Bug Summary

File:lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp
Warning:line 1754, column 33
The result of the '<<' expression is undefined

Annotated Source Code

1//===-- X86DisassemblerDecoder.cpp - Disassembler decoder -----------------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file is part of the X86 Disassembler.
11// It contains the implementation of the instruction decoder.
12// Documentation for the disassembler can be found in X86Disassembler.h.
13//
14//===----------------------------------------------------------------------===//
15
16#include <cstdarg> /* for va_*() */
17#include <cstdio> /* for vsnprintf() */
18#include <cstdlib> /* for exit() */
19#include <cstring> /* for memset() */
20
21#include "X86DisassemblerDecoder.h"
22
23using namespace llvm::X86Disassembler;
24
25/// Specifies whether a ModR/M byte is needed and (if so) which
26/// instruction each possible value of the ModR/M byte corresponds to. Once
27/// this information is known, we have narrowed down to a single instruction.
28struct ModRMDecision {
29 uint8_t modrm_type;
30 uint16_t instructionIDs;
31};
32
33/// Specifies which set of ModR/M->instruction tables to look at
34/// given a particular opcode.
35struct OpcodeDecision {
36 ModRMDecision modRMDecisions[256];
37};
38
39/// Specifies which opcode->instruction tables to look at given
40/// a particular context (set of attributes). Since there are many possible
41/// contexts, the decoder first uses CONTEXTS_SYM to determine which context
42/// applies given a specific set of attributes. Hence there are only IC_max
43/// entries in this table, rather than 2^(ATTR_max).
44struct ContextDecision {
45 OpcodeDecision opcodeDecisions[IC_max];
46};
47
48#include "X86GenDisassemblerTables.inc"
49
50#ifndef NDEBUG
51#define debug(s)do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 51, s); } while (0)
do { Debug(__FILE__"/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp", __LINE__51, s); } while (0)
52#else
53#define debug(s)do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 53, s); } while (0)
do { } while (0)
54#endif
55
56/*
57 * contextForAttrs - Client for the instruction context table. Takes a set of
58 * attributes and returns the appropriate decode context.
59 *
60 * @param attrMask - Attributes, from the enumeration attributeBits.
61 * @return - The InstructionContext to use when looking up an
62 * an instruction with these attributes.
63 */
64static InstructionContext contextForAttrs(uint16_t attrMask) {
65 return static_cast<InstructionContext>(CONTEXTS_SYMx86DisassemblerContexts[attrMask]);
66}
67
68/*
69 * modRMRequired - Reads the appropriate instruction table to determine whether
70 * the ModR/M byte is required to decode a particular instruction.
71 *
72 * @param type - The opcode type (i.e., how many bytes it has).
73 * @param insnContext - The context for the instruction, as returned by
74 * contextForAttrs.
75 * @param opcode - The last byte of the instruction's opcode, not counting
76 * ModR/M extensions and escapes.
77 * @return - true if the ModR/M byte is required, false otherwise.
78 */
79static int modRMRequired(OpcodeType type,
80 InstructionContext insnContext,
81 uint16_t opcode) {
82 const struct ContextDecision* decision = nullptr;
83
84 switch (type) {
85 case ONEBYTE:
86 decision = &ONEBYTE_SYMx86DisassemblerOneByteOpcodes;
87 break;
88 case TWOBYTE:
89 decision = &TWOBYTE_SYMx86DisassemblerTwoByteOpcodes;
90 break;
91 case THREEBYTE_38:
92 decision = &THREEBYTE38_SYMx86DisassemblerThreeByte38Opcodes;
93 break;
94 case THREEBYTE_3A:
95 decision = &THREEBYTE3A_SYMx86DisassemblerThreeByte3AOpcodes;
96 break;
97 case XOP8_MAP:
98 decision = &XOP8_MAP_SYMx86DisassemblerXOP8Opcodes;
99 break;
100 case XOP9_MAP:
101 decision = &XOP9_MAP_SYMx86DisassemblerXOP9Opcodes;
102 break;
103 case XOPA_MAP:
104 decision = &XOPA_MAP_SYMx86DisassemblerXOPAOpcodes;
105 break;
106 }
107
108 return decision->opcodeDecisions[insnContext].modRMDecisions[opcode].
109 modrm_type != MODRM_ONEENTRY;
110}
111
112/*
113 * decode - Reads the appropriate instruction table to obtain the unique ID of
114 * an instruction.
115 *
116 * @param type - See modRMRequired().
117 * @param insnContext - See modRMRequired().
118 * @param opcode - See modRMRequired().
119 * @param modRM - The ModR/M byte if required, or any value if not.
120 * @return - The UID of the instruction, or 0 on failure.
121 */
122static InstrUID decode(OpcodeType type,
123 InstructionContext insnContext,
124 uint8_t opcode,
125 uint8_t modRM) {
126 const struct ModRMDecision* dec = nullptr;
127
128 switch (type) {
129 case ONEBYTE:
130 dec = &ONEBYTE_SYMx86DisassemblerOneByteOpcodes.opcodeDecisions[insnContext].modRMDecisions[opcode];
131 break;
132 case TWOBYTE:
133 dec = &TWOBYTE_SYMx86DisassemblerTwoByteOpcodes.opcodeDecisions[insnContext].modRMDecisions[opcode];
134 break;
135 case THREEBYTE_38:
136 dec = &THREEBYTE38_SYMx86DisassemblerThreeByte38Opcodes.opcodeDecisions[insnContext].modRMDecisions[opcode];
137 break;
138 case THREEBYTE_3A:
139 dec = &THREEBYTE3A_SYMx86DisassemblerThreeByte3AOpcodes.opcodeDecisions[insnContext].modRMDecisions[opcode];
140 break;
141 case XOP8_MAP:
142 dec = &XOP8_MAP_SYMx86DisassemblerXOP8Opcodes.opcodeDecisions[insnContext].modRMDecisions[opcode];
143 break;
144 case XOP9_MAP:
145 dec = &XOP9_MAP_SYMx86DisassemblerXOP9Opcodes.opcodeDecisions[insnContext].modRMDecisions[opcode];
146 break;
147 case XOPA_MAP:
148 dec = &XOPA_MAP_SYMx86DisassemblerXOPAOpcodes.opcodeDecisions[insnContext].modRMDecisions[opcode];
149 break;
150 }
151
152 switch (dec->modrm_type) {
153 default:
154 debug("Corrupt table! Unknown modrm_type")do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 154, "Corrupt table! Unknown modrm_type"); } while (0)
;
155 return 0;
156 case MODRM_ONEENTRY:
157 return modRMTable[dec->instructionIDs];
158 case MODRM_SPLITRM:
159 if (modFromModRM(modRM)(((modRM) & 0xc0) >> 6) == 0x3)
160 return modRMTable[dec->instructionIDs+1];
161 return modRMTable[dec->instructionIDs];
162 case MODRM_SPLITREG:
163 if (modFromModRM(modRM)(((modRM) & 0xc0) >> 6) == 0x3)
164 return modRMTable[dec->instructionIDs+((modRM & 0x38) >> 3)+8];
165 return modRMTable[dec->instructionIDs+((modRM & 0x38) >> 3)];
166 case MODRM_SPLITMISC:
167 if (modFromModRM(modRM)(((modRM) & 0xc0) >> 6) == 0x3)
168 return modRMTable[dec->instructionIDs+(modRM & 0x3f)+8];
169 return modRMTable[dec->instructionIDs+((modRM & 0x38) >> 3)];
170 case MODRM_FULL:
171 return modRMTable[dec->instructionIDs+modRM];
172 }
173}
174
175/*
176 * specifierForUID - Given a UID, returns the name and operand specification for
177 * that instruction.
178 *
179 * @param uid - The unique ID for the instruction. This should be returned by
180 * decode(); specifierForUID will not check bounds.
181 * @return - A pointer to the specification for that instruction.
182 */
183static const struct InstructionSpecifier *specifierForUID(InstrUID uid) {
184 return &INSTRUCTIONS_SYMx86DisassemblerInstrSpecifiers[uid];
185}
186
187/*
188 * consumeByte - Uses the reader function provided by the user to consume one
189 * byte from the instruction's memory and advance the cursor.
190 *
191 * @param insn - The instruction with the reader function to use. The cursor
192 * for this instruction is advanced.
193 * @param byte - A pointer to a pre-allocated memory buffer to be populated
194 * with the data read.
195 * @return - 0 if the read was successful; nonzero otherwise.
196 */
197static int consumeByte(struct InternalInstruction* insn, uint8_t* byte) {
198 int ret = insn->reader(insn->readerArg, byte, insn->readerCursor);
199
200 if (!ret)
201 ++(insn->readerCursor);
202
203 return ret;
204}
205
206/*
207 * lookAtByte - Like consumeByte, but does not advance the cursor.
208 *
209 * @param insn - See consumeByte().
210 * @param byte - See consumeByte().
211 * @return - See consumeByte().
212 */
213static int lookAtByte(struct InternalInstruction* insn, uint8_t* byte) {
214 return insn->reader(insn->readerArg, byte, insn->readerCursor);
215}
216
217static void unconsumeByte(struct InternalInstruction* insn) {
218 insn->readerCursor--;
219}
220
221#define CONSUME_FUNC(name, type)static int name(struct InternalInstruction* insn, type* ptr) {
type combined = 0; unsigned offset; for (offset = 0; offset <
sizeof(type); ++offset) { uint8_t byte; int ret = insn->reader
(insn->readerArg, &byte, insn->readerCursor + offset
); if (ret) return ret; combined = combined | ((uint64_t)byte
<< (offset * 8)); } *ptr = combined; insn->readerCursor
+= sizeof(type); return 0; }
\
222 static int name(struct InternalInstruction* insn, type* ptr) { \
223 type combined = 0; \
224 unsigned offset; \
225 for (offset = 0; offset < sizeof(type); ++offset) { \
226 uint8_t byte; \
227 int ret = insn->reader(insn->readerArg, \
228 &byte, \
229 insn->readerCursor + offset); \
230 if (ret) \
231 return ret; \
232 combined = combined | ((uint64_t)byte << (offset * 8)); \
233 } \
234 *ptr = combined; \
235 insn->readerCursor += sizeof(type); \
236 return 0; \
237 }
238
239/*
240 * consume* - Use the reader function provided by the user to consume data
241 * values of various sizes from the instruction's memory and advance the
242 * cursor appropriately. These readers perform endian conversion.
243 *
244 * @param insn - See consumeByte().
245 * @param ptr - A pointer to a pre-allocated memory of appropriate size to
246 * be populated with the data read.
247 * @return - See consumeByte().
248 */
249CONSUME_FUNC(consumeInt8, int8_t)static int consumeInt8(struct InternalInstruction* insn, int8_t
* ptr) { int8_t combined = 0; unsigned offset; for (offset = 0
; offset < sizeof(int8_t); ++offset) { uint8_t byte; int ret
= insn->reader(insn->readerArg, &byte, insn->readerCursor
+ offset); if (ret) return ret; combined = combined | ((uint64_t
)byte << (offset * 8)); } *ptr = combined; insn->readerCursor
+= sizeof(int8_t); return 0; }
250CONSUME_FUNC(consumeInt16, int16_t)static int consumeInt16(struct InternalInstruction* insn, int16_t
* ptr) { int16_t combined = 0; unsigned offset; for (offset =
0; offset < sizeof(int16_t); ++offset) { uint8_t byte; int
ret = insn->reader(insn->readerArg, &byte, insn->
readerCursor + offset); if (ret) return ret; combined = combined
| ((uint64_t)byte << (offset * 8)); } *ptr = combined;
insn->readerCursor += sizeof(int16_t); return 0; }
251CONSUME_FUNC(consumeInt32, int32_t)static int consumeInt32(struct InternalInstruction* insn, int32_t
* ptr) { int32_t combined = 0; unsigned offset; for (offset =
0; offset < sizeof(int32_t); ++offset) { uint8_t byte; int
ret = insn->reader(insn->readerArg, &byte, insn->
readerCursor + offset); if (ret) return ret; combined = combined
| ((uint64_t)byte << (offset * 8)); } *ptr = combined;
insn->readerCursor += sizeof(int32_t); return 0; }
252CONSUME_FUNC(consumeUInt16, uint16_t)static int consumeUInt16(struct InternalInstruction* insn, uint16_t
* ptr) { uint16_t combined = 0; unsigned offset; for (offset =
0; offset < sizeof(uint16_t); ++offset) { uint8_t byte; int
ret = insn->reader(insn->readerArg, &byte, insn->
readerCursor + offset); if (ret) return ret; combined = combined
| ((uint64_t)byte << (offset * 8)); } *ptr = combined;
insn->readerCursor += sizeof(uint16_t); return 0; }
253CONSUME_FUNC(consumeUInt32, uint32_t)static int consumeUInt32(struct InternalInstruction* insn, uint32_t
* ptr) { uint32_t combined = 0; unsigned offset; for (offset =
0; offset < sizeof(uint32_t); ++offset) { uint8_t byte; int
ret = insn->reader(insn->readerArg, &byte, insn->
readerCursor + offset); if (ret) return ret; combined = combined
| ((uint64_t)byte << (offset * 8)); } *ptr = combined;
insn->readerCursor += sizeof(uint32_t); return 0; }
254CONSUME_FUNC(consumeUInt64, uint64_t)static int consumeUInt64(struct InternalInstruction* insn, uint64_t
* ptr) { uint64_t combined = 0; unsigned offset; for (offset =
0; offset < sizeof(uint64_t); ++offset) { uint8_t byte; int
ret = insn->reader(insn->readerArg, &byte, insn->
readerCursor + offset); if (ret) return ret; combined = combined
| ((uint64_t)byte << (offset * 8)); } *ptr = combined;
insn->readerCursor += sizeof(uint64_t); return 0; }
255
256/*
257 * dbgprintf - Uses the logging function provided by the user to log a single
258 * message, typically without a carriage-return.
259 *
260 * @param insn - The instruction containing the logging function.
261 * @param format - See printf().
262 * @param ... - See printf().
263 */
264static void dbgprintf(struct InternalInstruction* insn,
265 const char* format,
266 ...) {
267 char buffer[256];
268 va_list ap;
269
270 if (!insn->dlog)
271 return;
272
273 va_start(ap, format)__builtin_va_start(ap, format);
274 (void)vsnprintf(buffer, sizeof(buffer), format, ap);
275 va_end(ap)__builtin_va_end(ap);
276
277 insn->dlog(insn->dlogArg, buffer);
278}
279
280/*
281 * setPrefixPresent - Marks that a particular prefix is present at a particular
282 * location.
283 *
284 * @param insn - The instruction to be marked as having the prefix.
285 * @param prefix - The prefix that is present.
286 * @param location - The location where the prefix is located (in the address
287 * space of the instruction's reader).
288 */
289static void setPrefixPresent(struct InternalInstruction* insn,
290 uint8_t prefix,
291 uint64_t location)
292{
293 insn->prefixPresent[prefix] = 1;
294 insn->prefixLocations[prefix] = location;
295}
296
297/*
298 * isPrefixAtLocation - Queries an instruction to determine whether a prefix is
299 * present at a given location.
300 *
301 * @param insn - The instruction to be queried.
302 * @param prefix - The prefix.
303 * @param location - The location to query.
304 * @return - Whether the prefix is at that location.
305 */
306static bool isPrefixAtLocation(struct InternalInstruction* insn,
307 uint8_t prefix,
308 uint64_t location)
309{
310 return insn->prefixPresent[prefix] == 1 &&
311 insn->prefixLocations[prefix] == location;
312}
313
314/*
315 * readPrefixes - Consumes all of an instruction's prefix bytes, and marks the
316 * instruction as having them. Also sets the instruction's default operand,
317 * address, and other relevant data sizes to report operands correctly.
318 *
319 * @param insn - The instruction whose prefixes are to be read.
320 * @return - 0 if the instruction could be read until the end of the prefix
321 * bytes, and no prefixes conflicted; nonzero otherwise.
322 */
323static int readPrefixes(struct InternalInstruction* insn) {
324 bool isPrefix = true;
325 bool prefixGroups[4] = { false };
326 uint64_t prefixLocation;
327 uint8_t byte = 0;
328 uint8_t nextByte;
329
330 bool hasAdSize = false;
331 bool hasOpSize = false;
332
333 dbgprintf(insn, "readPrefixes()");
334
335 while (isPrefix) {
336 prefixLocation = insn->readerCursor;
337
338 /* If we fail reading prefixes, just stop here and let the opcode reader deal with it */
339 if (consumeByte(insn, &byte))
340 break;
341
342 /*
343 * If the byte is a LOCK/REP/REPNE prefix and not a part of the opcode, then
344 * break and let it be disassembled as a normal "instruction".
345 */
346 if (insn->readerCursor - 1 == insn->startLocation && byte == 0xf0)
347 break;
348
349 if (insn->readerCursor - 1 == insn->startLocation
350 && (byte == 0xf2 || byte == 0xf3)
351 && !lookAtByte(insn, &nextByte))
352 {
353 /*
354 * If the byte is 0xf2 or 0xf3, and any of the following conditions are
355 * met:
356 * - it is followed by a LOCK (0xf0) prefix
357 * - it is followed by an xchg instruction
358 * then it should be disassembled as a xacquire/xrelease not repne/rep.
359 */
360 if ((byte == 0xf2 || byte == 0xf3) &&
361 ((nextByte == 0xf0) ||
362 ((nextByte & 0xfe) == 0x86 || (nextByte & 0xf8) == 0x90)))
363 insn->xAcquireRelease = true;
364 /*
365 * Also if the byte is 0xf3, and the following condition is met:
366 * - it is followed by a "mov mem, reg" (opcode 0x88/0x89) or
367 * "mov mem, imm" (opcode 0xc6/0xc7) instructions.
368 * then it should be disassembled as an xrelease not rep.
369 */
370 if (byte == 0xf3 &&
371 (nextByte == 0x88 || nextByte == 0x89 ||
372 nextByte == 0xc6 || nextByte == 0xc7))
373 insn->xAcquireRelease = true;
374 if (insn->mode == MODE_64BIT && (nextByte & 0xf0) == 0x40) {
375 if (consumeByte(insn, &nextByte))
376 return -1;
377 if (lookAtByte(insn, &nextByte))
378 return -1;
379 unconsumeByte(insn);
380 }
381 if (nextByte != 0x0f && nextByte != 0x90)
382 break;
383 }
384
385 switch (byte) {
386 case 0xf0: /* LOCK */
387 case 0xf2: /* REPNE/REPNZ */
388 case 0xf3: /* REP or REPE/REPZ */
389 if (prefixGroups[0])
390 dbgprintf(insn, "Redundant Group 1 prefix");
391 prefixGroups[0] = true;
392 setPrefixPresent(insn, byte, prefixLocation);
393 break;
394 case 0x2e: /* CS segment override -OR- Branch not taken */
395 case 0x36: /* SS segment override -OR- Branch taken */
396 case 0x3e: /* DS segment override */
397 case 0x26: /* ES segment override */
398 case 0x64: /* FS segment override */
399 case 0x65: /* GS segment override */
400 switch (byte) {
401 case 0x2e:
402 insn->segmentOverride = SEG_OVERRIDE_CS;
403 break;
404 case 0x36:
405 insn->segmentOverride = SEG_OVERRIDE_SS;
406 break;
407 case 0x3e:
408 insn->segmentOverride = SEG_OVERRIDE_DS;
409 break;
410 case 0x26:
411 insn->segmentOverride = SEG_OVERRIDE_ES;
412 break;
413 case 0x64:
414 insn->segmentOverride = SEG_OVERRIDE_FS;
415 break;
416 case 0x65:
417 insn->segmentOverride = SEG_OVERRIDE_GS;
418 break;
419 default:
420 debug("Unhandled override")do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 420, "Unhandled override"); } while (0)
;
421 return -1;
422 }
423 if (prefixGroups[1])
424 dbgprintf(insn, "Redundant Group 2 prefix");
425 prefixGroups[1] = true;
426 setPrefixPresent(insn, byte, prefixLocation);
427 break;
428 case 0x66: /* Operand-size override */
429 if (prefixGroups[2])
430 dbgprintf(insn, "Redundant Group 3 prefix");
431 prefixGroups[2] = true;
432 hasOpSize = true;
433 setPrefixPresent(insn, byte, prefixLocation);
434 break;
435 case 0x67: /* Address-size override */
436 if (prefixGroups[3])
437 dbgprintf(insn, "Redundant Group 4 prefix");
438 prefixGroups[3] = true;
439 hasAdSize = true;
440 setPrefixPresent(insn, byte, prefixLocation);
441 break;
442 default: /* Not a prefix byte */
443 isPrefix = false;
444 break;
445 }
446
447 if (isPrefix)
448 dbgprintf(insn, "Found prefix 0x%hhx", byte);
449 }
450
451 insn->vectorExtensionType = TYPE_NO_VEX_XOP;
452
453 if (byte == 0x62) {
454 uint8_t byte1, byte2;
455
456 if (consumeByte(insn, &byte1)) {
457 dbgprintf(insn, "Couldn't read second byte of EVEX prefix");
458 return -1;
459 }
460
461 if (lookAtByte(insn, &byte2)) {
462 dbgprintf(insn, "Couldn't read third byte of EVEX prefix");
463 return -1;
464 }
465
466 if ((insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) &&
467 ((~byte1 & 0xc) == 0xc) && ((byte2 & 0x4) == 0x4)) {
468 insn->vectorExtensionType = TYPE_EVEX;
469 } else {
470 unconsumeByte(insn); /* unconsume byte1 */
471 unconsumeByte(insn); /* unconsume byte */
472 insn->necessaryPrefixLocation = insn->readerCursor - 2;
473 }
474
475 if (insn->vectorExtensionType == TYPE_EVEX) {
476 insn->vectorExtensionPrefix[0] = byte;
477 insn->vectorExtensionPrefix[1] = byte1;
478 if (consumeByte(insn, &insn->vectorExtensionPrefix[2])) {
479 dbgprintf(insn, "Couldn't read third byte of EVEX prefix");
480 return -1;
481 }
482 if (consumeByte(insn, &insn->vectorExtensionPrefix[3])) {
483 dbgprintf(insn, "Couldn't read fourth byte of EVEX prefix");
484 return -1;
485 }
486
487 /* We simulate the REX prefix for simplicity's sake */
488 if (insn->mode == MODE_64BIT) {
489 insn->rexPrefix = 0x40
490 | (wFromEVEX3of4(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x80) >> 7) << 3)
491 | (rFromEVEX2of4(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x80) >>
7)
<< 2)
492 | (xFromEVEX2of4(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x40) >>
6)
<< 1)
493 | (bFromEVEX2of4(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x20) >>
5)
<< 0);
494 }
495
496 dbgprintf(insn, "Found EVEX prefix 0x%hhx 0x%hhx 0x%hhx 0x%hhx",
497 insn->vectorExtensionPrefix[0], insn->vectorExtensionPrefix[1],
498 insn->vectorExtensionPrefix[2], insn->vectorExtensionPrefix[3]);
499 }
500 } else if (byte == 0xc4) {
501 uint8_t byte1;
502
503 if (lookAtByte(insn, &byte1)) {
504 dbgprintf(insn, "Couldn't read second byte of VEX");
505 return -1;
506 }
507
508 if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) {
509 insn->vectorExtensionType = TYPE_VEX_3B;
510 insn->necessaryPrefixLocation = insn->readerCursor - 1;
511 } else {
512 unconsumeByte(insn);
513 insn->necessaryPrefixLocation = insn->readerCursor - 1;
514 }
515
516 if (insn->vectorExtensionType == TYPE_VEX_3B) {
517 insn->vectorExtensionPrefix[0] = byte;
518 consumeByte(insn, &insn->vectorExtensionPrefix[1]);
519 consumeByte(insn, &insn->vectorExtensionPrefix[2]);
520
521 /* We simulate the REX prefix for simplicity's sake */
522
523 if (insn->mode == MODE_64BIT) {
524 insn->rexPrefix = 0x40
525 | (wFromVEX3of3(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x80) >> 7) << 3)
526 | (rFromVEX2of3(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x80) >>
7)
<< 2)
527 | (xFromVEX2of3(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x40) >>
6)
<< 1)
528 | (bFromVEX2of3(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x20) >>
5)
<< 0);
529 }
530
531 dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx 0x%hhx",
532 insn->vectorExtensionPrefix[0], insn->vectorExtensionPrefix[1],
533 insn->vectorExtensionPrefix[2]);
534 }
535 } else if (byte == 0xc5) {
536 uint8_t byte1;
537
538 if (lookAtByte(insn, &byte1)) {
539 dbgprintf(insn, "Couldn't read second byte of VEX");
540 return -1;
541 }
542
543 if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) {
544 insn->vectorExtensionType = TYPE_VEX_2B;
545 } else {
546 unconsumeByte(insn);
547 }
548
549 if (insn->vectorExtensionType == TYPE_VEX_2B) {
550 insn->vectorExtensionPrefix[0] = byte;
551 consumeByte(insn, &insn->vectorExtensionPrefix[1]);
552
553 if (insn->mode == MODE_64BIT) {
554 insn->rexPrefix = 0x40
555 | (rFromVEX2of2(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x80) >>
7)
<< 2);
556 }
557
558 switch (ppFromVEX2of2(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x3)) {
559 default:
560 break;
561 case VEX_PREFIX_66:
562 hasOpSize = true;
563 break;
564 }
565
566 dbgprintf(insn, "Found VEX prefix 0x%hhx 0x%hhx",
567 insn->vectorExtensionPrefix[0],
568 insn->vectorExtensionPrefix[1]);
569 }
570 } else if (byte == 0x8f) {
571 uint8_t byte1;
572
573 if (lookAtByte(insn, &byte1)) {
574 dbgprintf(insn, "Couldn't read second byte of XOP");
575 return -1;
576 }
577
578 if ((byte1 & 0x38) != 0x0) { /* 0 in these 3 bits is a POP instruction. */
579 insn->vectorExtensionType = TYPE_XOP;
580 insn->necessaryPrefixLocation = insn->readerCursor - 1;
581 } else {
582 unconsumeByte(insn);
583 insn->necessaryPrefixLocation = insn->readerCursor - 1;
584 }
585
586 if (insn->vectorExtensionType == TYPE_XOP) {
587 insn->vectorExtensionPrefix[0] = byte;
588 consumeByte(insn, &insn->vectorExtensionPrefix[1]);
589 consumeByte(insn, &insn->vectorExtensionPrefix[2]);
590
591 /* We simulate the REX prefix for simplicity's sake */
592
593 if (insn->mode == MODE_64BIT) {
594 insn->rexPrefix = 0x40
595 | (wFromXOP3of3(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x80) >> 7) << 3)
596 | (rFromXOP2of3(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x80) >>
7)
<< 2)
597 | (xFromXOP2of3(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x40) >>
6)
<< 1)
598 | (bFromXOP2of3(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x20) >>
5)
<< 0);
599 }
600
601 switch (ppFromXOP3of3(insn->vectorExtensionPrefix[2])((insn->vectorExtensionPrefix[2]) & 0x3)) {
602 default:
603 break;
604 case VEX_PREFIX_66:
605 hasOpSize = true;
606 break;
607 }
608
609 dbgprintf(insn, "Found XOP prefix 0x%hhx 0x%hhx 0x%hhx",
610 insn->vectorExtensionPrefix[0], insn->vectorExtensionPrefix[1],
611 insn->vectorExtensionPrefix[2]);
612 }
613 } else {
614 if (insn->mode == MODE_64BIT) {
615 if ((byte & 0xf0) == 0x40) {
616 uint8_t opcodeByte;
617
618 if (lookAtByte(insn, &opcodeByte) || ((opcodeByte & 0xf0) == 0x40)) {
619 dbgprintf(insn, "Redundant REX prefix");
620 return -1;
621 }
622
623 insn->rexPrefix = byte;
624 insn->necessaryPrefixLocation = insn->readerCursor - 2;
625
626 dbgprintf(insn, "Found REX prefix 0x%hhx", byte);
627 } else {
628 unconsumeByte(insn);
629 insn->necessaryPrefixLocation = insn->readerCursor - 1;
630 }
631 } else {
632 unconsumeByte(insn);
633 insn->necessaryPrefixLocation = insn->readerCursor - 1;
634 }
635 }
636
637 if (insn->mode == MODE_16BIT) {
638 insn->registerSize = (hasOpSize ? 4 : 2);
639 insn->addressSize = (hasAdSize ? 4 : 2);
640 insn->displacementSize = (hasAdSize ? 4 : 2);
641 insn->immediateSize = (hasOpSize ? 4 : 2);
642 } else if (insn->mode == MODE_32BIT) {
643 insn->registerSize = (hasOpSize ? 2 : 4);
644 insn->addressSize = (hasAdSize ? 2 : 4);
645 insn->displacementSize = (hasAdSize ? 2 : 4);
646 insn->immediateSize = (hasOpSize ? 2 : 4);
647 } else if (insn->mode == MODE_64BIT) {
648 if (insn->rexPrefix && wFromREX(insn->rexPrefix)(((insn->rexPrefix) & 0x8) >> 3)) {
649 insn->registerSize = 8;
650 insn->addressSize = (hasAdSize ? 4 : 8);
651 insn->displacementSize = 4;
652 insn->immediateSize = 4;
653 } else {
654 insn->registerSize = (hasOpSize ? 2 : 4);
655 insn->addressSize = (hasAdSize ? 4 : 8);
656 insn->displacementSize = (hasOpSize ? 2 : 4);
657 insn->immediateSize = (hasOpSize ? 2 : 4);
658 }
659 }
660
661 return 0;
662}
663
664/*
665 * readOpcode - Reads the opcode (excepting the ModR/M byte in the case of
666 * extended or escape opcodes).
667 *
668 * @param insn - The instruction whose opcode is to be read.
669 * @return - 0 if the opcode could be read successfully; nonzero otherwise.
670 */
671static int readOpcode(struct InternalInstruction* insn) {
672 /* Determine the length of the primary opcode */
673
674 uint8_t current;
675
676 dbgprintf(insn, "readOpcode()");
677
678 insn->opcodeType = ONEBYTE;
679
680 if (insn->vectorExtensionType == TYPE_EVEX) {
681 switch (mmFromEVEX2of4(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x3)) {
682 default:
683 dbgprintf(insn, "Unhandled mm field for instruction (0x%hhx)",
684 mmFromEVEX2of4(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x3));
685 return -1;
686 case VEX_LOB_0F:
687 insn->opcodeType = TWOBYTE;
688 return consumeByte(insn, &insn->opcode);
689 case VEX_LOB_0F38:
690 insn->opcodeType = THREEBYTE_38;
691 return consumeByte(insn, &insn->opcode);
692 case VEX_LOB_0F3A:
693 insn->opcodeType = THREEBYTE_3A;
694 return consumeByte(insn, &insn->opcode);
695 }
696 } else if (insn->vectorExtensionType == TYPE_VEX_3B) {
697 switch (mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x1f)) {
698 default:
699 dbgprintf(insn, "Unhandled m-mmmm field for instruction (0x%hhx)",
700 mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x1f));
701 return -1;
702 case VEX_LOB_0F:
703 insn->opcodeType = TWOBYTE;
704 return consumeByte(insn, &insn->opcode);
705 case VEX_LOB_0F38:
706 insn->opcodeType = THREEBYTE_38;
707 return consumeByte(insn, &insn->opcode);
708 case VEX_LOB_0F3A:
709 insn->opcodeType = THREEBYTE_3A;
710 return consumeByte(insn, &insn->opcode);
711 }
712 } else if (insn->vectorExtensionType == TYPE_VEX_2B) {
713 insn->opcodeType = TWOBYTE;
714 return consumeByte(insn, &insn->opcode);
715 } else if (insn->vectorExtensionType == TYPE_XOP) {
716 switch (mmmmmFromXOP2of3(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x1f)) {
717 default:
718 dbgprintf(insn, "Unhandled m-mmmm field for instruction (0x%hhx)",
719 mmmmmFromVEX2of3(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x1f));
720 return -1;
721 case XOP_MAP_SELECT_8:
722 insn->opcodeType = XOP8_MAP;
723 return consumeByte(insn, &insn->opcode);
724 case XOP_MAP_SELECT_9:
725 insn->opcodeType = XOP9_MAP;
726 return consumeByte(insn, &insn->opcode);
727 case XOP_MAP_SELECT_A:
728 insn->opcodeType = XOPA_MAP;
729 return consumeByte(insn, &insn->opcode);
730 }
731 }
732
733 if (consumeByte(insn, &current))
734 return -1;
735
736 if (current == 0x0f) {
737 dbgprintf(insn, "Found a two-byte escape prefix (0x%hhx)", current);
738
739 if (consumeByte(insn, &current))
740 return -1;
741
742 if (current == 0x38) {
743 dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current);
744
745 if (consumeByte(insn, &current))
746 return -1;
747
748 insn->opcodeType = THREEBYTE_38;
749 } else if (current == 0x3a) {
750 dbgprintf(insn, "Found a three-byte escape prefix (0x%hhx)", current);
751
752 if (consumeByte(insn, &current))
753 return -1;
754
755 insn->opcodeType = THREEBYTE_3A;
756 } else {
757 dbgprintf(insn, "Didn't find a three-byte escape prefix");
758
759 insn->opcodeType = TWOBYTE;
760 }
761 }
762
763 /*
764 * At this point we have consumed the full opcode.
765 * Anything we consume from here on must be unconsumed.
766 */
767
768 insn->opcode = current;
769
770 return 0;
771}
772
773static int readModRM(struct InternalInstruction* insn);
774
775/*
776 * getIDWithAttrMask - Determines the ID of an instruction, consuming
777 * the ModR/M byte as appropriate for extended and escape opcodes,
778 * and using a supplied attribute mask.
779 *
780 * @param instructionID - A pointer whose target is filled in with the ID of the
781 * instruction.
782 * @param insn - The instruction whose ID is to be determined.
783 * @param attrMask - The attribute mask to search.
784 * @return - 0 if the ModR/M could be read when needed or was not
785 * needed; nonzero otherwise.
786 */
787static int getIDWithAttrMask(uint16_t* instructionID,
788 struct InternalInstruction* insn,
789 uint16_t attrMask) {
790 bool hasModRMExtension;
791
792 InstructionContext instructionClass = contextForAttrs(attrMask);
793
794 hasModRMExtension = modRMRequired(insn->opcodeType,
795 instructionClass,
796 insn->opcode);
797
798 if (hasModRMExtension) {
799 if (readModRM(insn))
800 return -1;
801
802 *instructionID = decode(insn->opcodeType,
803 instructionClass,
804 insn->opcode,
805 insn->modRM);
806 } else {
807 *instructionID = decode(insn->opcodeType,
808 instructionClass,
809 insn->opcode,
810 0);
811 }
812
813 return 0;
814}
815
816/*
817 * is16BitEquivalent - Determines whether two instruction names refer to
818 * equivalent instructions but one is 16-bit whereas the other is not.
819 *
820 * @param orig - The instruction that is not 16-bit
821 * @param equiv - The instruction that is 16-bit
822 */
823static bool is16BitEquivalent(const char *orig, const char *equiv) {
824 off_t i;
825
826 for (i = 0;; i++) {
827 if (orig[i] == '\0' && equiv[i] == '\0')
828 return true;
829 if (orig[i] == '\0' || equiv[i] == '\0')
830 return false;
831 if (orig[i] != equiv[i]) {
832 if ((orig[i] == 'Q' || orig[i] == 'L') && equiv[i] == 'W')
833 continue;
834 if ((orig[i] == '6' || orig[i] == '3') && equiv[i] == '1')
835 continue;
836 if ((orig[i] == '4' || orig[i] == '2') && equiv[i] == '6')
837 continue;
838 return false;
839 }
840 }
841}
842
843/*
844 * is64Bit - Determines whether this instruction is a 64-bit instruction.
845 *
846 * @param name - The instruction that is not 16-bit
847 */
848static bool is64Bit(const char *name) {
849 off_t i;
850
851 for (i = 0;; ++i) {
852 if (name[i] == '\0')
853 return false;
854 if (name[i] == '6' && name[i+1] == '4')
855 return true;
856 }
857}
858
859/*
860 * getID - Determines the ID of an instruction, consuming the ModR/M byte as
861 * appropriate for extended and escape opcodes. Determines the attributes and
862 * context for the instruction before doing so.
863 *
864 * @param insn - The instruction whose ID is to be determined.
865 * @return - 0 if the ModR/M could be read when needed or was not needed;
866 * nonzero otherwise.
867 */
868static int getID(struct InternalInstruction* insn, const void *miiArg) {
869 uint16_t attrMask;
870 uint16_t instructionID;
871
872 dbgprintf(insn, "getID()");
873
874 attrMask = ATTR_NONE;
875
876 if (insn->mode == MODE_64BIT)
877 attrMask |= ATTR_64BIT;
878
879 if (insn->vectorExtensionType != TYPE_NO_VEX_XOP) {
880 attrMask |= (insn->vectorExtensionType == TYPE_EVEX) ? ATTR_EVEX : ATTR_VEX;
881
882 if (insn->vectorExtensionType == TYPE_EVEX) {
883 switch (ppFromEVEX3of4(insn->vectorExtensionPrefix[2])((insn->vectorExtensionPrefix[2]) & 0x3)) {
884 case VEX_PREFIX_66:
885 attrMask |= ATTR_OPSIZE;
886 break;
887 case VEX_PREFIX_F3:
888 attrMask |= ATTR_XS;
889 break;
890 case VEX_PREFIX_F2:
891 attrMask |= ATTR_XD;
892 break;
893 }
894
895 if (zFromEVEX4of4(insn->vectorExtensionPrefix[3])(((insn->vectorExtensionPrefix[3]) & 0x80) >> 7))
896 attrMask |= ATTR_EVEXKZ;
897 if (bFromEVEX4of4(insn->vectorExtensionPrefix[3])(((insn->vectorExtensionPrefix[3]) & 0x10) >> 4))
898 attrMask |= ATTR_EVEXB;
899 if (aaaFromEVEX4of4(insn->vectorExtensionPrefix[3])((insn->vectorExtensionPrefix[3]) & 0x7))
900 attrMask |= ATTR_EVEXK;
901 if (lFromEVEX4of4(insn->vectorExtensionPrefix[3])(((insn->vectorExtensionPrefix[3]) & 0x20) >> 5))
902 attrMask |= ATTR_EVEXL;
903 if (l2FromEVEX4of4(insn->vectorExtensionPrefix[3])(((insn->vectorExtensionPrefix[3]) & 0x40) >> 6))
904 attrMask |= ATTR_EVEXL2;
905 } else if (insn->vectorExtensionType == TYPE_VEX_3B) {
906 switch (ppFromVEX3of3(insn->vectorExtensionPrefix[2])((insn->vectorExtensionPrefix[2]) & 0x3)) {
907 case VEX_PREFIX_66:
908 attrMask |= ATTR_OPSIZE;
909 break;
910 case VEX_PREFIX_F3:
911 attrMask |= ATTR_XS;
912 break;
913 case VEX_PREFIX_F2:
914 attrMask |= ATTR_XD;
915 break;
916 }
917
918 if (lFromVEX3of3(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x4) >> 2))
919 attrMask |= ATTR_VEXL;
920 } else if (insn->vectorExtensionType == TYPE_VEX_2B) {
921 switch (ppFromVEX2of2(insn->vectorExtensionPrefix[1])((insn->vectorExtensionPrefix[1]) & 0x3)) {
922 case VEX_PREFIX_66:
923 attrMask |= ATTR_OPSIZE;
924 break;
925 case VEX_PREFIX_F3:
926 attrMask |= ATTR_XS;
927 break;
928 case VEX_PREFIX_F2:
929 attrMask |= ATTR_XD;
930 break;
931 }
932
933 if (lFromVEX2of2(insn->vectorExtensionPrefix[1])(((insn->vectorExtensionPrefix[1]) & 0x4) >> 2))
934 attrMask |= ATTR_VEXL;
935 } else if (insn->vectorExtensionType == TYPE_XOP) {
936 switch (ppFromXOP3of3(insn->vectorExtensionPrefix[2])((insn->vectorExtensionPrefix[2]) & 0x3)) {
937 case VEX_PREFIX_66:
938 attrMask |= ATTR_OPSIZE;
939 break;
940 case VEX_PREFIX_F3:
941 attrMask |= ATTR_XS;
942 break;
943 case VEX_PREFIX_F2:
944 attrMask |= ATTR_XD;
945 break;
946 }
947
948 if (lFromXOP3of3(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x4) >> 2))
949 attrMask |= ATTR_VEXL;
950 } else {
951 return -1;
952 }
953 } else {
954 if (insn->mode != MODE_16BIT && isPrefixAtLocation(insn, 0x66, insn->necessaryPrefixLocation))
955 attrMask |= ATTR_OPSIZE;
956 else if (isPrefixAtLocation(insn, 0x67, insn->necessaryPrefixLocation))
957 attrMask |= ATTR_ADSIZE;
958 else if (isPrefixAtLocation(insn, 0xf3, insn->necessaryPrefixLocation))
959 attrMask |= ATTR_XS;
960 else if (isPrefixAtLocation(insn, 0xf2, insn->necessaryPrefixLocation))
961 attrMask |= ATTR_XD;
962 }
963
964 if (insn->rexPrefix & 0x08)
965 attrMask |= ATTR_REXW;
966
967 /*
968 * JCXZ/JECXZ need special handling for 16-bit mode because the meaning
969 * of the AdSize prefix is inverted w.r.t. 32-bit mode.
970 */
971 if (insn->mode == MODE_16BIT && insn->opcodeType == ONEBYTE &&
972 insn->opcode == 0xE3)
973 attrMask ^= ATTR_ADSIZE;
974
975 /*
976 * In 64-bit mode all f64 superscripted opcodes ignore opcode size prefix
977 * CALL/JMP/JCC instructions need to ignore 0x66 and consume 4 bytes
978 */
979
980 if (insn->mode == MODE_64BIT &&
981 isPrefixAtLocation(insn, 0x66, insn->necessaryPrefixLocation)) {
982 switch (insn->opcode) {
983 case 0xE8:
984 case 0xE9:
985 // Take care of psubsb and other mmx instructions.
986 if (insn->opcodeType == ONEBYTE) {
987 attrMask ^= ATTR_OPSIZE;
988 insn->immediateSize = 4;
989 insn->displacementSize = 4;
990 }
991 break;
992 case 0x82:
993 case 0x83:
994 case 0x84:
995 case 0x85:
996 case 0x86:
997 case 0x87:
998 case 0x88:
999 case 0x89:
1000 case 0x8A:
1001 case 0x8B:
1002 case 0x8C:
1003 case 0x8D:
1004 case 0x8E:
1005 case 0x8F:
1006 // Take care of lea and three byte ops.
1007 if (insn->opcodeType == TWOBYTE) {
1008 attrMask ^= ATTR_OPSIZE;
1009 insn->immediateSize = 4;
1010 insn->displacementSize = 4;
1011 }
1012 break;
1013 }
1014 }
1015
1016 if (getIDWithAttrMask(&instructionID, insn, attrMask))
1017 return -1;
1018
1019 /* The following clauses compensate for limitations of the tables. */
1020
1021 if (insn->mode != MODE_64BIT &&
1022 insn->vectorExtensionType != TYPE_NO_VEX_XOP) {
1023 /*
1024 * The tables can't distinquish between cases where the W-bit is used to
1025 * select register size and cases where its a required part of the opcode.
1026 */
1027 if ((insn->vectorExtensionType == TYPE_EVEX &&
1028 wFromEVEX3of4(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x80) >> 7)) ||
1029 (insn->vectorExtensionType == TYPE_VEX_3B &&
1030 wFromVEX3of3(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x80) >> 7)) ||
1031 (insn->vectorExtensionType == TYPE_XOP &&
1032 wFromXOP3of3(insn->vectorExtensionPrefix[2])(((insn->vectorExtensionPrefix[2]) & 0x80) >> 7))) {
1033
1034 uint16_t instructionIDWithREXW;
1035 if (getIDWithAttrMask(&instructionIDWithREXW,
1036 insn, attrMask | ATTR_REXW)) {
1037 insn->instructionID = instructionID;
1038 insn->spec = specifierForUID(instructionID);
1039 return 0;
1040 }
1041
1042 auto SpecName = GetInstrName(instructionIDWithREXW, miiArg);
1043 // If not a 64-bit instruction. Switch the opcode.
1044 if (!is64Bit(SpecName.data())) {
1045 insn->instructionID = instructionIDWithREXW;
1046 insn->spec = specifierForUID(instructionIDWithREXW);
1047 return 0;
1048 }
1049 }
1050 }
1051
1052 /*
1053 * Absolute moves need special handling.
1054 * -For 16-bit mode because the meaning of the AdSize and OpSize prefixes are
1055 * inverted w.r.t.
1056 * -For 32-bit mode we need to ensure the ADSIZE prefix is observed in
1057 * any position.
1058 */
1059 if (insn->opcodeType == ONEBYTE && ((insn->opcode & 0xFC) == 0xA0)) {
1060 /* Make sure we observed the prefixes in any position. */
1061 if (insn->prefixPresent[0x67])
1062 attrMask |= ATTR_ADSIZE;
1063 if (insn->prefixPresent[0x66])
1064 attrMask |= ATTR_OPSIZE;
1065
1066 /* In 16-bit, invert the attributes. */
1067 if (insn->mode == MODE_16BIT)
1068 attrMask ^= ATTR_ADSIZE | ATTR_OPSIZE;
1069
1070 if (getIDWithAttrMask(&instructionID, insn, attrMask))
1071 return -1;
1072
1073 insn->instructionID = instructionID;
1074 insn->spec = specifierForUID(instructionID);
1075 return 0;
1076 }
1077
1078 if ((insn->mode == MODE_16BIT || insn->prefixPresent[0x66]) &&
1079 !(attrMask & ATTR_OPSIZE)) {
1080 /*
1081 * The instruction tables make no distinction between instructions that
1082 * allow OpSize anywhere (i.e., 16-bit operations) and that need it in a
1083 * particular spot (i.e., many MMX operations). In general we're
1084 * conservative, but in the specific case where OpSize is present but not
1085 * in the right place we check if there's a 16-bit operation.
1086 */
1087
1088 const struct InstructionSpecifier *spec;
1089 uint16_t instructionIDWithOpsize;
1090 llvm::StringRef specName, specWithOpSizeName;
1091
1092 spec = specifierForUID(instructionID);
1093
1094 if (getIDWithAttrMask(&instructionIDWithOpsize,
1095 insn,
1096 attrMask | ATTR_OPSIZE)) {
1097 /*
1098 * ModRM required with OpSize but not present; give up and return version
1099 * without OpSize set
1100 */
1101
1102 insn->instructionID = instructionID;
1103 insn->spec = spec;
1104 return 0;
1105 }
1106
1107 specName = GetInstrName(instructionID, miiArg);
1108 specWithOpSizeName = GetInstrName(instructionIDWithOpsize, miiArg);
1109
1110 if (is16BitEquivalent(specName.data(), specWithOpSizeName.data()) &&
1111 (insn->mode == MODE_16BIT) ^ insn->prefixPresent[0x66]) {
1112 insn->instructionID = instructionIDWithOpsize;
1113 insn->spec = specifierForUID(instructionIDWithOpsize);
1114 } else {
1115 insn->instructionID = instructionID;
1116 insn->spec = spec;
1117 }
1118 return 0;
1119 }
1120
1121 if (insn->opcodeType == ONEBYTE && insn->opcode == 0x90 &&
1122 insn->rexPrefix & 0x01) {
1123 /*
1124 * NOOP shouldn't decode as NOOP if REX.b is set. Instead
1125 * it should decode as XCHG %r8, %eax.
1126 */
1127
1128 const struct InstructionSpecifier *spec;
1129 uint16_t instructionIDWithNewOpcode;
1130 const struct InstructionSpecifier *specWithNewOpcode;
1131
1132 spec = specifierForUID(instructionID);
1133
1134 /* Borrow opcode from one of the other XCHGar opcodes */
1135 insn->opcode = 0x91;
1136
1137 if (getIDWithAttrMask(&instructionIDWithNewOpcode,
1138 insn,
1139 attrMask)) {
1140 insn->opcode = 0x90;
1141
1142 insn->instructionID = instructionID;
1143 insn->spec = spec;
1144 return 0;
1145 }
1146
1147 specWithNewOpcode = specifierForUID(instructionIDWithNewOpcode);
1148
1149 /* Change back */
1150 insn->opcode = 0x90;
1151
1152 insn->instructionID = instructionIDWithNewOpcode;
1153 insn->spec = specWithNewOpcode;
1154
1155 return 0;
1156 }
1157
1158 insn->instructionID = instructionID;
1159 insn->spec = specifierForUID(insn->instructionID);
1160
1161 return 0;
1162}
1163
1164/*
1165 * readSIB - Consumes the SIB byte to determine addressing information for an
1166 * instruction.
1167 *
1168 * @param insn - The instruction whose SIB byte is to be read.
1169 * @return - 0 if the SIB byte was successfully read; nonzero otherwise.
1170 */
1171static int readSIB(struct InternalInstruction* insn) {
1172 SIBIndex sibIndexBase = SIB_INDEX_NONE;
1173 SIBBase sibBaseBase = SIB_BASE_NONE;
1174 uint8_t index, base;
1175
1176 dbgprintf(insn, "readSIB()");
1177
1178 if (insn->consumedSIB)
1179 return 0;
1180
1181 insn->consumedSIB = true;
1182
1183 switch (insn->addressSize) {
1184 case 2:
1185 dbgprintf(insn, "SIB-based addressing doesn't work in 16-bit mode");
1186 return -1;
1187 case 4:
1188 sibIndexBase = SIB_INDEX_EAX;
1189 sibBaseBase = SIB_BASE_EAX;
1190 break;
1191 case 8:
1192 sibIndexBase = SIB_INDEX_RAX;
1193 sibBaseBase = SIB_BASE_RAX;
1194 break;
1195 }
1196
1197 if (consumeByte(insn, &insn->sib))
1198 return -1;
1199
1200 index = indexFromSIB(insn->sib)(((insn->sib) & 0x38) >> 3) | (xFromREX(insn->rexPrefix)(((insn->rexPrefix) & 0x2) >> 1) << 3);
1201
1202 // FIXME: The fifth bit (bit index 4) is only to be used for instructions
1203 // that understand VSIB indexing. ORing the bit in here is mildy dangerous
1204 // because performing math on an 'enum SIBIndex' can produce garbage.
1205 // Excluding the "none" value, it should cover 6 spaces of register names:
1206 // - 16 possibilities for 16-bit GPR starting at SIB_INDEX_BX_SI
1207 // - 16 possibilities for 32-bit GPR starting at SIB_INDEX_EAX
1208 // - 16 possibilities for 64-bit GPR starting at SIB_INDEX_RAX
1209 // - 32 possibilities for each of XMM, YMM, ZMM registers
1210 // When sibIndexBase gets assigned SIB_INDEX_RAX as it does in 64-bit mode,
1211 // summing in a fully decoded index between 0 and 31 can end up with a value
1212 // that looks like something in the low half of the XMM range.
1213 // translateRMMemory() tries to reverse the damage, with only partial success,
1214 // as evidenced by known bugs in "test/MC/Disassembler/X86/x86-64.txt"
1215 if (insn->vectorExtensionType == TYPE_EVEX)
1216 index |= v2FromEVEX4of4(insn->vectorExtensionPrefix[3])(((~insn->vectorExtensionPrefix[3]) & 0x8) >> 3) << 4;
1217
1218 if (index == 0x4) {
1219 insn->sibIndex = SIB_INDEX_NONE;
1220 } else {
1221 insn->sibIndex = (SIBIndex)(sibIndexBase + index);
1222 }
1223
1224 insn->sibScale = 1 << scaleFromSIB(insn->sib)(((insn->sib) & 0xc0) >> 6);
1225
1226 base = baseFromSIB(insn->sib)((insn->sib) & 0x7) | (bFromREX(insn->rexPrefix)((insn->rexPrefix) & 0x1) << 3);
1227
1228 switch (base) {
1229 case 0x5:
1230 case 0xd:
1231 switch (modFromModRM(insn->modRM)(((insn->modRM) & 0xc0) >> 6)) {
1232 case 0x0:
1233 insn->eaDisplacement = EA_DISP_32;
1234 insn->sibBase = SIB_BASE_NONE;
1235 break;
1236 case 0x1:
1237 insn->eaDisplacement = EA_DISP_8;
1238 insn->sibBase = (SIBBase)(sibBaseBase + base);
1239 break;
1240 case 0x2:
1241 insn->eaDisplacement = EA_DISP_32;
1242 insn->sibBase = (SIBBase)(sibBaseBase + base);
1243 break;
1244 case 0x3:
1245 debug("Cannot have Mod = 0b11 and a SIB byte")do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 1245, "Cannot have Mod = 0b11 and a SIB byte"); } while (0)
;
1246 return -1;
1247 }
1248 break;
1249 default:
1250 insn->sibBase = (SIBBase)(sibBaseBase + base);
1251 break;
1252 }
1253
1254 return 0;
1255}
1256
1257/*
1258 * readDisplacement - Consumes the displacement of an instruction.
1259 *
1260 * @param insn - The instruction whose displacement is to be read.
1261 * @return - 0 if the displacement byte was successfully read; nonzero
1262 * otherwise.
1263 */
1264static int readDisplacement(struct InternalInstruction* insn) {
1265 int8_t d8;
1266 int16_t d16;
1267 int32_t d32;
1268
1269 dbgprintf(insn, "readDisplacement()");
1270
1271 if (insn->consumedDisplacement)
1272 return 0;
1273
1274 insn->consumedDisplacement = true;
1275 insn->displacementOffset = insn->readerCursor - insn->startLocation;
1276
1277 switch (insn->eaDisplacement) {
1278 case EA_DISP_NONE:
1279 insn->consumedDisplacement = false;
1280 break;
1281 case EA_DISP_8:
1282 if (consumeInt8(insn, &d8))
1283 return -1;
1284 insn->displacement = d8;
1285 break;
1286 case EA_DISP_16:
1287 if (consumeInt16(insn, &d16))
1288 return -1;
1289 insn->displacement = d16;
1290 break;
1291 case EA_DISP_32:
1292 if (consumeInt32(insn, &d32))
1293 return -1;
1294 insn->displacement = d32;
1295 break;
1296 }
1297
1298 insn->consumedDisplacement = true;
1299 return 0;
1300}
1301
1302/*
1303 * readModRM - Consumes all addressing information (ModR/M byte, SIB byte, and
1304 * displacement) for an instruction and interprets it.
1305 *
1306 * @param insn - The instruction whose addressing information is to be read.
1307 * @return - 0 if the information was successfully read; nonzero otherwise.
1308 */
1309static int readModRM(struct InternalInstruction* insn) {
1310 uint8_t mod, rm, reg;
1311
1312 dbgprintf(insn, "readModRM()");
1313
1314 if (insn->consumedModRM)
1315 return 0;
1316
1317 if (consumeByte(insn, &insn->modRM))
1318 return -1;
1319 insn->consumedModRM = true;
1320
1321 mod = modFromModRM(insn->modRM)(((insn->modRM) & 0xc0) >> 6);
1322 rm = rmFromModRM(insn->modRM)((insn->modRM) & 0x7);
1323 reg = regFromModRM(insn->modRM)(((insn->modRM) & 0x38) >> 3);
1324
1325 /*
1326 * This goes by insn->registerSize to pick the correct register, which messes
1327 * up if we're using (say) XMM or 8-bit register operands. That gets fixed in
1328 * fixupReg().
1329 */
1330 switch (insn->registerSize) {
1331 case 2:
1332 insn->regBase = MODRM_REG_AX;
1333 insn->eaRegBase = EA_REG_AX;
1334 break;
1335 case 4:
1336 insn->regBase = MODRM_REG_EAX;
1337 insn->eaRegBase = EA_REG_EAX;
1338 break;
1339 case 8:
1340 insn->regBase = MODRM_REG_RAX;
1341 insn->eaRegBase = EA_REG_RAX;
1342 break;
1343 }
1344
1345 reg |= rFromREX(insn->rexPrefix)(((insn->rexPrefix) & 0x4) >> 2) << 3;
1346 rm |= bFromREX(insn->rexPrefix)((insn->rexPrefix) & 0x1) << 3;
1347 if (insn->vectorExtensionType == TYPE_EVEX) {
1348 reg |= r2FromEVEX2of4(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x10) >>
4)
<< 4;
1349 rm |= xFromEVEX2of4(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x40) >>
6)
<< 4;
1350 }
1351
1352 insn->reg = (Reg)(insn->regBase + reg);
1353
1354 switch (insn->addressSize) {
1355 case 2:
1356 insn->eaBaseBase = EA_BASE_BX_SI;
1357
1358 switch (mod) {
1359 case 0x0:
1360 if (rm == 0x6) {
1361 insn->eaBase = EA_BASE_NONE;
1362 insn->eaDisplacement = EA_DISP_16;
1363 if (readDisplacement(insn))
1364 return -1;
1365 } else {
1366 insn->eaBase = (EABase)(insn->eaBaseBase + rm);
1367 insn->eaDisplacement = EA_DISP_NONE;
1368 }
1369 break;
1370 case 0x1:
1371 insn->eaBase = (EABase)(insn->eaBaseBase + rm);
1372 insn->eaDisplacement = EA_DISP_8;
1373 insn->displacementSize = 1;
1374 if (readDisplacement(insn))
1375 return -1;
1376 break;
1377 case 0x2:
1378 insn->eaBase = (EABase)(insn->eaBaseBase + rm);
1379 insn->eaDisplacement = EA_DISP_16;
1380 if (readDisplacement(insn))
1381 return -1;
1382 break;
1383 case 0x3:
1384 insn->eaBase = (EABase)(insn->eaRegBase + rm);
1385 if (readDisplacement(insn))
1386 return -1;
1387 break;
1388 }
1389 break;
1390 case 4:
1391 case 8:
1392 insn->eaBaseBase = (insn->addressSize == 4 ? EA_BASE_EAX : EA_BASE_RAX);
1393
1394 switch (mod) {
1395 case 0x0:
1396 insn->eaDisplacement = EA_DISP_NONE; /* readSIB may override this */
1397 // In determining whether RIP-relative mode is used (rm=5),
1398 // or whether a SIB byte is present (rm=4),
1399 // the extension bits (REX.b and EVEX.x) are ignored.
1400 switch (rm & 7) {
1401 case 0x4: // SIB byte is present
1402 insn->eaBase = (insn->addressSize == 4 ?
1403 EA_BASE_sib : EA_BASE_sib64);
1404 if (readSIB(insn) || readDisplacement(insn))
1405 return -1;
1406 break;
1407 case 0x5: // RIP-relative
1408 insn->eaBase = EA_BASE_NONE;
1409 insn->eaDisplacement = EA_DISP_32;
1410 if (readDisplacement(insn))
1411 return -1;
1412 break;
1413 default:
1414 insn->eaBase = (EABase)(insn->eaBaseBase + rm);
1415 break;
1416 }
1417 break;
1418 case 0x1:
1419 insn->displacementSize = 1;
1420 /* FALLTHROUGH */
1421 case 0x2:
1422 insn->eaDisplacement = (mod == 0x1 ? EA_DISP_8 : EA_DISP_32);
1423 switch (rm & 7) {
1424 case 0x4: // SIB byte is present
1425 insn->eaBase = EA_BASE_sib;
1426 if (readSIB(insn) || readDisplacement(insn))
1427 return -1;
1428 break;
1429 default:
1430 insn->eaBase = (EABase)(insn->eaBaseBase + rm);
1431 if (readDisplacement(insn))
1432 return -1;
1433 break;
1434 }
1435 break;
1436 case 0x3:
1437 insn->eaDisplacement = EA_DISP_NONE;
1438 insn->eaBase = (EABase)(insn->eaRegBase + rm);
1439 break;
1440 }
1441 break;
1442 } /* switch (insn->addressSize) */
1443
1444 return 0;
1445}
1446
1447#define GENERIC_FIXUP_FUNC(name, base, prefix)static uint16_t name(struct InternalInstruction *insn, OperandType
type, uint8_t index, uint8_t *valid) { *valid = 1; switch (type
) { default: do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 1447, "Unhandled register type"); } while (0); *valid = 0; return
0; case TYPE_Rv: return base + index; case TYPE_R8: if (insn
->rexPrefix && index >= 4 && index <=
7) { return prefix_SPL + (index - 4); } else { return prefix_AL
+ index; } case TYPE_R16: return prefix_AX + index; case TYPE_R32
: return prefix_EAX + index; case TYPE_R64: return prefix_RAX
+ index; case TYPE_ZMM: return prefix_ZMM0 + index; case TYPE_YMM
: return prefix_YMM0 + index; case TYPE_XMM: return prefix_XMM0
+ index; case TYPE_VK: if (index > 7) *valid = 0; return prefix_K0
+ index; case TYPE_MM64: return prefix_MM0 + (index & 0x7
); case TYPE_SEGMENTREG: if (index > 5) *valid = 0; return
prefix_ES + index; case TYPE_DEBUGREG: return prefix_DR0 + index
; case TYPE_CONTROLREG: return prefix_CR0 + index; case TYPE_BNDR
: if (index > 3) *valid = 0; return prefix_BND0 + index; }
}
\
1448 static uint16_t name(struct InternalInstruction *insn, \
1449 OperandType type, \
1450 uint8_t index, \
1451 uint8_t *valid) { \
1452 *valid = 1; \
1453 switch (type) { \
1454 default: \
1455 debug("Unhandled register type")do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 1455, "Unhandled register type"); } while (0)
; \
1456 *valid = 0; \
1457 return 0; \
1458 case TYPE_Rv: \
1459 return base + index; \
1460 case TYPE_R8: \
1461 if (insn->rexPrefix && \
1462 index >= 4 && index <= 7) { \
1463 return prefix##_SPL + (index - 4); \
1464 } else { \
1465 return prefix##_AL + index; \
1466 } \
1467 case TYPE_R16: \
1468 return prefix##_AX + index; \
1469 case TYPE_R32: \
1470 return prefix##_EAX + index; \
1471 case TYPE_R64: \
1472 return prefix##_RAX + index; \
1473 case TYPE_ZMM: \
1474 return prefix##_ZMM0 + index; \
1475 case TYPE_YMM: \
1476 return prefix##_YMM0 + index; \
1477 case TYPE_XMM: \
1478 return prefix##_XMM0 + index; \
1479 case TYPE_VK: \
1480 if (index > 7) \
1481 *valid = 0; \
1482 return prefix##_K0 + index; \
1483 case TYPE_MM64: \
1484 return prefix##_MM0 + (index & 0x7); \
1485 case TYPE_SEGMENTREG: \
1486 if (index > 5) \
1487 *valid = 0; \
1488 return prefix##_ES + index; \
1489 case TYPE_DEBUGREG: \
1490 return prefix##_DR0 + index; \
1491 case TYPE_CONTROLREG: \
1492 return prefix##_CR0 + index; \
1493 case TYPE_BNDR: \
1494 if (index > 3) \
1495 *valid = 0; \
1496 return prefix##_BND0 + index; \
1497 } \
1498 }
1499
1500/*
1501 * fixup*Value - Consults an operand type to determine the meaning of the
1502 * reg or R/M field. If the operand is an XMM operand, for example, an
1503 * operand would be XMM0 instead of AX, which readModRM() would otherwise
1504 * misinterpret it as.
1505 *
1506 * @param insn - The instruction containing the operand.
1507 * @param type - The operand type.
1508 * @param index - The existing value of the field as reported by readModRM().
1509 * @param valid - The address of a uint8_t. The target is set to 1 if the
1510 * field is valid for the register class; 0 if not.
1511 * @return - The proper value.
1512 */
1513GENERIC_FIXUP_FUNC(fixupRegValue, insn->regBase, MODRM_REG)static uint16_t fixupRegValue(struct InternalInstruction *insn
, OperandType type, uint8_t index, uint8_t *valid) { *valid =
1; switch (type) { default: do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 1513, "Unhandled register type"); } while (0); *valid = 0; return
0; case TYPE_Rv: return insn->regBase + index; case TYPE_R8
: if (insn->rexPrefix && index >= 4 && index
<= 7) { return MODRM_REG_SPL + (index - 4); } else { return
MODRM_REG_AL + index; } case TYPE_R16: return MODRM_REG_AX +
index; case TYPE_R32: return MODRM_REG_EAX + index; case TYPE_R64
: return MODRM_REG_RAX + index; case TYPE_ZMM: return MODRM_REG_ZMM0
+ index; case TYPE_YMM: return MODRM_REG_YMM0 + index; case TYPE_XMM
: return MODRM_REG_XMM0 + index; case TYPE_VK: if (index >
7) *valid = 0; return MODRM_REG_K0 + index; case TYPE_MM64: return
MODRM_REG_MM0 + (index & 0x7); case TYPE_SEGMENTREG: if (
index > 5) *valid = 0; return MODRM_REG_ES + index; case TYPE_DEBUGREG
: return MODRM_REG_DR0 + index; case TYPE_CONTROLREG: return MODRM_REG_CR0
+ index; case TYPE_BNDR: if (index > 3) *valid = 0; return
MODRM_REG_BND0 + index; } }
1514GENERIC_FIXUP_FUNC(fixupRMValue, insn->eaRegBase, EA_REG)static uint16_t fixupRMValue(struct InternalInstruction *insn
, OperandType type, uint8_t index, uint8_t *valid) { *valid =
1; switch (type) { default: do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 1514, "Unhandled register type"); } while (0); *valid = 0; return
0; case TYPE_Rv: return insn->eaRegBase + index; case TYPE_R8
: if (insn->rexPrefix && index >= 4 && index
<= 7) { return EA_REG_SPL + (index - 4); } else { return EA_REG_AL
+ index; } case TYPE_R16: return EA_REG_AX + index; case TYPE_R32
: return EA_REG_EAX + index; case TYPE_R64: return EA_REG_RAX
+ index; case TYPE_ZMM: return EA_REG_ZMM0 + index; case TYPE_YMM
: return EA_REG_YMM0 + index; case TYPE_XMM: return EA_REG_XMM0
+ index; case TYPE_VK: if (index > 7) *valid = 0; return EA_REG_K0
+ index; case TYPE_MM64: return EA_REG_MM0 + (index & 0x7
); case TYPE_SEGMENTREG: if (index > 5) *valid = 0; return
EA_REG_ES + index; case TYPE_DEBUGREG: return EA_REG_DR0 + index
; case TYPE_CONTROLREG: return EA_REG_CR0 + index; case TYPE_BNDR
: if (index > 3) *valid = 0; return EA_REG_BND0 + index; }
}
1515
1516/*
1517 * fixupReg - Consults an operand specifier to determine which of the
1518 * fixup*Value functions to use in correcting readModRM()'ss interpretation.
1519 *
1520 * @param insn - See fixup*Value().
1521 * @param op - The operand specifier.
1522 * @return - 0 if fixup was successful; -1 if the register returned was
1523 * invalid for its class.
1524 */
1525static int fixupReg(struct InternalInstruction *insn,
1526 const struct OperandSpecifier *op) {
1527 uint8_t valid;
1528
1529 dbgprintf(insn, "fixupReg()");
1530
1531 switch ((OperandEncoding)op->encoding) {
1532 default:
1533 debug("Expected a REG or R/M encoding in fixupReg")do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 1533, "Expected a REG or R/M encoding in fixupReg"); } while
(0)
;
1534 return -1;
1535 case ENCODING_VVVV:
1536 insn->vvvv = (Reg)fixupRegValue(insn,
1537 (OperandType)op->type,
1538 insn->vvvv,
1539 &valid);
1540 if (!valid)
1541 return -1;
1542 break;
1543 case ENCODING_REG:
1544 insn->reg = (Reg)fixupRegValue(insn,
1545 (OperandType)op->type,
1546 insn->reg - insn->regBase,
1547 &valid);
1548 if (!valid)
1549 return -1;
1550 break;
1551 CASE_ENCODING_RMcase ENCODING_RM: case ENCODING_RM_CD2: case ENCODING_RM_CD4:
case ENCODING_RM_CD8: case ENCODING_RM_CD16: case ENCODING_RM_CD32
: case ENCODING_RM_CD64
:
1552 CASE_ENCODING_VSIBcase ENCODING_VSIB: case ENCODING_VSIB_CD2: case ENCODING_VSIB_CD4
: case ENCODING_VSIB_CD8: case ENCODING_VSIB_CD16: case ENCODING_VSIB_CD32
: case ENCODING_VSIB_CD64
:
1553 if (insn->eaBase >= insn->eaRegBase) {
1554 insn->eaBase = (EABase)fixupRMValue(insn,
1555 (OperandType)op->type,
1556 insn->eaBase - insn->eaRegBase,
1557 &valid);
1558 if (!valid)
1559 return -1;
1560 }
1561 break;
1562 }
1563
1564 return 0;
1565}
1566
1567/*
1568 * readOpcodeRegister - Reads an operand from the opcode field of an
1569 * instruction and interprets it appropriately given the operand width.
1570 * Handles AddRegFrm instructions.
1571 *
1572 * @param insn - the instruction whose opcode field is to be read.
1573 * @param size - The width (in bytes) of the register being specified.
1574 * 1 means AL and friends, 2 means AX, 4 means EAX, and 8 means
1575 * RAX.
1576 * @return - 0 on success; nonzero otherwise.
1577 */
1578static int readOpcodeRegister(struct InternalInstruction* insn, uint8_t size) {
1579 dbgprintf(insn, "readOpcodeRegister()");
1580
1581 if (size == 0)
1582 size = insn->registerSize;
1583
1584 switch (size) {
1585 case 1:
1586 insn->opcodeRegister = (Reg)(MODRM_REG_AL + ((bFromREX(insn->rexPrefix)((insn->rexPrefix) & 0x1) << 3)
1587 | (insn->opcode & 7)));
1588 if (insn->rexPrefix &&
1589 insn->opcodeRegister >= MODRM_REG_AL + 0x4 &&
1590 insn->opcodeRegister < MODRM_REG_AL + 0x8) {
1591 insn->opcodeRegister = (Reg)(MODRM_REG_SPL
1592 + (insn->opcodeRegister - MODRM_REG_AL - 4));
1593 }
1594
1595 break;
1596 case 2:
1597 insn->opcodeRegister = (Reg)(MODRM_REG_AX
1598 + ((bFromREX(insn->rexPrefix)((insn->rexPrefix) & 0x1) << 3)
1599 | (insn->opcode & 7)));
1600 break;
1601 case 4:
1602 insn->opcodeRegister = (Reg)(MODRM_REG_EAX
1603 + ((bFromREX(insn->rexPrefix)((insn->rexPrefix) & 0x1) << 3)
1604 | (insn->opcode & 7)));
1605 break;
1606 case 8:
1607 insn->opcodeRegister = (Reg)(MODRM_REG_RAX
1608 + ((bFromREX(insn->rexPrefix)((insn->rexPrefix) & 0x1) << 3)
1609 | (insn->opcode & 7)));
1610 break;
1611 }
1612
1613 return 0;
1614}
1615
1616/*
1617 * readImmediate - Consumes an immediate operand from an instruction, given the
1618 * desired operand size.
1619 *
1620 * @param insn - The instruction whose operand is to be read.
1621 * @param size - The width (in bytes) of the operand.
1622 * @return - 0 if the immediate was successfully consumed; nonzero
1623 * otherwise.
1624 */
1625static int readImmediate(struct InternalInstruction* insn, uint8_t size) {
1626 uint8_t imm8;
1627 uint16_t imm16;
1628 uint32_t imm32;
1629 uint64_t imm64;
1630
1631 dbgprintf(insn, "readImmediate()");
1632
1633 if (insn->numImmediatesConsumed == 2) {
1634 debug("Already consumed two immediates")do { Debug("/tmp/buildd/llvm-toolchain-snapshot-5.0~svn301389/lib/Target/X86/Disassembler/X86DisassemblerDecoder.cpp"
, 1634, "Already consumed two immediates"); } while (0)
;
1635 return -1;
1636 }
1637
1638 if (size == 0)
1639 size = insn->immediateSize;
1640 else
1641 insn->immediateSize = size;
1642 insn->immediateOffset = insn->readerCursor - insn->startLocation;
1643
1644 switch (size) {
1645 case 1:
1646 if (consumeByte(insn, &imm8))
1647 return -1;
1648 insn->immediates[insn->numImmediatesConsumed] = imm8;
1649 break;
1650 case 2:
1651 if (consumeUInt16(insn, &imm16))
1652 return -1;
1653 insn->immediates[insn->numImmediatesConsumed] = imm16;
1654 break;
1655 case 4:
1656 if (consumeUInt32(insn, &imm32))
1657 return -1;
1658 insn->immediates[insn->numImmediatesConsumed] = imm32;
1659 break;
1660 case 8:
1661 if (consumeUInt64(insn, &imm64))
1662 return -1;
1663 insn->immediates[insn->numImmediatesConsumed] = imm64;
1664 break;
1665 }
1666
1667 insn->numImmediatesConsumed++;
1668
1669 return 0;
1670}
1671
1672/*
1673 * readVVVV - Consumes vvvv from an instruction if it has a VEX prefix.
1674 *
1675 * @param insn - The instruction whose operand is to be read.
1676 * @return - 0 if the vvvv was successfully consumed; nonzero
1677 * otherwise.
1678 */
1679static int readVVVV(struct InternalInstruction* insn) {
1680 dbgprintf(insn, "readVVVV()");
1681
1682 int vvvv;
1683 if (insn->vectorExtensionType == TYPE_EVEX)
1684 vvvv = (v2FromEVEX4of4(insn->vectorExtensionPrefix[3])(((~insn->vectorExtensionPrefix[3]) & 0x8) >> 3) << 4 |
1685 vvvvFromEVEX3of4(insn->vectorExtensionPrefix[2])(((~(insn->vectorExtensionPrefix[2])) & 0x78) >>
3)
);
1686 else if (insn->vectorExtensionType == TYPE_VEX_3B)
1687 vvvv = vvvvFromVEX3of3(insn->vectorExtensionPrefix[2])(((~(insn->vectorExtensionPrefix[2])) & 0x78) >>
3)
;
1688 else if (insn->vectorExtensionType == TYPE_VEX_2B)
1689 vvvv = vvvvFromVEX2of2(insn->vectorExtensionPrefix[1])(((~(insn->vectorExtensionPrefix[1])) & 0x78) >>
3)
;
1690 else if (insn->vectorExtensionType == TYPE_XOP)
1691 vvvv = vvvvFromXOP3of3(insn->vectorExtensionPrefix[2])(((~(insn->vectorExtensionPrefix[2])) & 0x78) >>
3)
;
1692 else
1693 return -1;
1694
1695 if (insn->mode != MODE_64BIT)
1696 vvvv &= 0x7;
1697
1698 insn->vvvv = static_cast<Reg>(vvvv);
1699 return 0;
1700}
1701
1702/*
1703 * readMaskRegister - Reads an mask register from the opcode field of an
1704 * instruction.
1705 *
1706 * @param insn - The instruction whose opcode field is to be read.
1707 * @return - 0 on success; nonzero otherwise.
1708 */
1709static int readMaskRegister(struct InternalInstruction* insn) {
1710 dbgprintf(insn, "readMaskRegister()");
1711
1712 if (insn->vectorExtensionType != TYPE_EVEX)
1713 return -1;
1714
1715 insn->writemask =
1716 static_cast<Reg>(aaaFromEVEX4of4(insn->vectorExtensionPrefix[3])((insn->vectorExtensionPrefix[3]) & 0x7));
1717 return 0;
1718}
1719
1720/*
1721 * readOperands - Consults the specifier for an instruction and consumes all
1722 * operands for that instruction, interpreting them as it goes.
1723 *
1724 * @param insn - The instruction whose operands are to be read and interpreted.
1725 * @return - 0 if all operands could be read; nonzero otherwise.
1726 */
1727static int readOperands(struct InternalInstruction* insn) {
1728 int hasVVVV, needVVVV;
1729 int sawRegImm = 0;
1730
1731 dbgprintf(insn, "readOperands()");
1732
1733 /* If non-zero vvvv specified, need to make sure one of the operands
1734 uses it. */
1735 hasVVVV = !readVVVV(insn);
1736 needVVVV = hasVVVV && (insn->vvvv != 0);
1737
1738 for (const auto &Op : x86OperandSets[insn->spec->operands]) {
1739 switch (Op.encoding) {
1
Control jumps to 'case ENCODING_DUP:' at line 1833
3
Control jumps to 'case ENCODING_DUP:' at line 1833
5
Control jumps to 'case ENCODING_DUP:' at line 1833
7
Control jumps to 'case ENCODING_VSIB_CD64:' at line 1744
1740 case ENCODING_NONE:
1741 case ENCODING_SI:
1742 case ENCODING_DI:
1743 break;
1744 CASE_ENCODING_VSIBcase ENCODING_VSIB: case ENCODING_VSIB_CD2: case ENCODING_VSIB_CD4
: case ENCODING_VSIB_CD8: case ENCODING_VSIB_CD16: case ENCODING_VSIB_CD32
: case ENCODING_VSIB_CD64
:
1745 // VSIB can use the V2 bit so check only the other bits.
1746 if (needVVVV)
8
Taking false branch
1747 needVVVV = hasVVVV & ((insn->vvvv & 0xf) != 0);
1748 if (readModRM(insn))
9
Assuming the condition is false
10
Taking false branch
1749 return -1;
1750 if (fixupReg(insn, &Op))
11
Taking false branch
1751 return -1;
1752 // Apply the AVX512 compressed displacement scaling factor.
1753 if (Op.encoding != ENCODING_REG && insn->eaDisplacement == EA_DISP_8)
12
Assuming the condition is true
13
Taking true branch
1754 insn->displacement *= 1 << (Op.encoding - ENCODING_VSIB);
14
The result of the '<<' expression is undefined
1755 break;
1756 case ENCODING_REG:
1757 CASE_ENCODING_RMcase ENCODING_RM: case ENCODING_RM_CD2: case ENCODING_RM_CD4:
case ENCODING_RM_CD8: case ENCODING_RM_CD16: case ENCODING_RM_CD32
: case ENCODING_RM_CD64
:
1758 if (readModRM(insn))
1759 return -1;
1760 if (fixupReg(insn, &Op))
1761 return -1;
1762 // Apply the AVX512 compressed displacement scaling factor.
1763 if (Op.encoding != ENCODING_REG && insn->eaDisplacement == EA_DISP_8)
1764 insn->displacement *= 1 << (Op.encoding - ENCODING_RM);
1765 break;
1766 case ENCODING_IB:
1767 if (sawRegImm) {
1768 /* Saw a register immediate so don't read again and instead split the
1769 previous immediate. FIXME: This is a hack. */
1770 insn->immediates[insn->numImmediatesConsumed] =
1771 insn->immediates[insn->numImmediatesConsumed - 1] & 0xf;
1772 ++insn->numImmediatesConsumed;
1773 break;
1774 }
1775 if (readImmediate(insn, 1))
1776 return -1;
1777 if (Op.type == TYPE_XMM || Op.type == TYPE_YMM)
1778 sawRegImm = 1;
1779 break;
1780 case ENCODING_IW:
1781 if (readImmediate(insn, 2))
1782 return -1;
1783 break;
1784 case ENCODING_ID:
1785 if (readImmediate(insn, 4))
1786 return -1;
1787 break;
1788 case ENCODING_IO:
1789 if (readImmediate(insn, 8))
1790 return -1;
1791 break;
1792 case ENCODING_Iv:
1793 if (readImmediate(insn, insn->immediateSize))
1794 return -1;
1795 break;
1796 case ENCODING_Ia:
1797 if (readImmediate(insn, insn->addressSize))
1798 return -1;
1799 break;
1800 case ENCODING_RB:
1801 if (readOpcodeRegister(insn, 1))
1802 return -1;
1803 break;
1804 case ENCODING_RW:
1805 if (readOpcodeRegister(insn, 2))
1806 return -1;
1807 break;
1808 case ENCODING_RD:
1809 if (readOpcodeRegister(insn, 4))
1810 return -1;
1811 break;
1812 case ENCODING_RO:
1813 if (readOpcodeRegister(insn, 8))
1814 return -1;
1815 break;
1816 case ENCODING_Rv:
1817 if (readOpcodeRegister(insn, 0))
1818 return -1;
1819 break;
1820 case ENCODING_FP:
1821 break;
1822 case ENCODING_VVVV:
1823 needVVVV = 0; /* Mark that we have found a VVVV operand. */
1824 if (!hasVVVV)
1825 return -1;
1826 if (fixupReg(insn, &Op))
1827 return -1;
1828 break;
1829 case ENCODING_WRITEMASK:
1830 if (readMaskRegister(insn))
1831 return -1;
1832 break;
1833 case ENCODING_DUP:
1834 break;
2
Execution continues on line 1738
4
Execution continues on line 1738
6
Execution continues on line 1738
1835 default:
1836 dbgprintf(insn, "Encountered an operand with an unknown encoding.");
1837 return -1;
1838 }
1839 }
1840
1841 /* If we didn't find ENCODING_VVVV operand, but non-zero vvvv present, fail */
1842 if (needVVVV) return -1;
1843
1844 return 0;
1845}
1846
1847/*
1848 * decodeInstruction - Reads and interprets a full instruction provided by the
1849 * user.
1850 *
1851 * @param insn - A pointer to the instruction to be populated. Must be
1852 * pre-allocated.
1853 * @param reader - The function to be used to read the instruction's bytes.
1854 * @param readerArg - A generic argument to be passed to the reader to store
1855 * any internal state.
1856 * @param logger - If non-NULL, the function to be used to write log messages
1857 * and warnings.
1858 * @param loggerArg - A generic argument to be passed to the logger to store
1859 * any internal state.
1860 * @param startLoc - The address (in the reader's address space) of the first
1861 * byte in the instruction.
1862 * @param mode - The mode (real mode, IA-32e, or IA-32e in 64-bit mode) to
1863 * decode the instruction in.
1864 * @return - 0 if the instruction's memory could be read; nonzero if
1865 * not.
1866 */
1867int llvm::X86Disassembler::decodeInstruction(
1868 struct InternalInstruction *insn, byteReader_t reader,
1869 const void *readerArg, dlog_t logger, void *loggerArg, const void *miiArg,
1870 uint64_t startLoc, DisassemblerMode mode) {
1871 memset(insn, 0, sizeof(struct InternalInstruction));
1872
1873 insn->reader = reader;
1874 insn->readerArg = readerArg;
1875 insn->dlog = logger;
1876 insn->dlogArg = loggerArg;
1877 insn->startLocation = startLoc;
1878 insn->readerCursor = startLoc;
1879 insn->mode = mode;
1880 insn->numImmediatesConsumed = 0;
1881
1882 if (readPrefixes(insn) ||
1883 readOpcode(insn) ||
1884 getID(insn, miiArg) ||
1885 insn->instructionID == 0 ||
1886 readOperands(insn))
1887 return -1;
1888
1889 insn->operands = x86OperandSets[insn->spec->operands];
1890
1891 insn->length = insn->readerCursor - insn->startLocation;
1892
1893 dbgprintf(insn, "Read from 0x%llx to 0x%llx: length %zu",
1894 startLoc, insn->readerCursor, insn->length);
1895
1896 if (insn->length > 15)
1897 dbgprintf(insn, "Instruction exceeds 15-byte limit");
1898
1899 return 0;
1900}