xref: /freebsd/contrib/llvm-project/llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
1 //===-- X86Disassembler.cpp - Disassembler for x86 and x86_64 -------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file is part of the X86 Disassembler.
10 // It contains code to translate the data produced by the decoder into
11 //  MCInsts.
12 //
13 //
14 // The X86 disassembler is a table-driven disassembler for the 16-, 32-, and
15 // 64-bit X86 instruction sets.  The main decode sequence for an assembly
16 // instruction in this disassembler is:
17 //
18 // 1. Read the prefix bytes and determine the attributes of the instruction.
19 //    These attributes, recorded in enum attributeBits
20 //    (X86DisassemblerDecoderCommon.h), form a bitmask.  The table CONTEXTS_SYM
21 //    provides a mapping from bitmasks to contexts, which are represented by
22 //    enum InstructionContext (ibid.).
23 //
24 // 2. Read the opcode, and determine what kind of opcode it is.  The
25 //    disassembler distinguishes four kinds of opcodes, which are enumerated in
26 //    OpcodeType (X86DisassemblerDecoderCommon.h): one-byte (0xnn), two-byte
27 //    (0x0f 0xnn), three-byte-38 (0x0f 0x38 0xnn), or three-byte-3a
28 //    (0x0f 0x3a 0xnn).  Mandatory prefixes are treated as part of the context.
29 //
30 // 3. Depending on the opcode type, look in one of four ClassDecision structures
31 //    (X86DisassemblerDecoderCommon.h).  Use the opcode class to determine which
32 //    OpcodeDecision (ibid.) to look the opcode in.  Look up the opcode, to get
33 //    a ModRMDecision (ibid.).
34 //
35 // 4. Some instructions, such as escape opcodes or extended opcodes, or even
36 //    instructions that have ModRM*Reg / ModRM*Mem forms in LLVM, need the
37 //    ModR/M byte to complete decode.  The ModRMDecision's type is an entry from
38 //    ModRMDecisionType (X86DisassemblerDecoderCommon.h) that indicates if the
39 //    ModR/M byte is required and how to interpret it.
40 //
41 // 5. After resolving the ModRMDecision, the disassembler has a unique ID
42 //    of type InstrUID (X86DisassemblerDecoderCommon.h).  Looking this ID up in
43 //    INSTRUCTIONS_SYM yields the name of the instruction and the encodings and
44 //    meanings of its operands.
45 //
46 // 6. For each operand, its encoding is an entry from OperandEncoding
47 //    (X86DisassemblerDecoderCommon.h) and its type is an entry from
48 //    OperandType (ibid.).  The encoding indicates how to read it from the
49 //    instruction; the type indicates how to interpret the value once it has
50 //    been read.  For example, a register operand could be stored in the R/M
51 //    field of the ModR/M byte, the REG field of the ModR/M byte, or added to
52 //    the main opcode.  This is orthogonal from its meaning (an GPR or an XMM
53 //    register, for instance).  Given this information, the operands can be
54 //    extracted and interpreted.
55 //
56 // 7. As the last step, the disassembler translates the instruction information
57 //    and operands into a format understandable by the client - in this case, an
58 //    MCInst for use by the MC infrastructure.
59 //
60 // The disassembler is broken broadly into two parts: the table emitter that
61 // emits the instruction decode tables discussed above during compilation, and
62 // the disassembler itself.  The table emitter is documented in more detail in
63 // utils/TableGen/X86DisassemblerEmitter.h.
64 //
65 // X86Disassembler.cpp contains the code responsible for step 7, and for
66 //   invoking the decoder to execute steps 1-6.
67 // X86DisassemblerDecoderCommon.h contains the definitions needed by both the
68 //   table emitter and the disassembler.
69 // X86DisassemblerDecoder.h contains the public interface of the decoder,
70 //   factored out into C for possible use by other projects.
71 // X86DisassemblerDecoder.c contains the source code of the decoder, which is
72 //   responsible for steps 1-6.
73 //
74 //===----------------------------------------------------------------------===//
75 
76 #include "MCTargetDesc/X86BaseInfo.h"
77 #include "MCTargetDesc/X86MCTargetDesc.h"
78 #include "TargetInfo/X86TargetInfo.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/MC/TargetRegistry.h"
87 #include "llvm/Support/Debug.h"
88 #include "llvm/Support/Format.h"
89 #include "llvm/Support/raw_ostream.h"
90 
91 using namespace llvm;
92 using namespace llvm::X86Disassembler;
93 
94 #define DEBUG_TYPE "x86-disassembler"
95 
96 #define debug(s) LLVM_DEBUG(dbgs() << __LINE__ << ": " << s);
97 
98 // Specifies whether a ModR/M byte is needed and (if so) which
99 // instruction each possible value of the ModR/M byte corresponds to.  Once
100 // this information is known, we have narrowed down to a single instruction.
101 struct ModRMDecision {
102   uint8_t modrm_type;
103   uint16_t instructionIDs;
104 };
105 
106 // Specifies which set of ModR/M->instruction tables to look at
107 // given a particular opcode.
108 struct OpcodeDecision {
109   ModRMDecision modRMDecisions[256];
110 };
111 
112 // Specifies which opcode->instruction tables to look at given
113 // a particular context (set of attributes).  Since there are many possible
114 // contexts, the decoder first uses CONTEXTS_SYM to determine which context
115 // applies given a specific set of attributes.  Hence there are only IC_max
116 // entries in this table, rather than 2^(ATTR_max).
117 struct ContextDecision {
118   OpcodeDecision opcodeDecisions[IC_max];
119 };
120 
121 #include "X86GenDisassemblerTables.inc"
122 
decode(OpcodeType type,InstructionContext insnContext,uint8_t opcode,uint8_t modRM)123 static InstrUID decode(OpcodeType type, InstructionContext insnContext,
124                        uint8_t opcode, uint8_t modRM) {
125   const struct ModRMDecision *dec;
126 
127   switch (type) {
128   case ONEBYTE:
129     dec = &ONEBYTE_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
130     break;
131   case TWOBYTE:
132     dec = &TWOBYTE_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
133     break;
134   case THREEBYTE_38:
135     dec = &THREEBYTE38_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
136     break;
137   case THREEBYTE_3A:
138     dec = &THREEBYTE3A_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
139     break;
140   case XOP8_MAP:
141     dec = &XOP8_MAP_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
142     break;
143   case XOP9_MAP:
144     dec = &XOP9_MAP_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
145     break;
146   case XOPA_MAP:
147     dec = &XOPA_MAP_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
148     break;
149   case THREEDNOW_MAP:
150     dec =
151         &THREEDNOW_MAP_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
152     break;
153   case MAP4:
154     dec = &MAP4_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
155     break;
156   case MAP5:
157     dec = &MAP5_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
158     break;
159   case MAP6:
160     dec = &MAP6_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
161     break;
162   case MAP7:
163     dec = &MAP7_SYM.opcodeDecisions[insnContext].modRMDecisions[opcode];
164     break;
165   }
166 
167   switch (dec->modrm_type) {
168   default:
169     llvm_unreachable("Corrupt table!  Unknown modrm_type");
170     return 0;
171   case MODRM_ONEENTRY:
172     return modRMTable[dec->instructionIDs];
173   case MODRM_SPLITRM:
174     if (modFromModRM(modRM) == 0x3)
175       return modRMTable[dec->instructionIDs + 1];
176     return modRMTable[dec->instructionIDs];
177   case MODRM_SPLITREG:
178     if (modFromModRM(modRM) == 0x3)
179       return modRMTable[dec->instructionIDs + ((modRM & 0x38) >> 3) + 8];
180     return modRMTable[dec->instructionIDs + ((modRM & 0x38) >> 3)];
181   case MODRM_SPLITMISC:
182     if (modFromModRM(modRM) == 0x3)
183       return modRMTable[dec->instructionIDs + (modRM & 0x3f) + 8];
184     return modRMTable[dec->instructionIDs + ((modRM & 0x38) >> 3)];
185   case MODRM_FULL:
186     return modRMTable[dec->instructionIDs + modRM];
187   }
188 }
189 
peek(struct InternalInstruction * insn,uint8_t & byte)190 static bool peek(struct InternalInstruction *insn, uint8_t &byte) {
191   uint64_t offset = insn->readerCursor - insn->startLocation;
192   if (offset >= insn->bytes.size())
193     return true;
194   byte = insn->bytes[offset];
195   return false;
196 }
197 
consume(InternalInstruction * insn,T & ptr)198 template <typename T> static bool consume(InternalInstruction *insn, T &ptr) {
199   auto r = insn->bytes;
200   uint64_t offset = insn->readerCursor - insn->startLocation;
201   if (offset + sizeof(T) > r.size())
202     return true;
203   ptr = support::endian::read<T>(&r[offset], llvm::endianness::little);
204   insn->readerCursor += sizeof(T);
205   return false;
206 }
207 
isREX(struct InternalInstruction * insn,uint8_t prefix)208 static bool isREX(struct InternalInstruction *insn, uint8_t prefix) {
209   return insn->mode == MODE_64BIT && prefix >= 0x40 && prefix <= 0x4f;
210 }
211 
isREX2(struct InternalInstruction * insn,uint8_t prefix)212 static bool isREX2(struct InternalInstruction *insn, uint8_t prefix) {
213   return insn->mode == MODE_64BIT && prefix == 0xd5;
214 }
215 
216 // Consumes all of an instruction's prefix bytes, and marks the
217 // instruction as having them.  Also sets the instruction's default operand,
218 // address, and other relevant data sizes to report operands correctly.
219 //
220 // insn must not be empty.
readPrefixes(struct InternalInstruction * insn)221 static int readPrefixes(struct InternalInstruction *insn) {
222   bool isPrefix = true;
223   uint8_t byte = 0;
224   uint8_t nextByte;
225 
226   LLVM_DEBUG(dbgs() << "readPrefixes()");
227 
228   while (isPrefix) {
229     // If we fail reading prefixes, just stop here and let the opcode reader
230     // deal with it.
231     if (consume(insn, byte))
232       break;
233 
234     // If the byte is a LOCK/REP/REPNE prefix and not a part of the opcode, then
235     // break and let it be disassembled as a normal "instruction".
236     if (insn->readerCursor - 1 == insn->startLocation && byte == 0xf0) // LOCK
237       break;
238 
239     if ((byte == 0xf2 || byte == 0xf3) && !peek(insn, nextByte)) {
240       // If the byte is 0xf2 or 0xf3, and any of the following conditions are
241       // met:
242       // - it is followed by a LOCK (0xf0) prefix
243       // - it is followed by an xchg instruction
244       // then it should be disassembled as a xacquire/xrelease not repne/rep.
245       if (((nextByte == 0xf0) ||
246            ((nextByte & 0xfe) == 0x86 || (nextByte & 0xf8) == 0x90))) {
247         insn->xAcquireRelease = true;
248         if (!(byte == 0xf3 && nextByte == 0x90)) // PAUSE instruction support
249           break;
250       }
251       // Also if the byte is 0xf3, and the following condition is met:
252       // - it is followed by a "mov mem, reg" (opcode 0x88/0x89) or
253       //                       "mov mem, imm" (opcode 0xc6/0xc7) instructions.
254       // then it should be disassembled as an xrelease not rep.
255       if (byte == 0xf3 && (nextByte == 0x88 || nextByte == 0x89 ||
256                            nextByte == 0xc6 || nextByte == 0xc7)) {
257         insn->xAcquireRelease = true;
258         break;
259       }
260       if (isREX(insn, nextByte)) {
261         uint8_t nnextByte;
262         // Go to REX prefix after the current one
263         if (consume(insn, nnextByte))
264           return -1;
265         // We should be able to read next byte after REX prefix
266         if (peek(insn, nnextByte))
267           return -1;
268         --insn->readerCursor;
269       }
270     }
271 
272     switch (byte) {
273     case 0xf0: // LOCK
274       insn->hasLockPrefix = true;
275       break;
276     case 0xf2: // REPNE/REPNZ
277     case 0xf3: { // REP or REPE/REPZ
278       uint8_t nextByte;
279       if (peek(insn, nextByte))
280         break;
281       // TODO:
282       //  1. There could be several 0x66
283       //  2. if (nextByte == 0x66) and nextNextByte != 0x0f then
284       //      it's not mandatory prefix
285       //  3. if (nextByte >= 0x40 && nextByte <= 0x4f) it's REX and we need
286       //     0x0f exactly after it to be mandatory prefix
287       //  4. if (nextByte == 0xd5) it's REX2 and we need
288       //     0x0f exactly after it to be mandatory prefix
289       if (isREX(insn, nextByte) || isREX2(insn, nextByte) || nextByte == 0x0f ||
290           nextByte == 0x66)
291         // The last of 0xf2 /0xf3 is mandatory prefix
292         insn->mandatoryPrefix = byte;
293       insn->repeatPrefix = byte;
294       break;
295     }
296     case 0x2e: // CS segment override -OR- Branch not taken
297       insn->segmentOverride = SEG_OVERRIDE_CS;
298       break;
299     case 0x36: // SS segment override -OR- Branch taken
300       insn->segmentOverride = SEG_OVERRIDE_SS;
301       break;
302     case 0x3e: // DS segment override
303       insn->segmentOverride = SEG_OVERRIDE_DS;
304       break;
305     case 0x26: // ES segment override
306       insn->segmentOverride = SEG_OVERRIDE_ES;
307       break;
308     case 0x64: // FS segment override
309       insn->segmentOverride = SEG_OVERRIDE_FS;
310       break;
311     case 0x65: // GS segment override
312       insn->segmentOverride = SEG_OVERRIDE_GS;
313       break;
314     case 0x66: { // Operand-size override {
315       uint8_t nextByte;
316       insn->hasOpSize = true;
317       if (peek(insn, nextByte))
318         break;
319       // 0x66 can't overwrite existing mandatory prefix and should be ignored
320       if (!insn->mandatoryPrefix && (nextByte == 0x0f || isREX(insn, nextByte)))
321         insn->mandatoryPrefix = byte;
322       break;
323     }
324     case 0x67: // Address-size override
325       insn->hasAdSize = true;
326       break;
327     default: // Not a prefix byte
328       isPrefix = false;
329       break;
330     }
331 
332     if (isPrefix)
333       LLVM_DEBUG(dbgs() << format("Found prefix 0x%hhx", byte));
334   }
335 
336   insn->vectorExtensionType = TYPE_NO_VEX_XOP;
337 
338   if (byte == 0x62) {
339     uint8_t byte1, byte2;
340     if (consume(insn, byte1)) {
341       LLVM_DEBUG(dbgs() << "Couldn't read second byte of EVEX prefix");
342       return -1;
343     }
344 
345     if (peek(insn, byte2)) {
346       LLVM_DEBUG(dbgs() << "Couldn't read third byte of EVEX prefix");
347       return -1;
348     }
349 
350     if ((insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0)) {
351       insn->vectorExtensionType = TYPE_EVEX;
352     } else {
353       --insn->readerCursor; // unconsume byte1
354       --insn->readerCursor; // unconsume byte
355     }
356 
357     if (insn->vectorExtensionType == TYPE_EVEX) {
358       insn->vectorExtensionPrefix[0] = byte;
359       insn->vectorExtensionPrefix[1] = byte1;
360       if (consume(insn, insn->vectorExtensionPrefix[2])) {
361         LLVM_DEBUG(dbgs() << "Couldn't read third byte of EVEX prefix");
362         return -1;
363       }
364       if (consume(insn, insn->vectorExtensionPrefix[3])) {
365         LLVM_DEBUG(dbgs() << "Couldn't read fourth byte of EVEX prefix");
366         return -1;
367       }
368 
369       if (insn->mode == MODE_64BIT) {
370         // We simulate the REX prefix for simplicity's sake
371         insn->rexPrefix = 0x40 |
372                           (wFromEVEX3of4(insn->vectorExtensionPrefix[2]) << 3) |
373                           (rFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 2) |
374                           (xFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 1) |
375                           (bFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 0);
376 
377         // We simulate the REX2 prefix for simplicity's sake
378         insn->rex2ExtensionPrefix[1] =
379             (r2FromEVEX2of4(insn->vectorExtensionPrefix[1]) << 6) |
380             (x2FromEVEX3of4(insn->vectorExtensionPrefix[2]) << 5) |
381             (b2FromEVEX2of4(insn->vectorExtensionPrefix[1]) << 4);
382       }
383 
384       LLVM_DEBUG(
385           dbgs() << format(
386               "Found EVEX prefix 0x%hhx 0x%hhx 0x%hhx 0x%hhx",
387               insn->vectorExtensionPrefix[0], insn->vectorExtensionPrefix[1],
388               insn->vectorExtensionPrefix[2], insn->vectorExtensionPrefix[3]));
389     }
390   } else if (byte == 0xc4) {
391     uint8_t byte1;
392     if (peek(insn, byte1)) {
393       LLVM_DEBUG(dbgs() << "Couldn't read second byte of VEX");
394       return -1;
395     }
396 
397     if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0)
398       insn->vectorExtensionType = TYPE_VEX_3B;
399     else
400       --insn->readerCursor;
401 
402     if (insn->vectorExtensionType == TYPE_VEX_3B) {
403       insn->vectorExtensionPrefix[0] = byte;
404       consume(insn, insn->vectorExtensionPrefix[1]);
405       consume(insn, insn->vectorExtensionPrefix[2]);
406 
407       // We simulate the REX prefix for simplicity's sake
408 
409       if (insn->mode == MODE_64BIT)
410         insn->rexPrefix = 0x40 |
411                           (wFromVEX3of3(insn->vectorExtensionPrefix[2]) << 3) |
412                           (rFromVEX2of3(insn->vectorExtensionPrefix[1]) << 2) |
413                           (xFromVEX2of3(insn->vectorExtensionPrefix[1]) << 1) |
414                           (bFromVEX2of3(insn->vectorExtensionPrefix[1]) << 0);
415 
416       LLVM_DEBUG(dbgs() << format("Found VEX prefix 0x%hhx 0x%hhx 0x%hhx",
417                                   insn->vectorExtensionPrefix[0],
418                                   insn->vectorExtensionPrefix[1],
419                                   insn->vectorExtensionPrefix[2]));
420     }
421   } else if (byte == 0xc5) {
422     uint8_t byte1;
423     if (peek(insn, byte1)) {
424       LLVM_DEBUG(dbgs() << "Couldn't read second byte of VEX");
425       return -1;
426     }
427 
428     if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0)
429       insn->vectorExtensionType = TYPE_VEX_2B;
430     else
431       --insn->readerCursor;
432 
433     if (insn->vectorExtensionType == TYPE_VEX_2B) {
434       insn->vectorExtensionPrefix[0] = byte;
435       consume(insn, insn->vectorExtensionPrefix[1]);
436 
437       if (insn->mode == MODE_64BIT)
438         insn->rexPrefix =
439             0x40 | (rFromVEX2of2(insn->vectorExtensionPrefix[1]) << 2);
440 
441       switch (ppFromVEX2of2(insn->vectorExtensionPrefix[1])) {
442       default:
443         break;
444       case VEX_PREFIX_66:
445         insn->hasOpSize = true;
446         break;
447       }
448 
449       LLVM_DEBUG(dbgs() << format("Found VEX prefix 0x%hhx 0x%hhx",
450                                   insn->vectorExtensionPrefix[0],
451                                   insn->vectorExtensionPrefix[1]));
452     }
453   } else if (byte == 0x8f) {
454     uint8_t byte1;
455     if (peek(insn, byte1)) {
456       LLVM_DEBUG(dbgs() << "Couldn't read second byte of XOP");
457       return -1;
458     }
459 
460     if ((byte1 & 0x38) != 0x0) // 0 in these 3 bits is a POP instruction.
461       insn->vectorExtensionType = TYPE_XOP;
462     else
463       --insn->readerCursor;
464 
465     if (insn->vectorExtensionType == TYPE_XOP) {
466       insn->vectorExtensionPrefix[0] = byte;
467       consume(insn, insn->vectorExtensionPrefix[1]);
468       consume(insn, insn->vectorExtensionPrefix[2]);
469 
470       // We simulate the REX prefix for simplicity's sake
471 
472       if (insn->mode == MODE_64BIT)
473         insn->rexPrefix = 0x40 |
474                           (wFromXOP3of3(insn->vectorExtensionPrefix[2]) << 3) |
475                           (rFromXOP2of3(insn->vectorExtensionPrefix[1]) << 2) |
476                           (xFromXOP2of3(insn->vectorExtensionPrefix[1]) << 1) |
477                           (bFromXOP2of3(insn->vectorExtensionPrefix[1]) << 0);
478 
479       switch (ppFromXOP3of3(insn->vectorExtensionPrefix[2])) {
480       default:
481         break;
482       case VEX_PREFIX_66:
483         insn->hasOpSize = true;
484         break;
485       }
486 
487       LLVM_DEBUG(dbgs() << format("Found XOP prefix 0x%hhx 0x%hhx 0x%hhx",
488                                   insn->vectorExtensionPrefix[0],
489                                   insn->vectorExtensionPrefix[1],
490                                   insn->vectorExtensionPrefix[2]));
491     }
492   } else if (isREX2(insn, byte)) {
493     uint8_t byte1;
494     if (peek(insn, byte1)) {
495       LLVM_DEBUG(dbgs() << "Couldn't read second byte of REX2");
496       return -1;
497     }
498     insn->rex2ExtensionPrefix[0] = byte;
499     consume(insn, insn->rex2ExtensionPrefix[1]);
500 
501     // We simulate the REX prefix for simplicity's sake
502     insn->rexPrefix = 0x40 | (wFromREX2(insn->rex2ExtensionPrefix[1]) << 3) |
503                       (rFromREX2(insn->rex2ExtensionPrefix[1]) << 2) |
504                       (xFromREX2(insn->rex2ExtensionPrefix[1]) << 1) |
505                       (bFromREX2(insn->rex2ExtensionPrefix[1]) << 0);
506     LLVM_DEBUG(dbgs() << format("Found REX2 prefix 0x%hhx 0x%hhx",
507                                 insn->rex2ExtensionPrefix[0],
508                                 insn->rex2ExtensionPrefix[1]));
509   } else if (isREX(insn, byte)) {
510     if (peek(insn, nextByte))
511       return -1;
512     insn->rexPrefix = byte;
513     LLVM_DEBUG(dbgs() << format("Found REX prefix 0x%hhx", byte));
514   } else
515     --insn->readerCursor;
516 
517   if (insn->mode == MODE_16BIT) {
518     insn->registerSize = (insn->hasOpSize ? 4 : 2);
519     insn->addressSize = (insn->hasAdSize ? 4 : 2);
520     insn->displacementSize = (insn->hasAdSize ? 4 : 2);
521     insn->immediateSize = (insn->hasOpSize ? 4 : 2);
522   } else if (insn->mode == MODE_32BIT) {
523     insn->registerSize = (insn->hasOpSize ? 2 : 4);
524     insn->addressSize = (insn->hasAdSize ? 2 : 4);
525     insn->displacementSize = (insn->hasAdSize ? 2 : 4);
526     insn->immediateSize = (insn->hasOpSize ? 2 : 4);
527   } else if (insn->mode == MODE_64BIT) {
528     insn->displacementSize = 4;
529     if (insn->rexPrefix && wFromREX(insn->rexPrefix)) {
530       insn->registerSize = 8;
531       insn->addressSize = (insn->hasAdSize ? 4 : 8);
532       insn->immediateSize = 4;
533       insn->hasOpSize = false;
534     } else {
535       insn->registerSize = (insn->hasOpSize ? 2 : 4);
536       insn->addressSize = (insn->hasAdSize ? 4 : 8);
537       insn->immediateSize = (insn->hasOpSize ? 2 : 4);
538     }
539   }
540 
541   return 0;
542 }
543 
544 // Consumes the SIB byte to determine addressing information.
readSIB(struct InternalInstruction * insn)545 static int readSIB(struct InternalInstruction *insn) {
546   SIBBase sibBaseBase = SIB_BASE_NONE;
547   uint8_t index, base;
548 
549   LLVM_DEBUG(dbgs() << "readSIB()");
550   switch (insn->addressSize) {
551   case 2:
552   default:
553     llvm_unreachable("SIB-based addressing doesn't work in 16-bit mode");
554   case 4:
555     insn->sibIndexBase = SIB_INDEX_EAX;
556     sibBaseBase = SIB_BASE_EAX;
557     break;
558   case 8:
559     insn->sibIndexBase = SIB_INDEX_RAX;
560     sibBaseBase = SIB_BASE_RAX;
561     break;
562   }
563 
564   if (consume(insn, insn->sib))
565     return -1;
566 
567   index = indexFromSIB(insn->sib) | (xFromREX(insn->rexPrefix) << 3) |
568           (x2FromREX2(insn->rex2ExtensionPrefix[1]) << 4);
569 
570   if (index == 0x4) {
571     insn->sibIndex = SIB_INDEX_NONE;
572   } else {
573     insn->sibIndex = (SIBIndex)(insn->sibIndexBase + index);
574   }
575 
576   insn->sibScale = 1 << scaleFromSIB(insn->sib);
577 
578   base = baseFromSIB(insn->sib) | (bFromREX(insn->rexPrefix) << 3) |
579          (b2FromREX2(insn->rex2ExtensionPrefix[1]) << 4);
580 
581   switch (base) {
582   case 0x5:
583   case 0xd:
584     switch (modFromModRM(insn->modRM)) {
585     case 0x0:
586       insn->eaDisplacement = EA_DISP_32;
587       insn->sibBase = SIB_BASE_NONE;
588       break;
589     case 0x1:
590       insn->eaDisplacement = EA_DISP_8;
591       insn->sibBase = (SIBBase)(sibBaseBase + base);
592       break;
593     case 0x2:
594       insn->eaDisplacement = EA_DISP_32;
595       insn->sibBase = (SIBBase)(sibBaseBase + base);
596       break;
597     default:
598       llvm_unreachable("Cannot have Mod = 0b11 and a SIB byte");
599     }
600     break;
601   default:
602     insn->sibBase = (SIBBase)(sibBaseBase + base);
603     break;
604   }
605 
606   return 0;
607 }
608 
readDisplacement(struct InternalInstruction * insn)609 static int readDisplacement(struct InternalInstruction *insn) {
610   int8_t d8;
611   int16_t d16;
612   int32_t d32;
613   LLVM_DEBUG(dbgs() << "readDisplacement()");
614 
615   insn->displacementOffset = insn->readerCursor - insn->startLocation;
616   switch (insn->eaDisplacement) {
617   case EA_DISP_NONE:
618     break;
619   case EA_DISP_8:
620     if (consume(insn, d8))
621       return -1;
622     insn->displacement = d8;
623     break;
624   case EA_DISP_16:
625     if (consume(insn, d16))
626       return -1;
627     insn->displacement = d16;
628     break;
629   case EA_DISP_32:
630     if (consume(insn, d32))
631       return -1;
632     insn->displacement = d32;
633     break;
634   }
635 
636   return 0;
637 }
638 
639 // Consumes all addressing information (ModR/M byte, SIB byte, and displacement.
readModRM(struct InternalInstruction * insn)640 static int readModRM(struct InternalInstruction *insn) {
641   uint8_t mod, rm, reg;
642   LLVM_DEBUG(dbgs() << "readModRM()");
643 
644   if (insn->consumedModRM)
645     return 0;
646 
647   if (consume(insn, insn->modRM))
648     return -1;
649   insn->consumedModRM = true;
650 
651   mod = modFromModRM(insn->modRM);
652   rm = rmFromModRM(insn->modRM);
653   reg = regFromModRM(insn->modRM);
654 
655   // This goes by insn->registerSize to pick the correct register, which messes
656   // up if we're using (say) XMM or 8-bit register operands. That gets fixed in
657   // fixupReg().
658   switch (insn->registerSize) {
659   case 2:
660     insn->regBase = MODRM_REG_AX;
661     insn->eaRegBase = EA_REG_AX;
662     break;
663   case 4:
664     insn->regBase = MODRM_REG_EAX;
665     insn->eaRegBase = EA_REG_EAX;
666     break;
667   case 8:
668     insn->regBase = MODRM_REG_RAX;
669     insn->eaRegBase = EA_REG_RAX;
670     break;
671   }
672 
673   reg |= (rFromREX(insn->rexPrefix) << 3) |
674          (r2FromREX2(insn->rex2ExtensionPrefix[1]) << 4);
675   rm |= (bFromREX(insn->rexPrefix) << 3) |
676         (b2FromREX2(insn->rex2ExtensionPrefix[1]) << 4);
677 
678   if (insn->vectorExtensionType == TYPE_EVEX && insn->mode == MODE_64BIT)
679     reg |= r2FromEVEX2of4(insn->vectorExtensionPrefix[1]) << 4;
680 
681   insn->reg = (Reg)(insn->regBase + reg);
682 
683   switch (insn->addressSize) {
684   case 2: {
685     EABase eaBaseBase = EA_BASE_BX_SI;
686 
687     switch (mod) {
688     case 0x0:
689       if (rm == 0x6) {
690         insn->eaBase = EA_BASE_NONE;
691         insn->eaDisplacement = EA_DISP_16;
692         if (readDisplacement(insn))
693           return -1;
694       } else {
695         insn->eaBase = (EABase)(eaBaseBase + rm);
696         insn->eaDisplacement = EA_DISP_NONE;
697       }
698       break;
699     case 0x1:
700       insn->eaBase = (EABase)(eaBaseBase + rm);
701       insn->eaDisplacement = EA_DISP_8;
702       insn->displacementSize = 1;
703       if (readDisplacement(insn))
704         return -1;
705       break;
706     case 0x2:
707       insn->eaBase = (EABase)(eaBaseBase + rm);
708       insn->eaDisplacement = EA_DISP_16;
709       if (readDisplacement(insn))
710         return -1;
711       break;
712     case 0x3:
713       insn->eaBase = (EABase)(insn->eaRegBase + rm);
714       if (readDisplacement(insn))
715         return -1;
716       break;
717     }
718     break;
719   }
720   case 4:
721   case 8: {
722     EABase eaBaseBase = (insn->addressSize == 4 ? EA_BASE_EAX : EA_BASE_RAX);
723 
724     switch (mod) {
725     case 0x0:
726       insn->eaDisplacement = EA_DISP_NONE; // readSIB may override this
727       // In determining whether RIP-relative mode is used (rm=5),
728       // or whether a SIB byte is present (rm=4),
729       // the extension bits (REX.b and EVEX.x) are ignored.
730       switch (rm & 7) {
731       case 0x4: // SIB byte is present
732         insn->eaBase = (insn->addressSize == 4 ? EA_BASE_sib : EA_BASE_sib64);
733         if (readSIB(insn) || readDisplacement(insn))
734           return -1;
735         break;
736       case 0x5: // RIP-relative
737         insn->eaBase = EA_BASE_NONE;
738         insn->eaDisplacement = EA_DISP_32;
739         if (readDisplacement(insn))
740           return -1;
741         break;
742       default:
743         insn->eaBase = (EABase)(eaBaseBase + rm);
744         break;
745       }
746       break;
747     case 0x1:
748       insn->displacementSize = 1;
749       [[fallthrough]];
750     case 0x2:
751       insn->eaDisplacement = (mod == 0x1 ? EA_DISP_8 : EA_DISP_32);
752       switch (rm & 7) {
753       case 0x4: // SIB byte is present
754         insn->eaBase = EA_BASE_sib;
755         if (readSIB(insn) || readDisplacement(insn))
756           return -1;
757         break;
758       default:
759         insn->eaBase = (EABase)(eaBaseBase + rm);
760         if (readDisplacement(insn))
761           return -1;
762         break;
763       }
764       break;
765     case 0x3:
766       insn->eaDisplacement = EA_DISP_NONE;
767       insn->eaBase = (EABase)(insn->eaRegBase + rm);
768       break;
769     }
770     break;
771   }
772   } // switch (insn->addressSize)
773 
774   return 0;
775 }
776 
777 #define GENERIC_FIXUP_FUNC(name, base, prefix)                                 \
778   static uint16_t name(struct InternalInstruction *insn, OperandType type,     \
779                        uint8_t index, uint8_t *valid) {                        \
780     *valid = 1;                                                                \
781     switch (type) {                                                            \
782     default:                                                                   \
783       debug("Unhandled register type");                                        \
784       *valid = 0;                                                              \
785       return 0;                                                                \
786     case TYPE_Rv:                                                              \
787       return base + index;                                                     \
788     case TYPE_R8:                                                              \
789       if (insn->rexPrefix && index >= 4 && index <= 7)                         \
790         return prefix##_SPL + (index - 4);                                     \
791       else                                                                     \
792         return prefix##_AL + index;                                            \
793     case TYPE_R16:                                                             \
794       return prefix##_AX + index;                                              \
795     case TYPE_R32:                                                             \
796       return prefix##_EAX + index;                                             \
797     case TYPE_R64:                                                             \
798       return prefix##_RAX + index;                                             \
799     case TYPE_ZMM:                                                             \
800       return prefix##_ZMM0 + index;                                            \
801     case TYPE_YMM:                                                             \
802       return prefix##_YMM0 + index;                                            \
803     case TYPE_XMM:                                                             \
804       return prefix##_XMM0 + index;                                            \
805     case TYPE_TMM:                                                             \
806       if (index > 7)                                                           \
807         *valid = 0;                                                            \
808       return prefix##_TMM0 + index;                                            \
809     case TYPE_VK:                                                              \
810       index &= 0xf;                                                            \
811       if (index > 7)                                                           \
812         *valid = 0;                                                            \
813       return prefix##_K0 + index;                                              \
814     case TYPE_VK_PAIR:                                                         \
815       if (index > 7)                                                           \
816         *valid = 0;                                                            \
817       return prefix##_K0_K1 + (index / 2);                                     \
818     case TYPE_MM64:                                                            \
819       return prefix##_MM0 + (index & 0x7);                                     \
820     case TYPE_SEGMENTREG:                                                      \
821       if ((index & 7) > 5)                                                     \
822         *valid = 0;                                                            \
823       return prefix##_ES + (index & 7);                                        \
824     case TYPE_DEBUGREG:                                                        \
825       if (index > 15)                                                          \
826         *valid = 0;                                                            \
827       return prefix##_DR0 + index;                                             \
828     case TYPE_CONTROLREG:                                                      \
829       if (index > 15)                                                          \
830         *valid = 0;                                                            \
831       return prefix##_CR0 + index;                                             \
832     case TYPE_MVSIBX:                                                          \
833       return prefix##_XMM0 + index;                                            \
834     case TYPE_MVSIBY:                                                          \
835       return prefix##_YMM0 + index;                                            \
836     case TYPE_MVSIBZ:                                                          \
837       return prefix##_ZMM0 + index;                                            \
838     }                                                                          \
839   }
840 
841 // Consult an operand type to determine the meaning of the reg or R/M field. If
842 // the operand is an XMM operand, for example, an operand would be XMM0 instead
843 // of AX, which readModRM() would otherwise misinterpret it as.
844 //
845 // @param insn  - The instruction containing the operand.
846 // @param type  - The operand type.
847 // @param index - The existing value of the field as reported by readModRM().
848 // @param valid - The address of a uint8_t.  The target is set to 1 if the
849 //                field is valid for the register class; 0 if not.
850 // @return      - The proper value.
851 GENERIC_FIXUP_FUNC(fixupRegValue, insn->regBase, MODRM_REG)
852 GENERIC_FIXUP_FUNC(fixupRMValue, insn->eaRegBase, EA_REG)
853 
854 // Consult an operand specifier to determine which of the fixup*Value functions
855 // to use in correcting readModRM()'ss interpretation.
856 //
857 // @param insn  - See fixup*Value().
858 // @param op    - The operand specifier.
859 // @return      - 0 if fixup was successful; -1 if the register returned was
860 //                invalid for its class.
fixupReg(struct InternalInstruction * insn,const struct OperandSpecifier * op)861 static int fixupReg(struct InternalInstruction *insn,
862                     const struct OperandSpecifier *op) {
863   uint8_t valid;
864   LLVM_DEBUG(dbgs() << "fixupReg()");
865 
866   switch ((OperandEncoding)op->encoding) {
867   default:
868     debug("Expected a REG or R/M encoding in fixupReg");
869     return -1;
870   case ENCODING_VVVV:
871     insn->vvvv =
872         (Reg)fixupRegValue(insn, (OperandType)op->type, insn->vvvv, &valid);
873     if (!valid)
874       return -1;
875     break;
876   case ENCODING_REG:
877     insn->reg = (Reg)fixupRegValue(insn, (OperandType)op->type,
878                                    insn->reg - insn->regBase, &valid);
879     if (!valid)
880       return -1;
881     break;
882   CASE_ENCODING_RM:
883     if (insn->vectorExtensionType == TYPE_EVEX && insn->mode == MODE_64BIT &&
884         modFromModRM(insn->modRM) == 3) {
885       // EVEX_X can extend the register id to 32 for a non-GPR register that is
886       // encoded in RM.
887       // mode : MODE_64_BIT
888       //  Only 8 vector registers are available in 32 bit mode
889       // mod : 3
890       //  RM encodes a register
891       switch (op->type) {
892       case TYPE_Rv:
893       case TYPE_R8:
894       case TYPE_R16:
895       case TYPE_R32:
896       case TYPE_R64:
897         break;
898       default:
899         insn->eaBase =
900             (EABase)(insn->eaBase +
901                      (xFromEVEX2of4(insn->vectorExtensionPrefix[1]) << 4));
902         break;
903       }
904     }
905     [[fallthrough]];
906   case ENCODING_SIB:
907     if (insn->eaBase >= insn->eaRegBase) {
908       insn->eaBase = (EABase)fixupRMValue(
909           insn, (OperandType)op->type, insn->eaBase - insn->eaRegBase, &valid);
910       if (!valid)
911         return -1;
912     }
913     break;
914   }
915 
916   return 0;
917 }
918 
919 // Read the opcode (except the ModR/M byte in the case of extended or escape
920 // opcodes).
readOpcode(struct InternalInstruction * insn)921 static bool readOpcode(struct InternalInstruction *insn) {
922   uint8_t current;
923   LLVM_DEBUG(dbgs() << "readOpcode()");
924 
925   insn->opcodeType = ONEBYTE;
926   if (insn->vectorExtensionType == TYPE_EVEX) {
927     switch (mmmFromEVEX2of4(insn->vectorExtensionPrefix[1])) {
928     default:
929       LLVM_DEBUG(
930           dbgs() << format("Unhandled mmm field for instruction (0x%hhx)",
931                            mmmFromEVEX2of4(insn->vectorExtensionPrefix[1])));
932       return true;
933     case VEX_LOB_0F:
934       insn->opcodeType = TWOBYTE;
935       return consume(insn, insn->opcode);
936     case VEX_LOB_0F38:
937       insn->opcodeType = THREEBYTE_38;
938       return consume(insn, insn->opcode);
939     case VEX_LOB_0F3A:
940       insn->opcodeType = THREEBYTE_3A;
941       return consume(insn, insn->opcode);
942     case VEX_LOB_MAP4:
943       insn->opcodeType = MAP4;
944       return consume(insn, insn->opcode);
945     case VEX_LOB_MAP5:
946       insn->opcodeType = MAP5;
947       return consume(insn, insn->opcode);
948     case VEX_LOB_MAP6:
949       insn->opcodeType = MAP6;
950       return consume(insn, insn->opcode);
951     case VEX_LOB_MAP7:
952       insn->opcodeType = MAP7;
953       return consume(insn, insn->opcode);
954     }
955   } else if (insn->vectorExtensionType == TYPE_VEX_3B) {
956     switch (mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])) {
957     default:
958       LLVM_DEBUG(
959           dbgs() << format("Unhandled m-mmmm field for instruction (0x%hhx)",
960                            mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])));
961       return true;
962     case VEX_LOB_0F:
963       insn->opcodeType = TWOBYTE;
964       return consume(insn, insn->opcode);
965     case VEX_LOB_0F38:
966       insn->opcodeType = THREEBYTE_38;
967       return consume(insn, insn->opcode);
968     case VEX_LOB_0F3A:
969       insn->opcodeType = THREEBYTE_3A;
970       return consume(insn, insn->opcode);
971     case VEX_LOB_MAP5:
972       insn->opcodeType = MAP5;
973       return consume(insn, insn->opcode);
974     case VEX_LOB_MAP6:
975       insn->opcodeType = MAP6;
976       return consume(insn, insn->opcode);
977     case VEX_LOB_MAP7:
978       insn->opcodeType = MAP7;
979       return consume(insn, insn->opcode);
980     }
981   } else if (insn->vectorExtensionType == TYPE_VEX_2B) {
982     insn->opcodeType = TWOBYTE;
983     return consume(insn, insn->opcode);
984   } else if (insn->vectorExtensionType == TYPE_XOP) {
985     switch (mmmmmFromXOP2of3(insn->vectorExtensionPrefix[1])) {
986     default:
987       LLVM_DEBUG(
988           dbgs() << format("Unhandled m-mmmm field for instruction (0x%hhx)",
989                            mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])));
990       return true;
991     case XOP_MAP_SELECT_8:
992       insn->opcodeType = XOP8_MAP;
993       return consume(insn, insn->opcode);
994     case XOP_MAP_SELECT_9:
995       insn->opcodeType = XOP9_MAP;
996       return consume(insn, insn->opcode);
997     case XOP_MAP_SELECT_A:
998       insn->opcodeType = XOPA_MAP;
999       return consume(insn, insn->opcode);
1000     }
1001   } else if (mFromREX2(insn->rex2ExtensionPrefix[1])) {
1002     // m bit indicates opcode map 1
1003     insn->opcodeType = TWOBYTE;
1004     return consume(insn, insn->opcode);
1005   }
1006 
1007   if (consume(insn, current))
1008     return true;
1009 
1010   if (current == 0x0f) {
1011     LLVM_DEBUG(
1012         dbgs() << format("Found a two-byte escape prefix (0x%hhx)", current));
1013     if (consume(insn, current))
1014       return true;
1015 
1016     if (current == 0x38) {
1017       LLVM_DEBUG(dbgs() << format("Found a three-byte escape prefix (0x%hhx)",
1018                                   current));
1019       if (consume(insn, current))
1020         return true;
1021 
1022       insn->opcodeType = THREEBYTE_38;
1023     } else if (current == 0x3a) {
1024       LLVM_DEBUG(dbgs() << format("Found a three-byte escape prefix (0x%hhx)",
1025                                   current));
1026       if (consume(insn, current))
1027         return true;
1028 
1029       insn->opcodeType = THREEBYTE_3A;
1030     } else if (current == 0x0f) {
1031       LLVM_DEBUG(
1032           dbgs() << format("Found a 3dnow escape prefix (0x%hhx)", current));
1033 
1034       // Consume operands before the opcode to comply with the 3DNow encoding
1035       if (readModRM(insn))
1036         return true;
1037 
1038       if (consume(insn, current))
1039         return true;
1040 
1041       insn->opcodeType = THREEDNOW_MAP;
1042     } else {
1043       LLVM_DEBUG(dbgs() << "Didn't find a three-byte escape prefix");
1044       insn->opcodeType = TWOBYTE;
1045     }
1046   } else if (insn->mandatoryPrefix)
1047     // The opcode with mandatory prefix must start with opcode escape.
1048     // If not it's legacy repeat prefix
1049     insn->mandatoryPrefix = 0;
1050 
1051   // At this point we have consumed the full opcode.
1052   // Anything we consume from here on must be unconsumed.
1053   insn->opcode = current;
1054 
1055   return false;
1056 }
1057 
1058 // Determine whether equiv is the 16-bit equivalent of orig (32-bit or 64-bit).
is16BitEquivalent(const char * orig,const char * equiv)1059 static bool is16BitEquivalent(const char *orig, const char *equiv) {
1060   for (int i = 0;; i++) {
1061     if (orig[i] == '\0' && equiv[i] == '\0')
1062       return true;
1063     if (orig[i] == '\0' || equiv[i] == '\0')
1064       return false;
1065     if (orig[i] != equiv[i]) {
1066       if ((orig[i] == 'Q' || orig[i] == 'L') && equiv[i] == 'W')
1067         continue;
1068       if ((orig[i] == '6' || orig[i] == '3') && equiv[i] == '1')
1069         continue;
1070       if ((orig[i] == '4' || orig[i] == '2') && equiv[i] == '6')
1071         continue;
1072       return false;
1073     }
1074   }
1075 }
1076 
1077 // Determine whether this instruction is a 64-bit instruction.
is64Bit(const char * name)1078 static bool is64Bit(const char *name) {
1079   for (int i = 0;; ++i) {
1080     if (name[i] == '\0')
1081       return false;
1082     if (name[i] == '6' && name[i + 1] == '4')
1083       return true;
1084   }
1085 }
1086 
1087 // Determine the ID of an instruction, consuming the ModR/M byte as appropriate
1088 // for extended and escape opcodes, and using a supplied attribute mask.
getInstructionIDWithAttrMask(uint16_t * instructionID,struct InternalInstruction * insn,uint16_t attrMask)1089 static int getInstructionIDWithAttrMask(uint16_t *instructionID,
1090                                         struct InternalInstruction *insn,
1091                                         uint16_t attrMask) {
1092   auto insnCtx = InstructionContext(x86DisassemblerContexts[attrMask]);
1093   const ContextDecision *decision;
1094   switch (insn->opcodeType) {
1095   case ONEBYTE:
1096     decision = &ONEBYTE_SYM;
1097     break;
1098   case TWOBYTE:
1099     decision = &TWOBYTE_SYM;
1100     break;
1101   case THREEBYTE_38:
1102     decision = &THREEBYTE38_SYM;
1103     break;
1104   case THREEBYTE_3A:
1105     decision = &THREEBYTE3A_SYM;
1106     break;
1107   case XOP8_MAP:
1108     decision = &XOP8_MAP_SYM;
1109     break;
1110   case XOP9_MAP:
1111     decision = &XOP9_MAP_SYM;
1112     break;
1113   case XOPA_MAP:
1114     decision = &XOPA_MAP_SYM;
1115     break;
1116   case THREEDNOW_MAP:
1117     decision = &THREEDNOW_MAP_SYM;
1118     break;
1119   case MAP4:
1120     decision = &MAP4_SYM;
1121     break;
1122   case MAP5:
1123     decision = &MAP5_SYM;
1124     break;
1125   case MAP6:
1126     decision = &MAP6_SYM;
1127     break;
1128   case MAP7:
1129     decision = &MAP7_SYM;
1130     break;
1131   }
1132 
1133   if (decision->opcodeDecisions[insnCtx]
1134           .modRMDecisions[insn->opcode]
1135           .modrm_type != MODRM_ONEENTRY) {
1136     if (readModRM(insn))
1137       return -1;
1138     *instructionID =
1139         decode(insn->opcodeType, insnCtx, insn->opcode, insn->modRM);
1140   } else {
1141     *instructionID = decode(insn->opcodeType, insnCtx, insn->opcode, 0);
1142   }
1143 
1144   return 0;
1145 }
1146 
isCCMPOrCTEST(InternalInstruction * insn)1147 static bool isCCMPOrCTEST(InternalInstruction *insn) {
1148   if (insn->opcodeType != MAP4)
1149     return false;
1150   if (insn->opcode == 0x83 && regFromModRM(insn->modRM) == 7)
1151     return true;
1152   switch (insn->opcode & 0xfe) {
1153   default:
1154     return false;
1155   case 0x38:
1156   case 0x3a:
1157   case 0x84:
1158     return true;
1159   case 0x80:
1160     return regFromModRM(insn->modRM) == 7;
1161   case 0xf6:
1162     return regFromModRM(insn->modRM) == 0;
1163   }
1164 }
1165 
isNF(InternalInstruction * insn)1166 static bool isNF(InternalInstruction *insn) {
1167   if (!nfFromEVEX4of4(insn->vectorExtensionPrefix[3]))
1168     return false;
1169   if (insn->opcodeType == MAP4)
1170     return true;
1171   // Below NF instructions are not in map4.
1172   if (insn->opcodeType == THREEBYTE_38 &&
1173       ppFromEVEX3of4(insn->vectorExtensionPrefix[2]) == VEX_PREFIX_NONE) {
1174     switch (insn->opcode) {
1175     case 0xf2: // ANDN
1176     case 0xf3: // BLSI, BLSR, BLSMSK
1177     case 0xf5: // BZHI
1178     case 0xf7: // BEXTR
1179       return true;
1180     default:
1181       break;
1182     }
1183   }
1184   return false;
1185 }
1186 
1187 // Determine the ID of an instruction, consuming the ModR/M byte as appropriate
1188 // for extended and escape opcodes. Determines the attributes and context for
1189 // the instruction before doing so.
getInstructionID(struct InternalInstruction * insn,const MCInstrInfo * mii)1190 static int getInstructionID(struct InternalInstruction *insn,
1191                             const MCInstrInfo *mii) {
1192   uint16_t attrMask;
1193   uint16_t instructionID;
1194 
1195   LLVM_DEBUG(dbgs() << "getID()");
1196 
1197   attrMask = ATTR_NONE;
1198 
1199   if (insn->mode == MODE_64BIT)
1200     attrMask |= ATTR_64BIT;
1201 
1202   if (insn->vectorExtensionType != TYPE_NO_VEX_XOP) {
1203     attrMask |= (insn->vectorExtensionType == TYPE_EVEX) ? ATTR_EVEX : ATTR_VEX;
1204 
1205     if (insn->vectorExtensionType == TYPE_EVEX) {
1206       switch (ppFromEVEX3of4(insn->vectorExtensionPrefix[2])) {
1207       case VEX_PREFIX_66:
1208         attrMask |= ATTR_OPSIZE;
1209         break;
1210       case VEX_PREFIX_F3:
1211         attrMask |= ATTR_XS;
1212         break;
1213       case VEX_PREFIX_F2:
1214         attrMask |= ATTR_XD;
1215         break;
1216       }
1217 
1218       if (zFromEVEX4of4(insn->vectorExtensionPrefix[3]))
1219         attrMask |= ATTR_EVEXKZ;
1220       if (bFromEVEX4of4(insn->vectorExtensionPrefix[3]))
1221         attrMask |= ATTR_EVEXB;
1222       if (isNF(insn) && !readModRM(insn) &&
1223           !isCCMPOrCTEST(insn)) // NF bit is the MSB of aaa.
1224         attrMask |= ATTR_EVEXNF;
1225       // aaa is not used a opmask in MAP4
1226       else if (aaaFromEVEX4of4(insn->vectorExtensionPrefix[3]) &&
1227                (insn->opcodeType != MAP4))
1228         attrMask |= ATTR_EVEXK;
1229       if (lFromEVEX4of4(insn->vectorExtensionPrefix[3]))
1230         attrMask |= ATTR_VEXL;
1231       if (l2FromEVEX4of4(insn->vectorExtensionPrefix[3]))
1232         attrMask |= ATTR_EVEXL2;
1233     } else if (insn->vectorExtensionType == TYPE_VEX_3B) {
1234       switch (ppFromVEX3of3(insn->vectorExtensionPrefix[2])) {
1235       case VEX_PREFIX_66:
1236         attrMask |= ATTR_OPSIZE;
1237         break;
1238       case VEX_PREFIX_F3:
1239         attrMask |= ATTR_XS;
1240         break;
1241       case VEX_PREFIX_F2:
1242         attrMask |= ATTR_XD;
1243         break;
1244       }
1245 
1246       if (lFromVEX3of3(insn->vectorExtensionPrefix[2]))
1247         attrMask |= ATTR_VEXL;
1248     } else if (insn->vectorExtensionType == TYPE_VEX_2B) {
1249       switch (ppFromVEX2of2(insn->vectorExtensionPrefix[1])) {
1250       case VEX_PREFIX_66:
1251         attrMask |= ATTR_OPSIZE;
1252         if (insn->hasAdSize)
1253           attrMask |= ATTR_ADSIZE;
1254         break;
1255       case VEX_PREFIX_F3:
1256         attrMask |= ATTR_XS;
1257         break;
1258       case VEX_PREFIX_F2:
1259         attrMask |= ATTR_XD;
1260         break;
1261       }
1262 
1263       if (lFromVEX2of2(insn->vectorExtensionPrefix[1]))
1264         attrMask |= ATTR_VEXL;
1265     } else if (insn->vectorExtensionType == TYPE_XOP) {
1266       switch (ppFromXOP3of3(insn->vectorExtensionPrefix[2])) {
1267       case VEX_PREFIX_66:
1268         attrMask |= ATTR_OPSIZE;
1269         break;
1270       case VEX_PREFIX_F3:
1271         attrMask |= ATTR_XS;
1272         break;
1273       case VEX_PREFIX_F2:
1274         attrMask |= ATTR_XD;
1275         break;
1276       }
1277 
1278       if (lFromXOP3of3(insn->vectorExtensionPrefix[2]))
1279         attrMask |= ATTR_VEXL;
1280     } else {
1281       return -1;
1282     }
1283   } else if (!insn->mandatoryPrefix) {
1284     // If we don't have mandatory prefix we should use legacy prefixes here
1285     if (insn->hasOpSize && (insn->mode != MODE_16BIT))
1286       attrMask |= ATTR_OPSIZE;
1287     if (insn->hasAdSize)
1288       attrMask |= ATTR_ADSIZE;
1289     if (insn->opcodeType == ONEBYTE) {
1290       if (insn->repeatPrefix == 0xf3 && (insn->opcode == 0x90))
1291         // Special support for PAUSE
1292         attrMask |= ATTR_XS;
1293     } else {
1294       if (insn->repeatPrefix == 0xf2)
1295         attrMask |= ATTR_XD;
1296       else if (insn->repeatPrefix == 0xf3)
1297         attrMask |= ATTR_XS;
1298     }
1299   } else {
1300     switch (insn->mandatoryPrefix) {
1301     case 0xf2:
1302       attrMask |= ATTR_XD;
1303       break;
1304     case 0xf3:
1305       attrMask |= ATTR_XS;
1306       break;
1307     case 0x66:
1308       if (insn->mode != MODE_16BIT)
1309         attrMask |= ATTR_OPSIZE;
1310       if (insn->hasAdSize)
1311         attrMask |= ATTR_ADSIZE;
1312       break;
1313     case 0x67:
1314       attrMask |= ATTR_ADSIZE;
1315       break;
1316     }
1317   }
1318 
1319   if (insn->rexPrefix & 0x08) {
1320     attrMask |= ATTR_REXW;
1321     attrMask &= ~ATTR_ADSIZE;
1322   }
1323 
1324   // Absolute jump and pushp/popp need special handling
1325   if (insn->rex2ExtensionPrefix[0] == 0xd5 && insn->opcodeType == ONEBYTE &&
1326       (insn->opcode == 0xA1 || (insn->opcode & 0xf0) == 0x50))
1327     attrMask |= ATTR_REX2;
1328 
1329   if (insn->mode == MODE_16BIT) {
1330     // JCXZ/JECXZ need special handling for 16-bit mode because the meaning
1331     // of the AdSize prefix is inverted w.r.t. 32-bit mode.
1332     if (insn->opcodeType == ONEBYTE && insn->opcode == 0xE3)
1333       attrMask ^= ATTR_ADSIZE;
1334     // If we're in 16-bit mode and this is one of the relative jumps and opsize
1335     // prefix isn't present, we need to force the opsize attribute since the
1336     // prefix is inverted relative to 32-bit mode.
1337     if (!insn->hasOpSize && insn->opcodeType == ONEBYTE &&
1338         (insn->opcode == 0xE8 || insn->opcode == 0xE9))
1339       attrMask |= ATTR_OPSIZE;
1340 
1341     if (!insn->hasOpSize && insn->opcodeType == TWOBYTE &&
1342         insn->opcode >= 0x80 && insn->opcode <= 0x8F)
1343       attrMask |= ATTR_OPSIZE;
1344   }
1345 
1346 
1347   if (getInstructionIDWithAttrMask(&instructionID, insn, attrMask))
1348     return -1;
1349 
1350   // The following clauses compensate for limitations of the tables.
1351 
1352   if (insn->mode != MODE_64BIT &&
1353       insn->vectorExtensionType != TYPE_NO_VEX_XOP) {
1354     // The tables can't distinquish between cases where the W-bit is used to
1355     // select register size and cases where its a required part of the opcode.
1356     if ((insn->vectorExtensionType == TYPE_EVEX &&
1357          wFromEVEX3of4(insn->vectorExtensionPrefix[2])) ||
1358         (insn->vectorExtensionType == TYPE_VEX_3B &&
1359          wFromVEX3of3(insn->vectorExtensionPrefix[2])) ||
1360         (insn->vectorExtensionType == TYPE_XOP &&
1361          wFromXOP3of3(insn->vectorExtensionPrefix[2]))) {
1362 
1363       uint16_t instructionIDWithREXW;
1364       if (getInstructionIDWithAttrMask(&instructionIDWithREXW, insn,
1365                                        attrMask | ATTR_REXW)) {
1366         insn->instructionID = instructionID;
1367         insn->spec = &INSTRUCTIONS_SYM[instructionID];
1368         return 0;
1369       }
1370 
1371       auto SpecName = mii->getName(instructionIDWithREXW);
1372       // If not a 64-bit instruction. Switch the opcode.
1373       if (!is64Bit(SpecName.data())) {
1374         insn->instructionID = instructionIDWithREXW;
1375         insn->spec = &INSTRUCTIONS_SYM[instructionIDWithREXW];
1376         return 0;
1377       }
1378     }
1379   }
1380 
1381   // Absolute moves, umonitor, and movdir64b need special handling.
1382   // -For 16-bit mode because the meaning of the AdSize and OpSize prefixes are
1383   //  inverted w.r.t.
1384   // -For 32-bit mode we need to ensure the ADSIZE prefix is observed in
1385   //  any position.
1386   if ((insn->opcodeType == ONEBYTE && ((insn->opcode & 0xFC) == 0xA0)) ||
1387       (insn->opcodeType == TWOBYTE && (insn->opcode == 0xAE)) ||
1388       (insn->opcodeType == THREEBYTE_38 && insn->opcode == 0xF8) ||
1389       (insn->opcodeType == MAP4 && insn->opcode == 0xF8)) {
1390     // Make sure we observed the prefixes in any position.
1391     if (insn->hasAdSize)
1392       attrMask |= ATTR_ADSIZE;
1393     if (insn->hasOpSize)
1394       attrMask |= ATTR_OPSIZE;
1395 
1396     // In 16-bit, invert the attributes.
1397     if (insn->mode == MODE_16BIT) {
1398       attrMask ^= ATTR_ADSIZE;
1399 
1400       // The OpSize attribute is only valid with the absolute moves.
1401       if (insn->opcodeType == ONEBYTE && ((insn->opcode & 0xFC) == 0xA0))
1402         attrMask ^= ATTR_OPSIZE;
1403     }
1404 
1405     if (getInstructionIDWithAttrMask(&instructionID, insn, attrMask))
1406       return -1;
1407 
1408     insn->instructionID = instructionID;
1409     insn->spec = &INSTRUCTIONS_SYM[instructionID];
1410     return 0;
1411   }
1412 
1413   if ((insn->mode == MODE_16BIT || insn->hasOpSize) &&
1414       !(attrMask & ATTR_OPSIZE)) {
1415     // The instruction tables make no distinction between instructions that
1416     // allow OpSize anywhere (i.e., 16-bit operations) and that need it in a
1417     // particular spot (i.e., many MMX operations). In general we're
1418     // conservative, but in the specific case where OpSize is present but not in
1419     // the right place we check if there's a 16-bit operation.
1420     const struct InstructionSpecifier *spec;
1421     uint16_t instructionIDWithOpsize;
1422     llvm::StringRef specName, specWithOpSizeName;
1423 
1424     spec = &INSTRUCTIONS_SYM[instructionID];
1425 
1426     if (getInstructionIDWithAttrMask(&instructionIDWithOpsize, insn,
1427                                      attrMask | ATTR_OPSIZE)) {
1428       // ModRM required with OpSize but not present. Give up and return the
1429       // version without OpSize set.
1430       insn->instructionID = instructionID;
1431       insn->spec = spec;
1432       return 0;
1433     }
1434 
1435     specName = mii->getName(instructionID);
1436     specWithOpSizeName = mii->getName(instructionIDWithOpsize);
1437 
1438     if (is16BitEquivalent(specName.data(), specWithOpSizeName.data()) &&
1439         (insn->mode == MODE_16BIT) ^ insn->hasOpSize) {
1440       insn->instructionID = instructionIDWithOpsize;
1441       insn->spec = &INSTRUCTIONS_SYM[instructionIDWithOpsize];
1442     } else {
1443       insn->instructionID = instructionID;
1444       insn->spec = spec;
1445     }
1446     return 0;
1447   }
1448 
1449   if (insn->opcodeType == ONEBYTE && insn->opcode == 0x90 &&
1450       insn->rexPrefix & 0x01) {
1451     // NOOP shouldn't decode as NOOP if REX.b is set. Instead it should decode
1452     // as XCHG %r8, %eax.
1453     const struct InstructionSpecifier *spec;
1454     uint16_t instructionIDWithNewOpcode;
1455     const struct InstructionSpecifier *specWithNewOpcode;
1456 
1457     spec = &INSTRUCTIONS_SYM[instructionID];
1458 
1459     // Borrow opcode from one of the other XCHGar opcodes
1460     insn->opcode = 0x91;
1461 
1462     if (getInstructionIDWithAttrMask(&instructionIDWithNewOpcode, insn,
1463                                      attrMask)) {
1464       insn->opcode = 0x90;
1465 
1466       insn->instructionID = instructionID;
1467       insn->spec = spec;
1468       return 0;
1469     }
1470 
1471     specWithNewOpcode = &INSTRUCTIONS_SYM[instructionIDWithNewOpcode];
1472 
1473     // Change back
1474     insn->opcode = 0x90;
1475 
1476     insn->instructionID = instructionIDWithNewOpcode;
1477     insn->spec = specWithNewOpcode;
1478 
1479     return 0;
1480   }
1481 
1482   insn->instructionID = instructionID;
1483   insn->spec = &INSTRUCTIONS_SYM[insn->instructionID];
1484 
1485   return 0;
1486 }
1487 
1488 // Read an operand from the opcode field of an instruction and interprets it
1489 // appropriately given the operand width. Handles AddRegFrm instructions.
1490 //
1491 // @param insn  - the instruction whose opcode field is to be read.
1492 // @param size  - The width (in bytes) of the register being specified.
1493 //                1 means AL and friends, 2 means AX, 4 means EAX, and 8 means
1494 //                RAX.
1495 // @return      - 0 on success; nonzero otherwise.
readOpcodeRegister(struct InternalInstruction * insn,uint8_t size)1496 static int readOpcodeRegister(struct InternalInstruction *insn, uint8_t size) {
1497   LLVM_DEBUG(dbgs() << "readOpcodeRegister()");
1498 
1499   if (size == 0)
1500     size = insn->registerSize;
1501 
1502   auto setOpcodeRegister = [&](unsigned base) {
1503     insn->opcodeRegister =
1504         (Reg)(base + ((bFromREX(insn->rexPrefix) << 3) |
1505                       (b2FromREX2(insn->rex2ExtensionPrefix[1]) << 4) |
1506                       (insn->opcode & 7)));
1507   };
1508 
1509   switch (size) {
1510   case 1:
1511     setOpcodeRegister(MODRM_REG_AL);
1512     if (insn->rexPrefix && insn->opcodeRegister >= MODRM_REG_AL + 0x4 &&
1513         insn->opcodeRegister < MODRM_REG_AL + 0x8) {
1514       insn->opcodeRegister =
1515           (Reg)(MODRM_REG_SPL + (insn->opcodeRegister - MODRM_REG_AL - 4));
1516     }
1517 
1518     break;
1519   case 2:
1520     setOpcodeRegister(MODRM_REG_AX);
1521     break;
1522   case 4:
1523     setOpcodeRegister(MODRM_REG_EAX);
1524     break;
1525   case 8:
1526     setOpcodeRegister(MODRM_REG_RAX);
1527     break;
1528   }
1529 
1530   return 0;
1531 }
1532 
1533 // Consume an immediate operand from an instruction, given the desired operand
1534 // size.
1535 //
1536 // @param insn  - The instruction whose operand is to be read.
1537 // @param size  - The width (in bytes) of the operand.
1538 // @return      - 0 if the immediate was successfully consumed; nonzero
1539 //                otherwise.
readImmediate(struct InternalInstruction * insn,uint8_t size)1540 static int readImmediate(struct InternalInstruction *insn, uint8_t size) {
1541   uint8_t imm8;
1542   uint16_t imm16;
1543   uint32_t imm32;
1544   uint64_t imm64;
1545 
1546   LLVM_DEBUG(dbgs() << "readImmediate()");
1547 
1548   assert(insn->numImmediatesConsumed < 2 && "Already consumed two immediates");
1549 
1550   insn->immediateSize = size;
1551   insn->immediateOffset = insn->readerCursor - insn->startLocation;
1552 
1553   switch (size) {
1554   case 1:
1555     if (consume(insn, imm8))
1556       return -1;
1557     insn->immediates[insn->numImmediatesConsumed] = imm8;
1558     break;
1559   case 2:
1560     if (consume(insn, imm16))
1561       return -1;
1562     insn->immediates[insn->numImmediatesConsumed] = imm16;
1563     break;
1564   case 4:
1565     if (consume(insn, imm32))
1566       return -1;
1567     insn->immediates[insn->numImmediatesConsumed] = imm32;
1568     break;
1569   case 8:
1570     if (consume(insn, imm64))
1571       return -1;
1572     insn->immediates[insn->numImmediatesConsumed] = imm64;
1573     break;
1574   default:
1575     llvm_unreachable("invalid size");
1576   }
1577 
1578   insn->numImmediatesConsumed++;
1579 
1580   return 0;
1581 }
1582 
1583 // Consume vvvv from an instruction if it has a VEX prefix.
readVVVV(struct InternalInstruction * insn)1584 static int readVVVV(struct InternalInstruction *insn) {
1585   LLVM_DEBUG(dbgs() << "readVVVV()");
1586 
1587   int vvvv;
1588   if (insn->vectorExtensionType == TYPE_EVEX)
1589     vvvv = (v2FromEVEX4of4(insn->vectorExtensionPrefix[3]) << 4 |
1590             vvvvFromEVEX3of4(insn->vectorExtensionPrefix[2]));
1591   else if (insn->vectorExtensionType == TYPE_VEX_3B)
1592     vvvv = vvvvFromVEX3of3(insn->vectorExtensionPrefix[2]);
1593   else if (insn->vectorExtensionType == TYPE_VEX_2B)
1594     vvvv = vvvvFromVEX2of2(insn->vectorExtensionPrefix[1]);
1595   else if (insn->vectorExtensionType == TYPE_XOP)
1596     vvvv = vvvvFromXOP3of3(insn->vectorExtensionPrefix[2]);
1597   else
1598     return -1;
1599 
1600   if (insn->mode != MODE_64BIT)
1601     vvvv &= 0xf; // Can only clear bit 4. Bit 3 must be cleared later.
1602 
1603   insn->vvvv = static_cast<Reg>(vvvv);
1604   return 0;
1605 }
1606 
1607 // Read an mask register from the opcode field of an instruction.
1608 //
1609 // @param insn    - The instruction whose opcode field is to be read.
1610 // @return        - 0 on success; nonzero otherwise.
readMaskRegister(struct InternalInstruction * insn)1611 static int readMaskRegister(struct InternalInstruction *insn) {
1612   LLVM_DEBUG(dbgs() << "readMaskRegister()");
1613 
1614   if (insn->vectorExtensionType != TYPE_EVEX)
1615     return -1;
1616 
1617   insn->writemask =
1618       static_cast<Reg>(aaaFromEVEX4of4(insn->vectorExtensionPrefix[3]));
1619   return 0;
1620 }
1621 
1622 // Consults the specifier for an instruction and consumes all
1623 // operands for that instruction, interpreting them as it goes.
readOperands(struct InternalInstruction * insn)1624 static int readOperands(struct InternalInstruction *insn) {
1625   int hasVVVV, needVVVV;
1626   int sawRegImm = 0;
1627 
1628   LLVM_DEBUG(dbgs() << "readOperands()");
1629 
1630   // If non-zero vvvv specified, make sure one of the operands uses it.
1631   hasVVVV = !readVVVV(insn);
1632   needVVVV = hasVVVV && (insn->vvvv != 0);
1633 
1634   for (const auto &Op : x86OperandSets[insn->spec->operands]) {
1635     switch (Op.encoding) {
1636     case ENCODING_NONE:
1637     case ENCODING_SI:
1638     case ENCODING_DI:
1639       break;
1640     CASE_ENCODING_VSIB:
1641       // VSIB can use the V2 bit so check only the other bits.
1642       if (needVVVV)
1643         needVVVV = hasVVVV & ((insn->vvvv & 0xf) != 0);
1644       if (readModRM(insn))
1645         return -1;
1646 
1647       // Reject if SIB wasn't used.
1648       if (insn->eaBase != EA_BASE_sib && insn->eaBase != EA_BASE_sib64)
1649         return -1;
1650 
1651       // If sibIndex was set to SIB_INDEX_NONE, index offset is 4.
1652       if (insn->sibIndex == SIB_INDEX_NONE)
1653         insn->sibIndex = (SIBIndex)(insn->sibIndexBase + 4);
1654 
1655       // If EVEX.v2 is set this is one of the 16-31 registers.
1656       if (insn->vectorExtensionType == TYPE_EVEX && insn->mode == MODE_64BIT &&
1657           v2FromEVEX4of4(insn->vectorExtensionPrefix[3]))
1658         insn->sibIndex = (SIBIndex)(insn->sibIndex + 16);
1659 
1660       // Adjust the index register to the correct size.
1661       switch ((OperandType)Op.type) {
1662       default:
1663         debug("Unhandled VSIB index type");
1664         return -1;
1665       case TYPE_MVSIBX:
1666         insn->sibIndex =
1667             (SIBIndex)(SIB_INDEX_XMM0 + (insn->sibIndex - insn->sibIndexBase));
1668         break;
1669       case TYPE_MVSIBY:
1670         insn->sibIndex =
1671             (SIBIndex)(SIB_INDEX_YMM0 + (insn->sibIndex - insn->sibIndexBase));
1672         break;
1673       case TYPE_MVSIBZ:
1674         insn->sibIndex =
1675             (SIBIndex)(SIB_INDEX_ZMM0 + (insn->sibIndex - insn->sibIndexBase));
1676         break;
1677       }
1678 
1679       // Apply the AVX512 compressed displacement scaling factor.
1680       if (Op.encoding != ENCODING_REG && insn->eaDisplacement == EA_DISP_8)
1681         insn->displacement *= 1 << (Op.encoding - ENCODING_VSIB);
1682       break;
1683     case ENCODING_SIB:
1684       // Reject if SIB wasn't used.
1685       if (insn->eaBase != EA_BASE_sib && insn->eaBase != EA_BASE_sib64)
1686         return -1;
1687       if (readModRM(insn))
1688         return -1;
1689       if (fixupReg(insn, &Op))
1690         return -1;
1691       break;
1692     case ENCODING_REG:
1693     CASE_ENCODING_RM:
1694       if (readModRM(insn))
1695         return -1;
1696       if (fixupReg(insn, &Op))
1697         return -1;
1698       // Apply the AVX512 compressed displacement scaling factor.
1699       if (Op.encoding != ENCODING_REG && insn->eaDisplacement == EA_DISP_8)
1700         insn->displacement *= 1 << (Op.encoding - ENCODING_RM);
1701       break;
1702     case ENCODING_IB:
1703       if (sawRegImm) {
1704         // Saw a register immediate so don't read again and instead split the
1705         // previous immediate. FIXME: This is a hack.
1706         insn->immediates[insn->numImmediatesConsumed] =
1707             insn->immediates[insn->numImmediatesConsumed - 1] & 0xf;
1708         ++insn->numImmediatesConsumed;
1709         break;
1710       }
1711       if (readImmediate(insn, 1))
1712         return -1;
1713       if (Op.type == TYPE_XMM || Op.type == TYPE_YMM)
1714         sawRegImm = 1;
1715       break;
1716     case ENCODING_IW:
1717       if (readImmediate(insn, 2))
1718         return -1;
1719       break;
1720     case ENCODING_ID:
1721       if (readImmediate(insn, 4))
1722         return -1;
1723       break;
1724     case ENCODING_IO:
1725       if (readImmediate(insn, 8))
1726         return -1;
1727       break;
1728     case ENCODING_Iv:
1729       if (readImmediate(insn, insn->immediateSize))
1730         return -1;
1731       break;
1732     case ENCODING_Ia:
1733       if (readImmediate(insn, insn->addressSize))
1734         return -1;
1735       break;
1736     case ENCODING_IRC:
1737       insn->RC = (l2FromEVEX4of4(insn->vectorExtensionPrefix[3]) << 1) |
1738                  lFromEVEX4of4(insn->vectorExtensionPrefix[3]);
1739       break;
1740     case ENCODING_RB:
1741       if (readOpcodeRegister(insn, 1))
1742         return -1;
1743       break;
1744     case ENCODING_RW:
1745       if (readOpcodeRegister(insn, 2))
1746         return -1;
1747       break;
1748     case ENCODING_RD:
1749       if (readOpcodeRegister(insn, 4))
1750         return -1;
1751       break;
1752     case ENCODING_RO:
1753       if (readOpcodeRegister(insn, 8))
1754         return -1;
1755       break;
1756     case ENCODING_Rv:
1757       if (readOpcodeRegister(insn, 0))
1758         return -1;
1759       break;
1760     case ENCODING_CF:
1761       insn->immediates[1] = oszcFromEVEX3of4(insn->vectorExtensionPrefix[2]);
1762       needVVVV = false; // oszc shares the same bits with VVVV
1763       break;
1764     case ENCODING_CC:
1765       if (isCCMPOrCTEST(insn))
1766         insn->immediates[2] = scFromEVEX4of4(insn->vectorExtensionPrefix[3]);
1767       else
1768         insn->immediates[1] = insn->opcode & 0xf;
1769       break;
1770     case ENCODING_FP:
1771       break;
1772     case ENCODING_VVVV:
1773       needVVVV = 0; // Mark that we have found a VVVV operand.
1774       if (!hasVVVV)
1775         return -1;
1776       if (insn->mode != MODE_64BIT)
1777         insn->vvvv = static_cast<Reg>(insn->vvvv & 0x7);
1778       if (fixupReg(insn, &Op))
1779         return -1;
1780       break;
1781     case ENCODING_WRITEMASK:
1782       if (readMaskRegister(insn))
1783         return -1;
1784       break;
1785     case ENCODING_DUP:
1786       break;
1787     default:
1788       LLVM_DEBUG(dbgs() << "Encountered an operand with an unknown encoding.");
1789       return -1;
1790     }
1791   }
1792 
1793   // If we didn't find ENCODING_VVVV operand, but non-zero vvvv present, fail
1794   if (needVVVV)
1795     return -1;
1796 
1797   return 0;
1798 }
1799 
1800 namespace llvm {
1801 
1802 // Fill-ins to make the compiler happy. These constants are never actually
1803 // assigned; they are just filler to make an automatically-generated switch
1804 // statement work.
1805 namespace X86 {
1806   enum {
1807     BX_SI = 500,
1808     BX_DI = 501,
1809     BP_SI = 502,
1810     BP_DI = 503,
1811     sib   = 504,
1812     sib64 = 505
1813   };
1814 } // namespace X86
1815 
1816 } // namespace llvm
1817 
1818 static bool translateInstruction(MCInst &target,
1819                                 InternalInstruction &source,
1820                                 const MCDisassembler *Dis);
1821 
1822 namespace {
1823 
1824 /// Generic disassembler for all X86 platforms. All each platform class should
1825 /// have to do is subclass the constructor, and provide a different
1826 /// disassemblerMode value.
1827 class X86GenericDisassembler : public MCDisassembler {
1828   std::unique_ptr<const MCInstrInfo> MII;
1829 public:
1830   X86GenericDisassembler(const MCSubtargetInfo &STI, MCContext &Ctx,
1831                          std::unique_ptr<const MCInstrInfo> MII);
1832 public:
1833   DecodeStatus getInstruction(MCInst &instr, uint64_t &size,
1834                               ArrayRef<uint8_t> Bytes, uint64_t Address,
1835                               raw_ostream &cStream) const override;
1836 
1837 private:
1838   DisassemblerMode              fMode;
1839 };
1840 
1841 } // namespace
1842 
X86GenericDisassembler(const MCSubtargetInfo & STI,MCContext & Ctx,std::unique_ptr<const MCInstrInfo> MII)1843 X86GenericDisassembler::X86GenericDisassembler(
1844                                          const MCSubtargetInfo &STI,
1845                                          MCContext &Ctx,
1846                                          std::unique_ptr<const MCInstrInfo> MII)
1847   : MCDisassembler(STI, Ctx), MII(std::move(MII)) {
1848   const FeatureBitset &FB = STI.getFeatureBits();
1849   if (FB[X86::Is16Bit]) {
1850     fMode = MODE_16BIT;
1851     return;
1852   } else if (FB[X86::Is32Bit]) {
1853     fMode = MODE_32BIT;
1854     return;
1855   } else if (FB[X86::Is64Bit]) {
1856     fMode = MODE_64BIT;
1857     return;
1858   }
1859 
1860   llvm_unreachable("Invalid CPU mode");
1861 }
1862 
getInstruction(MCInst & Instr,uint64_t & Size,ArrayRef<uint8_t> Bytes,uint64_t Address,raw_ostream & CStream) const1863 MCDisassembler::DecodeStatus X86GenericDisassembler::getInstruction(
1864     MCInst &Instr, uint64_t &Size, ArrayRef<uint8_t> Bytes, uint64_t Address,
1865     raw_ostream &CStream) const {
1866   CommentStream = &CStream;
1867 
1868   InternalInstruction Insn;
1869   memset(&Insn, 0, sizeof(InternalInstruction));
1870   Insn.bytes = Bytes;
1871   Insn.startLocation = Address;
1872   Insn.readerCursor = Address;
1873   Insn.mode = fMode;
1874 
1875   if (Bytes.empty() || readPrefixes(&Insn) || readOpcode(&Insn) ||
1876       getInstructionID(&Insn, MII.get()) || Insn.instructionID == 0 ||
1877       readOperands(&Insn)) {
1878     Size = Insn.readerCursor - Address;
1879     return Fail;
1880   }
1881 
1882   Insn.operands = x86OperandSets[Insn.spec->operands];
1883   Insn.length = Insn.readerCursor - Insn.startLocation;
1884   Size = Insn.length;
1885   if (Size > 15)
1886     LLVM_DEBUG(dbgs() << "Instruction exceeds 15-byte limit");
1887 
1888   bool Ret = translateInstruction(Instr, Insn, this);
1889   if (!Ret) {
1890     unsigned Flags = X86::IP_NO_PREFIX;
1891     if (Insn.hasAdSize)
1892       Flags |= X86::IP_HAS_AD_SIZE;
1893     if (!Insn.mandatoryPrefix) {
1894       if (Insn.hasOpSize)
1895         Flags |= X86::IP_HAS_OP_SIZE;
1896       if (Insn.repeatPrefix == 0xf2)
1897         Flags |= X86::IP_HAS_REPEAT_NE;
1898       else if (Insn.repeatPrefix == 0xf3 &&
1899                // It should not be 'pause' f3 90
1900                Insn.opcode != 0x90)
1901         Flags |= X86::IP_HAS_REPEAT;
1902       if (Insn.hasLockPrefix)
1903         Flags |= X86::IP_HAS_LOCK;
1904     }
1905     Instr.setFlags(Flags);
1906   }
1907   return (!Ret) ? Success : Fail;
1908 }
1909 
1910 //
1911 // Private code that translates from struct InternalInstructions to MCInsts.
1912 //
1913 
1914 /// translateRegister - Translates an internal register to the appropriate LLVM
1915 ///   register, and appends it as an operand to an MCInst.
1916 ///
1917 /// @param mcInst     - The MCInst to append to.
1918 /// @param reg        - The Reg to append.
translateRegister(MCInst & mcInst,Reg reg)1919 static void translateRegister(MCInst &mcInst, Reg reg) {
1920 #define ENTRY(x) X86::x,
1921   static constexpr MCPhysReg llvmRegnums[] = {ALL_REGS};
1922 #undef ENTRY
1923 
1924   MCPhysReg llvmRegnum = llvmRegnums[reg];
1925   mcInst.addOperand(MCOperand::createReg(llvmRegnum));
1926 }
1927 
1928 static const uint8_t segmentRegnums[SEG_OVERRIDE_max] = {
1929   0,        // SEG_OVERRIDE_NONE
1930   X86::CS,
1931   X86::SS,
1932   X86::DS,
1933   X86::ES,
1934   X86::FS,
1935   X86::GS
1936 };
1937 
1938 /// translateSrcIndex   - Appends a source index operand to an MCInst.
1939 ///
1940 /// @param mcInst       - The MCInst to append to.
1941 /// @param insn         - The internal instruction.
translateSrcIndex(MCInst & mcInst,InternalInstruction & insn)1942 static bool translateSrcIndex(MCInst &mcInst, InternalInstruction &insn) {
1943   unsigned baseRegNo;
1944 
1945   if (insn.mode == MODE_64BIT)
1946     baseRegNo = insn.hasAdSize ? X86::ESI : X86::RSI;
1947   else if (insn.mode == MODE_32BIT)
1948     baseRegNo = insn.hasAdSize ? X86::SI : X86::ESI;
1949   else {
1950     assert(insn.mode == MODE_16BIT);
1951     baseRegNo = insn.hasAdSize ? X86::ESI : X86::SI;
1952   }
1953   MCOperand baseReg = MCOperand::createReg(baseRegNo);
1954   mcInst.addOperand(baseReg);
1955 
1956   MCOperand segmentReg;
1957   segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
1958   mcInst.addOperand(segmentReg);
1959   return false;
1960 }
1961 
1962 /// translateDstIndex   - Appends a destination index operand to an MCInst.
1963 ///
1964 /// @param mcInst       - The MCInst to append to.
1965 /// @param insn         - The internal instruction.
1966 
translateDstIndex(MCInst & mcInst,InternalInstruction & insn)1967 static bool translateDstIndex(MCInst &mcInst, InternalInstruction &insn) {
1968   unsigned baseRegNo;
1969 
1970   if (insn.mode == MODE_64BIT)
1971     baseRegNo = insn.hasAdSize ? X86::EDI : X86::RDI;
1972   else if (insn.mode == MODE_32BIT)
1973     baseRegNo = insn.hasAdSize ? X86::DI : X86::EDI;
1974   else {
1975     assert(insn.mode == MODE_16BIT);
1976     baseRegNo = insn.hasAdSize ? X86::EDI : X86::DI;
1977   }
1978   MCOperand baseReg = MCOperand::createReg(baseRegNo);
1979   mcInst.addOperand(baseReg);
1980   return false;
1981 }
1982 
1983 /// translateImmediate  - Appends an immediate operand to an MCInst.
1984 ///
1985 /// @param mcInst       - The MCInst to append to.
1986 /// @param immediate    - The immediate value to append.
1987 /// @param operand      - The operand, as stored in the descriptor table.
1988 /// @param insn         - The internal instruction.
translateImmediate(MCInst & mcInst,uint64_t immediate,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)1989 static void translateImmediate(MCInst &mcInst, uint64_t immediate,
1990                                const OperandSpecifier &operand,
1991                                InternalInstruction &insn,
1992                                const MCDisassembler *Dis) {
1993   // Sign-extend the immediate if necessary.
1994 
1995   OperandType type = (OperandType)operand.type;
1996 
1997   bool isBranch = false;
1998   uint64_t pcrel = 0;
1999   if (type == TYPE_REL) {
2000     isBranch = true;
2001     pcrel = insn.startLocation + insn.length;
2002     switch (operand.encoding) {
2003     default:
2004       break;
2005     case ENCODING_Iv:
2006       switch (insn.displacementSize) {
2007       default:
2008         break;
2009       case 1:
2010         if(immediate & 0x80)
2011           immediate |= ~(0xffull);
2012         break;
2013       case 2:
2014         if(immediate & 0x8000)
2015           immediate |= ~(0xffffull);
2016         break;
2017       case 4:
2018         if(immediate & 0x80000000)
2019           immediate |= ~(0xffffffffull);
2020         break;
2021       case 8:
2022         break;
2023       }
2024       break;
2025     case ENCODING_IB:
2026       if(immediate & 0x80)
2027         immediate |= ~(0xffull);
2028       break;
2029     case ENCODING_IW:
2030       if(immediate & 0x8000)
2031         immediate |= ~(0xffffull);
2032       break;
2033     case ENCODING_ID:
2034       if(immediate & 0x80000000)
2035         immediate |= ~(0xffffffffull);
2036       break;
2037     }
2038   }
2039   // By default sign-extend all X86 immediates based on their encoding.
2040   else if (type == TYPE_IMM) {
2041     switch (operand.encoding) {
2042     default:
2043       break;
2044     case ENCODING_IB:
2045       if(immediate & 0x80)
2046         immediate |= ~(0xffull);
2047       break;
2048     case ENCODING_IW:
2049       if(immediate & 0x8000)
2050         immediate |= ~(0xffffull);
2051       break;
2052     case ENCODING_ID:
2053       if(immediate & 0x80000000)
2054         immediate |= ~(0xffffffffull);
2055       break;
2056     case ENCODING_IO:
2057       break;
2058     }
2059   }
2060 
2061   switch (type) {
2062   case TYPE_XMM:
2063     mcInst.addOperand(MCOperand::createReg(X86::XMM0 + (immediate >> 4)));
2064     return;
2065   case TYPE_YMM:
2066     mcInst.addOperand(MCOperand::createReg(X86::YMM0 + (immediate >> 4)));
2067     return;
2068   case TYPE_ZMM:
2069     mcInst.addOperand(MCOperand::createReg(X86::ZMM0 + (immediate >> 4)));
2070     return;
2071   default:
2072     // operand is 64 bits wide.  Do nothing.
2073     break;
2074   }
2075 
2076   if (!Dis->tryAddingSymbolicOperand(
2077           mcInst, immediate + pcrel, insn.startLocation, isBranch,
2078           insn.immediateOffset, insn.immediateSize, insn.length))
2079     mcInst.addOperand(MCOperand::createImm(immediate));
2080 
2081   if (type == TYPE_MOFFS) {
2082     MCOperand segmentReg;
2083     segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
2084     mcInst.addOperand(segmentReg);
2085   }
2086 }
2087 
2088 /// translateRMRegister - Translates a register stored in the R/M field of the
2089 ///   ModR/M byte to its LLVM equivalent and appends it to an MCInst.
2090 /// @param mcInst       - The MCInst to append to.
2091 /// @param insn         - The internal instruction to extract the R/M field
2092 ///                       from.
2093 /// @return             - 0 on success; -1 otherwise
translateRMRegister(MCInst & mcInst,InternalInstruction & insn)2094 static bool translateRMRegister(MCInst &mcInst,
2095                                 InternalInstruction &insn) {
2096   if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
2097     debug("A R/M register operand may not have a SIB byte");
2098     return true;
2099   }
2100 
2101   switch (insn.eaBase) {
2102   default:
2103     debug("Unexpected EA base register");
2104     return true;
2105   case EA_BASE_NONE:
2106     debug("EA_BASE_NONE for ModR/M base");
2107     return true;
2108 #define ENTRY(x) case EA_BASE_##x:
2109   ALL_EA_BASES
2110 #undef ENTRY
2111     debug("A R/M register operand may not have a base; "
2112           "the operand must be a register.");
2113     return true;
2114 #define ENTRY(x)                                                      \
2115   case EA_REG_##x:                                                    \
2116     mcInst.addOperand(MCOperand::createReg(X86::x)); break;
2117   ALL_REGS
2118 #undef ENTRY
2119   }
2120 
2121   return false;
2122 }
2123 
2124 /// translateRMMemory - Translates a memory operand stored in the Mod and R/M
2125 ///   fields of an internal instruction (and possibly its SIB byte) to a memory
2126 ///   operand in LLVM's format, and appends it to an MCInst.
2127 ///
2128 /// @param mcInst       - The MCInst to append to.
2129 /// @param insn         - The instruction to extract Mod, R/M, and SIB fields
2130 ///                       from.
2131 /// @param ForceSIB     - The instruction must use SIB.
2132 /// @return             - 0 on success; nonzero otherwise
translateRMMemory(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis,bool ForceSIB=false)2133 static bool translateRMMemory(MCInst &mcInst, InternalInstruction &insn,
2134                               const MCDisassembler *Dis,
2135                               bool ForceSIB = false) {
2136   // Addresses in an MCInst are represented as five operands:
2137   //   1. basereg       (register)  The R/M base, or (if there is a SIB) the
2138   //                                SIB base
2139   //   2. scaleamount   (immediate) 1, or (if there is a SIB) the specified
2140   //                                scale amount
2141   //   3. indexreg      (register)  x86_registerNONE, or (if there is a SIB)
2142   //                                the index (which is multiplied by the
2143   //                                scale amount)
2144   //   4. displacement  (immediate) 0, or the displacement if there is one
2145   //   5. segmentreg    (register)  x86_registerNONE for now, but could be set
2146   //                                if we have segment overrides
2147 
2148   MCOperand baseReg;
2149   MCOperand scaleAmount;
2150   MCOperand indexReg;
2151   MCOperand displacement;
2152   MCOperand segmentReg;
2153   uint64_t pcrel = 0;
2154 
2155   if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
2156     if (insn.sibBase != SIB_BASE_NONE) {
2157       switch (insn.sibBase) {
2158       default:
2159         debug("Unexpected sibBase");
2160         return true;
2161 #define ENTRY(x)                                          \
2162       case SIB_BASE_##x:                                  \
2163         baseReg = MCOperand::createReg(X86::x); break;
2164       ALL_SIB_BASES
2165 #undef ENTRY
2166       }
2167     } else {
2168       baseReg = MCOperand::createReg(X86::NoRegister);
2169     }
2170 
2171     if (insn.sibIndex != SIB_INDEX_NONE) {
2172       switch (insn.sibIndex) {
2173       default:
2174         debug("Unexpected sibIndex");
2175         return true;
2176 #define ENTRY(x)                                          \
2177       case SIB_INDEX_##x:                                 \
2178         indexReg = MCOperand::createReg(X86::x); break;
2179       EA_BASES_32BIT
2180       EA_BASES_64BIT
2181       REGS_XMM
2182       REGS_YMM
2183       REGS_ZMM
2184 #undef ENTRY
2185       }
2186     } else {
2187       // Use EIZ/RIZ for a few ambiguous cases where the SIB byte is present,
2188       // but no index is used and modrm alone should have been enough.
2189       // -No base register in 32-bit mode. In 64-bit mode this is used to
2190       //  avoid rip-relative addressing.
2191       // -Any base register used other than ESP/RSP/R12D/R12. Using these as a
2192       //  base always requires a SIB byte.
2193       // -A scale other than 1 is used.
2194       if (!ForceSIB &&
2195           (insn.sibScale != 1 ||
2196            (insn.sibBase == SIB_BASE_NONE && insn.mode != MODE_64BIT) ||
2197            (insn.sibBase != SIB_BASE_NONE &&
2198             insn.sibBase != SIB_BASE_ESP && insn.sibBase != SIB_BASE_RSP &&
2199             insn.sibBase != SIB_BASE_R12D && insn.sibBase != SIB_BASE_R12))) {
2200         indexReg = MCOperand::createReg(insn.addressSize == 4 ? X86::EIZ :
2201                                                                 X86::RIZ);
2202       } else
2203         indexReg = MCOperand::createReg(X86::NoRegister);
2204     }
2205 
2206     scaleAmount = MCOperand::createImm(insn.sibScale);
2207   } else {
2208     switch (insn.eaBase) {
2209     case EA_BASE_NONE:
2210       if (insn.eaDisplacement == EA_DISP_NONE) {
2211         debug("EA_BASE_NONE and EA_DISP_NONE for ModR/M base");
2212         return true;
2213       }
2214       if (insn.mode == MODE_64BIT){
2215         pcrel = insn.startLocation + insn.length;
2216         Dis->tryAddingPcLoadReferenceComment(insn.displacement + pcrel,
2217                                              insn.startLocation +
2218                                                  insn.displacementOffset);
2219         // Section 2.2.1.6
2220         baseReg = MCOperand::createReg(insn.addressSize == 4 ? X86::EIP :
2221                                                                X86::RIP);
2222       }
2223       else
2224         baseReg = MCOperand::createReg(X86::NoRegister);
2225 
2226       indexReg = MCOperand::createReg(X86::NoRegister);
2227       break;
2228     case EA_BASE_BX_SI:
2229       baseReg = MCOperand::createReg(X86::BX);
2230       indexReg = MCOperand::createReg(X86::SI);
2231       break;
2232     case EA_BASE_BX_DI:
2233       baseReg = MCOperand::createReg(X86::BX);
2234       indexReg = MCOperand::createReg(X86::DI);
2235       break;
2236     case EA_BASE_BP_SI:
2237       baseReg = MCOperand::createReg(X86::BP);
2238       indexReg = MCOperand::createReg(X86::SI);
2239       break;
2240     case EA_BASE_BP_DI:
2241       baseReg = MCOperand::createReg(X86::BP);
2242       indexReg = MCOperand::createReg(X86::DI);
2243       break;
2244     default:
2245       indexReg = MCOperand::createReg(X86::NoRegister);
2246       switch (insn.eaBase) {
2247       default:
2248         debug("Unexpected eaBase");
2249         return true;
2250         // Here, we will use the fill-ins defined above.  However,
2251         //   BX_SI, BX_DI, BP_SI, and BP_DI are all handled above and
2252         //   sib and sib64 were handled in the top-level if, so they're only
2253         //   placeholders to keep the compiler happy.
2254 #define ENTRY(x)                                        \
2255       case EA_BASE_##x:                                 \
2256         baseReg = MCOperand::createReg(X86::x); break;
2257       ALL_EA_BASES
2258 #undef ENTRY
2259 #define ENTRY(x) case EA_REG_##x:
2260       ALL_REGS
2261 #undef ENTRY
2262         debug("A R/M memory operand may not be a register; "
2263               "the base field must be a base.");
2264         return true;
2265       }
2266     }
2267 
2268     scaleAmount = MCOperand::createImm(1);
2269   }
2270 
2271   displacement = MCOperand::createImm(insn.displacement);
2272 
2273   segmentReg = MCOperand::createReg(segmentRegnums[insn.segmentOverride]);
2274 
2275   mcInst.addOperand(baseReg);
2276   mcInst.addOperand(scaleAmount);
2277   mcInst.addOperand(indexReg);
2278 
2279   const uint8_t dispSize =
2280       (insn.eaDisplacement == EA_DISP_NONE) ? 0 : insn.displacementSize;
2281 
2282   if (!Dis->tryAddingSymbolicOperand(
2283           mcInst, insn.displacement + pcrel, insn.startLocation, false,
2284           insn.displacementOffset, dispSize, insn.length))
2285     mcInst.addOperand(displacement);
2286   mcInst.addOperand(segmentReg);
2287   return false;
2288 }
2289 
2290 /// translateRM - Translates an operand stored in the R/M (and possibly SIB)
2291 ///   byte of an instruction to LLVM form, and appends it to an MCInst.
2292 ///
2293 /// @param mcInst       - The MCInst to append to.
2294 /// @param operand      - The operand, as stored in the descriptor table.
2295 /// @param insn         - The instruction to extract Mod, R/M, and SIB fields
2296 ///                       from.
2297 /// @return             - 0 on success; nonzero otherwise
translateRM(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)2298 static bool translateRM(MCInst &mcInst, const OperandSpecifier &operand,
2299                         InternalInstruction &insn, const MCDisassembler *Dis) {
2300   switch (operand.type) {
2301   default:
2302     debug("Unexpected type for a R/M operand");
2303     return true;
2304   case TYPE_R8:
2305   case TYPE_R16:
2306   case TYPE_R32:
2307   case TYPE_R64:
2308   case TYPE_Rv:
2309   case TYPE_MM64:
2310   case TYPE_XMM:
2311   case TYPE_YMM:
2312   case TYPE_ZMM:
2313   case TYPE_TMM:
2314   case TYPE_VK_PAIR:
2315   case TYPE_VK:
2316   case TYPE_DEBUGREG:
2317   case TYPE_CONTROLREG:
2318   case TYPE_BNDR:
2319     return translateRMRegister(mcInst, insn);
2320   case TYPE_M:
2321   case TYPE_MVSIBX:
2322   case TYPE_MVSIBY:
2323   case TYPE_MVSIBZ:
2324     return translateRMMemory(mcInst, insn, Dis);
2325   case TYPE_MSIB:
2326     return translateRMMemory(mcInst, insn, Dis, true);
2327   }
2328 }
2329 
2330 /// translateFPRegister - Translates a stack position on the FPU stack to its
2331 ///   LLVM form, and appends it to an MCInst.
2332 ///
2333 /// @param mcInst       - The MCInst to append to.
2334 /// @param stackPos     - The stack position to translate.
translateFPRegister(MCInst & mcInst,uint8_t stackPos)2335 static void translateFPRegister(MCInst &mcInst,
2336                                 uint8_t stackPos) {
2337   mcInst.addOperand(MCOperand::createReg(X86::ST0 + stackPos));
2338 }
2339 
2340 /// translateMaskRegister - Translates a 3-bit mask register number to
2341 ///   LLVM form, and appends it to an MCInst.
2342 ///
2343 /// @param mcInst       - The MCInst to append to.
2344 /// @param maskRegNum   - Number of mask register from 0 to 7.
2345 /// @return             - false on success; true otherwise.
translateMaskRegister(MCInst & mcInst,uint8_t maskRegNum)2346 static bool translateMaskRegister(MCInst &mcInst,
2347                                 uint8_t maskRegNum) {
2348   if (maskRegNum >= 8) {
2349     debug("Invalid mask register number");
2350     return true;
2351   }
2352 
2353   mcInst.addOperand(MCOperand::createReg(X86::K0 + maskRegNum));
2354   return false;
2355 }
2356 
2357 /// translateOperand - Translates an operand stored in an internal instruction
2358 ///   to LLVM's format and appends it to an MCInst.
2359 ///
2360 /// @param mcInst       - The MCInst to append to.
2361 /// @param operand      - The operand, as stored in the descriptor table.
2362 /// @param insn         - The internal instruction.
2363 /// @return             - false on success; true otherwise.
translateOperand(MCInst & mcInst,const OperandSpecifier & operand,InternalInstruction & insn,const MCDisassembler * Dis)2364 static bool translateOperand(MCInst &mcInst, const OperandSpecifier &operand,
2365                              InternalInstruction &insn,
2366                              const MCDisassembler *Dis) {
2367   switch (operand.encoding) {
2368   default:
2369     debug("Unhandled operand encoding during translation");
2370     return true;
2371   case ENCODING_REG:
2372     translateRegister(mcInst, insn.reg);
2373     return false;
2374   case ENCODING_WRITEMASK:
2375     return translateMaskRegister(mcInst, insn.writemask);
2376   case ENCODING_SIB:
2377   CASE_ENCODING_RM:
2378   CASE_ENCODING_VSIB:
2379     return translateRM(mcInst, operand, insn, Dis);
2380   case ENCODING_IB:
2381   case ENCODING_IW:
2382   case ENCODING_ID:
2383   case ENCODING_IO:
2384   case ENCODING_Iv:
2385   case ENCODING_Ia:
2386     translateImmediate(mcInst,
2387                        insn.immediates[insn.numImmediatesTranslated++],
2388                        operand,
2389                        insn,
2390                        Dis);
2391     return false;
2392   case ENCODING_IRC:
2393     mcInst.addOperand(MCOperand::createImm(insn.RC));
2394     return false;
2395   case ENCODING_SI:
2396     return translateSrcIndex(mcInst, insn);
2397   case ENCODING_DI:
2398     return translateDstIndex(mcInst, insn);
2399   case ENCODING_RB:
2400   case ENCODING_RW:
2401   case ENCODING_RD:
2402   case ENCODING_RO:
2403   case ENCODING_Rv:
2404     translateRegister(mcInst, insn.opcodeRegister);
2405     return false;
2406   case ENCODING_CF:
2407     mcInst.addOperand(MCOperand::createImm(insn.immediates[1]));
2408     return false;
2409   case ENCODING_CC:
2410     if (isCCMPOrCTEST(&insn))
2411       mcInst.addOperand(MCOperand::createImm(insn.immediates[2]));
2412     else
2413       mcInst.addOperand(MCOperand::createImm(insn.immediates[1]));
2414     return false;
2415   case ENCODING_FP:
2416     translateFPRegister(mcInst, insn.modRM & 7);
2417     return false;
2418   case ENCODING_VVVV:
2419     translateRegister(mcInst, insn.vvvv);
2420     return false;
2421   case ENCODING_DUP:
2422     return translateOperand(mcInst, insn.operands[operand.type - TYPE_DUP0],
2423                             insn, Dis);
2424   }
2425 }
2426 
2427 /// translateInstruction - Translates an internal instruction and all its
2428 ///   operands to an MCInst.
2429 ///
2430 /// @param mcInst       - The MCInst to populate with the instruction's data.
2431 /// @param insn         - The internal instruction.
2432 /// @return             - false on success; true otherwise.
translateInstruction(MCInst & mcInst,InternalInstruction & insn,const MCDisassembler * Dis)2433 static bool translateInstruction(MCInst &mcInst,
2434                                 InternalInstruction &insn,
2435                                 const MCDisassembler *Dis) {
2436   if (!insn.spec) {
2437     debug("Instruction has no specification");
2438     return true;
2439   }
2440 
2441   mcInst.clear();
2442   mcInst.setOpcode(insn.instructionID);
2443   // If when reading the prefix bytes we determined the overlapping 0xf2 or 0xf3
2444   // prefix bytes should be disassembled as xrelease and xacquire then set the
2445   // opcode to those instead of the rep and repne opcodes.
2446   if (insn.xAcquireRelease) {
2447     if(mcInst.getOpcode() == X86::REP_PREFIX)
2448       mcInst.setOpcode(X86::XRELEASE_PREFIX);
2449     else if(mcInst.getOpcode() == X86::REPNE_PREFIX)
2450       mcInst.setOpcode(X86::XACQUIRE_PREFIX);
2451   }
2452 
2453   insn.numImmediatesTranslated = 0;
2454 
2455   for (const auto &Op : insn.operands) {
2456     if (Op.encoding != ENCODING_NONE) {
2457       if (translateOperand(mcInst, Op, insn, Dis)) {
2458         return true;
2459       }
2460     }
2461   }
2462 
2463   return false;
2464 }
2465 
createX86Disassembler(const Target & T,const MCSubtargetInfo & STI,MCContext & Ctx)2466 static MCDisassembler *createX86Disassembler(const Target &T,
2467                                              const MCSubtargetInfo &STI,
2468                                              MCContext &Ctx) {
2469   std::unique_ptr<const MCInstrInfo> MII(T.createMCInstrInfo());
2470   return new X86GenericDisassembler(STI, Ctx, std::move(MII));
2471 }
2472 
LLVMInitializeX86Disassembler()2473 extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeX86Disassembler() {
2474   // Register the disassembler.
2475   TargetRegistry::RegisterMCDisassembler(getTheX86_32Target(),
2476                                          createX86Disassembler);
2477   TargetRegistry::RegisterMCDisassembler(getTheX86_64Target(),
2478                                          createX86Disassembler);
2479 }
2480