REPNE_PREFIX)
980       mcInst.setOpcode(X86::XACQUIRE_PREFIX);
981   }
982 
983   insn.numImmediatesTranslated = 0;
984 
985   for (const auto &Op : insn.operands) {
986     if (Op.encoding != ENCODING_NONE) {
987       if (translateOperand(mcInst, Op, insn, Dis)) {
988         return true;
989       }
990     }
991   }
992 
993   return false;
994 }
995 
createX86Disassembler(const Target & T,const MCSubtargetInfo & STI,MCContext & Ctx)996 static MCDisassembler *createX86Disassembler(const Target &T,
997                                              const MCSubtargetInfo &STI,
998                                              MCContext &Ctx) {
999   std::unique_ptr<const MCInstrInfo> MII(T.createMCInstrInfo());
1000   return new X86Disassembler::X86GenericDisassembler(STI, Ctx, std::move(MII));
1001 }
1002 
LLVMInitializeX86Disassembler()1003 extern "C" void LLVMInitializeX86Disassembler() {
1004   // Register the disassembler.
1005   TargetRegistry::RegisterMCDisassembler(TheX86_32Target,
1006                                          createX86Disassembler);
1007   TargetRegistry::RegisterMCDisassembler(TheX86_64Target,
1008                                          createX86Disassembler);
1009 }
1010