blob: 903e36cfe6ce1d397966a75ce996106b0840d2f6 [file] [log] [blame]
Jia Liub22310f2012-02-18 12:03:15 +00001//===-- X86Disassembler.cpp - Disassembler for x86 and x86_64 -------------===//
Daniel Dunbar900f2ce2009-11-25 06:53:08 +00002//
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//===----------------------------------------------------------------------===//
Sean Callanan04cc3072009-12-19 02:59:52 +00009//
10// This file is part of the X86 Disassembler.
11// It contains code to translate the data produced by the decoder into
12// MCInsts.
13// Documentation for the disassembler can be found in X86Disassembler.h.
14//
15//===----------------------------------------------------------------------===//
16
17#include "X86Disassembler.h"
18#include "X86DisassemblerDecoder.h"
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +000019#include "llvm/MC/MCContext.h"
Sean Callanan04cc3072009-12-19 02:59:52 +000020#include "llvm/MC/MCDisassembler.h"
Chandler Carruthed0881b2012-12-03 16:50:05 +000021#include "llvm/MC/MCExpr.h"
Sean Callanan04cc3072009-12-19 02:59:52 +000022#include "llvm/MC/MCInst.h"
Benjamin Kramer478e8de2012-02-11 14:50:54 +000023#include "llvm/MC/MCInstrInfo.h"
James Molloy4c493e82011-09-07 17:24:38 +000024#include "llvm/MC/MCSubtargetInfo.h"
Sean Callanan010b3732010-04-02 21:23:51 +000025#include "llvm/Support/Debug.h"
Sean Callanan04cc3072009-12-19 02:59:52 +000026#include "llvm/Support/MemoryObject.h"
Evan Cheng2bb40352011-08-24 18:08:43 +000027#include "llvm/Support/TargetRegistry.h"
Sean Callanan04cc3072009-12-19 02:59:52 +000028#include "llvm/Support/raw_ostream.h"
Sean Callanan5c8f4cd2009-12-22 01:11:26 +000029
Evan Chengd9997ac2011-06-27 18:32:37 +000030#define GET_REGINFO_ENUM
31#include "X86GenRegisterInfo.inc"
Kevin Enderby5b03f722011-09-02 20:01:23 +000032#define GET_INSTRINFO_ENUM
33#include "X86GenInstrInfo.inc"
Sean Callanan5c8f4cd2009-12-22 01:11:26 +000034
Daniel Dunbar900f2ce2009-11-25 06:53:08 +000035using namespace llvm;
Sean Callanan04cc3072009-12-19 02:59:52 +000036using namespace llvm::X86Disassembler;
37
Sean Callanan010b3732010-04-02 21:23:51 +000038void x86DisassemblerDebug(const char *file,
39 unsigned line,
40 const char *s) {
41 dbgs() << file << ":" << line << ": " << s;
42}
43
Roman Divacky67923802012-09-05 21:17:34 +000044const char *x86DisassemblerGetInstrName(unsigned Opcode, const void *mii) {
Benjamin Kramer478e8de2012-02-11 14:50:54 +000045 const MCInstrInfo *MII = static_cast<const MCInstrInfo *>(mii);
46 return MII->getName(Opcode);
47}
48
Sean Callanan010b3732010-04-02 21:23:51 +000049#define debug(s) DEBUG(x86DisassemblerDebug(__FILE__, __LINE__, s));
50
Sean Callanan04cc3072009-12-19 02:59:52 +000051namespace llvm {
52
53// Fill-ins to make the compiler happy. These constants are never actually
54// assigned; they are just filler to make an automatically-generated switch
55// statement work.
56namespace X86 {
57 enum {
58 BX_SI = 500,
59 BX_DI = 501,
60 BP_SI = 502,
61 BP_DI = 503,
62 sib = 504,
63 sib64 = 505
64 };
65}
66
Sean Callanan5c8f4cd2009-12-22 01:11:26 +000067extern Target TheX86_32Target, TheX86_64Target;
68
Sean Callanan04cc3072009-12-19 02:59:52 +000069}
70
Sean Callanan010b3732010-04-02 21:23:51 +000071static bool translateInstruction(MCInst &target,
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +000072 InternalInstruction &source,
73 const MCDisassembler *Dis);
Sean Callanan04cc3072009-12-19 02:59:52 +000074
Benjamin Kramer478e8de2012-02-11 14:50:54 +000075X86GenericDisassembler::X86GenericDisassembler(const MCSubtargetInfo &STI,
76 DisassemblerMode mode,
77 const MCInstrInfo *MII)
78 : MCDisassembler(STI), MII(MII), fMode(mode) {}
Sean Callanan04cc3072009-12-19 02:59:52 +000079
80X86GenericDisassembler::~X86GenericDisassembler() {
Benjamin Kramer478e8de2012-02-11 14:50:54 +000081 delete MII;
Sean Callanan04cc3072009-12-19 02:59:52 +000082}
83
84/// regionReader - a callback function that wraps the readByte method from
85/// MemoryObject.
86///
87/// @param arg - The generic callback parameter. In this case, this should
88/// be a pointer to a MemoryObject.
89/// @param byte - A pointer to the byte to be read.
90/// @param address - The address to be read.
Roman Divacky67923802012-09-05 21:17:34 +000091static int regionReader(const void* arg, uint8_t* byte, uint64_t address) {
92 const MemoryObject* region = static_cast<const MemoryObject*>(arg);
Sean Callanan04cc3072009-12-19 02:59:52 +000093 return region->readByte(address, byte);
94}
95
96/// logger - a callback function that wraps the operator<< method from
97/// raw_ostream.
98///
99/// @param arg - The generic callback parameter. This should be a pointe
100/// to a raw_ostream.
101/// @param log - A string to be logged. logger() adds a newline.
102static void logger(void* arg, const char* log) {
103 if (!arg)
104 return;
105
106 raw_ostream &vStream = *(static_cast<raw_ostream*>(arg));
107 vStream << log << "\n";
108}
109
110//
111// Public interface for the disassembler
112//
113
Owen Andersona4043c42011-08-17 17:44:15 +0000114MCDisassembler::DecodeStatus
115X86GenericDisassembler::getInstruction(MCInst &instr,
116 uint64_t &size,
Derek Schuff56b662c2012-02-29 01:09:06 +0000117 const MemoryObject &region,
Owen Andersona4043c42011-08-17 17:44:15 +0000118 uint64_t address,
Owen Andersona0c3b972011-09-15 23:38:46 +0000119 raw_ostream &vStream,
120 raw_ostream &cStream) const {
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000121 CommentStream = &cStream;
122
Sean Callanan04cc3072009-12-19 02:59:52 +0000123 InternalInstruction internalInstr;
Benjamin Kramere5e189f2011-09-21 21:47:35 +0000124
125 dlog_t loggerFn = logger;
126 if (&vStream == &nulls())
127 loggerFn = 0; // Disable logging completely if it's going to nulls().
Sean Callanan04cc3072009-12-19 02:59:52 +0000128
129 int ret = decodeInstruction(&internalInstr,
130 regionReader,
Roman Divacky67923802012-09-05 21:17:34 +0000131 (const void*)&region,
Benjamin Kramere5e189f2011-09-21 21:47:35 +0000132 loggerFn,
Sean Callanan04cc3072009-12-19 02:59:52 +0000133 (void*)&vStream,
Roman Divacky67923802012-09-05 21:17:34 +0000134 (const void*)MII,
Sean Callanan04cc3072009-12-19 02:59:52 +0000135 address,
136 fMode);
137
Sean Callanan010b3732010-04-02 21:23:51 +0000138 if (ret) {
Sean Callanan04cc3072009-12-19 02:59:52 +0000139 size = internalInstr.readerCursor - address;
Owen Andersona4043c42011-08-17 17:44:15 +0000140 return Fail;
Sean Callanan04cc3072009-12-19 02:59:52 +0000141 }
142 else {
143 size = internalInstr.length;
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000144 return (!translateInstruction(instr, internalInstr, this)) ?
145 Success : Fail;
Sean Callanan04cc3072009-12-19 02:59:52 +0000146 }
147}
148
149//
150// Private code that translates from struct InternalInstructions to MCInsts.
151//
152
153/// translateRegister - Translates an internal register to the appropriate LLVM
154/// register, and appends it as an operand to an MCInst.
155///
156/// @param mcInst - The MCInst to append to.
157/// @param reg - The Reg to append.
158static void translateRegister(MCInst &mcInst, Reg reg) {
159#define ENTRY(x) X86::x,
160 uint8_t llvmRegnums[] = {
161 ALL_REGS
162 0
163 };
164#undef ENTRY
165
166 uint8_t llvmRegnum = llvmRegnums[reg];
167 mcInst.addOperand(MCOperand::CreateReg(llvmRegnum));
168}
169
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000170/// tryAddingSymbolicOperand - trys to add a symbolic operand in place of the
171/// immediate Value in the MCInst.
172///
173/// @param Value - The immediate Value, has had any PC adjustment made by
174/// the caller.
175/// @param isBranch - If the instruction is a branch instruction
176/// @param Address - The starting address of the instruction
177/// @param Offset - The byte offset to this immediate in the instruction
178/// @param Width - The byte width of this immediate in the instruction
179///
180/// If the getOpInfo() function was set when setupForSymbolicDisassembly() was
181/// called then that function is called to get any symbolic information for the
182/// immediate in the instruction using the Address, Offset and Width. If that
183/// returns non-zero then the symbolic information it returns is used to create
184/// an MCExpr and that is added as an operand to the MCInst. If getOpInfo()
185/// returns zero and isBranch is true then a symbol look up for immediate Value
186/// is done and if a symbol is found an MCExpr is created with that, else
187/// an MCExpr with the immediate Value is created. This function returns true
188/// if it adds an operand to the MCInst and false otherwise.
189static bool tryAddingSymbolicOperand(int64_t Value, bool isBranch,
190 uint64_t Address, uint64_t Offset,
191 uint64_t Width, MCInst &MI,
192 const MCDisassembler *Dis) {
Ahmed Bougachaad1084d2013-05-24 00:39:57 +0000193 return Dis->tryAddingSymbolicOperand(MI, Value, Address, isBranch,
194 Offset, Width);
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000195}
196
Kevin Enderbyb119c082012-02-29 22:58:34 +0000197/// tryAddingPcLoadReferenceComment - trys to add a comment as to what is being
198/// referenced by a load instruction with the base register that is the rip.
199/// These can often be addresses in a literal pool. The Address of the
200/// instruction and its immediate Value are used to determine the address
201/// being referenced in the literal pool entry. The SymbolLookUp call back will
202/// return a pointer to a literal 'C' string if the referenced address is an
203/// address into a section with 'C' string literals.
204static void tryAddingPcLoadReferenceComment(uint64_t Address, uint64_t Value,
205 const void *Decoder) {
206 const MCDisassembler *Dis = static_cast<const MCDisassembler*>(Decoder);
Ahmed Bougachaad1084d2013-05-24 00:39:57 +0000207 Dis->tryAddingPcLoadReferenceComment(Value, Address);
Kevin Enderbyb119c082012-02-29 22:58:34 +0000208}
209
Sean Callanan04cc3072009-12-19 02:59:52 +0000210/// translateImmediate - Appends an immediate operand to an MCInst.
211///
212/// @param mcInst - The MCInst to append to.
213/// @param immediate - The immediate value to append.
Sean Callanan4cd930f2010-05-05 22:47:27 +0000214/// @param operand - The operand, as stored in the descriptor table.
215/// @param insn - The internal instruction.
Benjamin Kramerde0a4fb2010-10-23 09:10:44 +0000216static void translateImmediate(MCInst &mcInst, uint64_t immediate,
217 const OperandSpecifier &operand,
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000218 InternalInstruction &insn,
219 const MCDisassembler *Dis) {
Sean Callanan4cd930f2010-05-05 22:47:27 +0000220 // Sign-extend the immediate if necessary.
221
Craig Topper6dedbae2012-03-04 02:16:41 +0000222 OperandType type = (OperandType)operand.type;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000223
Kevin Enderbyec4bd312012-04-18 23:12:11 +0000224 bool isBranch = false;
225 uint64_t pcrel = 0;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000226 if (type == TYPE_RELv) {
Kevin Enderbyec4bd312012-04-18 23:12:11 +0000227 isBranch = true;
228 pcrel = insn.startLocation +
Kevin Enderby216ac312012-07-24 21:40:01 +0000229 insn.immediateOffset + insn.immediateSize;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000230 switch (insn.displacementSize) {
231 default:
232 break;
Sean Callanan5e8603d2011-02-21 21:55:05 +0000233 case 1:
Craig Topper18854172013-08-25 22:23:38 +0000234 if(immediate & 0x80)
235 immediate |= ~(0xffull);
Sean Callanan4cd930f2010-05-05 22:47:27 +0000236 break;
Sean Callanan5e8603d2011-02-21 21:55:05 +0000237 case 2:
Craig Topper18854172013-08-25 22:23:38 +0000238 if(immediate & 0x8000)
239 immediate |= ~(0xffffull);
Sean Callanan4cd930f2010-05-05 22:47:27 +0000240 break;
Sean Callanan5e8603d2011-02-21 21:55:05 +0000241 case 4:
Craig Topper18854172013-08-25 22:23:38 +0000242 if(immediate & 0x80000000)
243 immediate |= ~(0xffffffffull);
Sean Callanan4cd930f2010-05-05 22:47:27 +0000244 break;
Sean Callanan5e8603d2011-02-21 21:55:05 +0000245 case 8:
Sean Callanan4cd930f2010-05-05 22:47:27 +0000246 break;
247 }
248 }
Kevin Enderby5b03f722011-09-02 20:01:23 +0000249 // By default sign-extend all X86 immediates based on their encoding.
250 else if (type == TYPE_IMM8 || type == TYPE_IMM16 || type == TYPE_IMM32 ||
251 type == TYPE_IMM64) {
252 uint32_t Opcode = mcInst.getOpcode();
253 switch (operand.encoding) {
254 default:
255 break;
256 case ENCODING_IB:
257 // Special case those X86 instructions that use the imm8 as a set of
258 // bits, bit count, etc. and are not sign-extend.
259 if (Opcode != X86::BLENDPSrri && Opcode != X86::BLENDPDrri &&
Bill Wendlingea6397f2012-07-19 00:11:40 +0000260 Opcode != X86::PBLENDWrri && Opcode != X86::MPSADBWrri &&
261 Opcode != X86::DPPSrri && Opcode != X86::DPPDrri &&
262 Opcode != X86::INSERTPSrr && Opcode != X86::VBLENDPSYrri &&
263 Opcode != X86::VBLENDPSYrmi && Opcode != X86::VBLENDPDYrri &&
264 Opcode != X86::VBLENDPDYrmi && Opcode != X86::VPBLENDWrri &&
265 Opcode != X86::VMPSADBWrri && Opcode != X86::VDPPSYrri &&
266 Opcode != X86::VDPPSYrmi && Opcode != X86::VDPPDrri &&
267 Opcode != X86::VINSERTPSrr)
Craig Topper18854172013-08-25 22:23:38 +0000268 if(immediate & 0x80)
269 immediate |= ~(0xffull);
Kevin Enderby5b03f722011-09-02 20:01:23 +0000270 break;
271 case ENCODING_IW:
Craig Topper18854172013-08-25 22:23:38 +0000272 if(immediate & 0x8000)
273 immediate |= ~(0xffffull);
Kevin Enderby5b03f722011-09-02 20:01:23 +0000274 break;
275 case ENCODING_ID:
Craig Topper18854172013-08-25 22:23:38 +0000276 if(immediate & 0x80000000)
277 immediate |= ~(0xffffffffull);
Kevin Enderby5b03f722011-09-02 20:01:23 +0000278 break;
279 case ENCODING_IO:
Kevin Enderby5b03f722011-09-02 20:01:23 +0000280 break;
281 }
282 }
Sean Callanan4cd930f2010-05-05 22:47:27 +0000283
284 switch (type) {
Craig Topperc30fdbc2012-08-31 15:40:30 +0000285 case TYPE_XMM32:
286 case TYPE_XMM64:
Craig Topper96e00e52011-09-14 05:55:28 +0000287 case TYPE_XMM128:
288 mcInst.addOperand(MCOperand::CreateReg(X86::XMM0 + (immediate >> 4)));
289 return;
290 case TYPE_XMM256:
291 mcInst.addOperand(MCOperand::CreateReg(X86::YMM0 + (immediate >> 4)));
292 return;
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000293 case TYPE_XMM512:
294 mcInst.addOperand(MCOperand::CreateReg(X86::ZMM0 + (immediate >> 4)));
295 return;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000296 case TYPE_REL8:
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000297 isBranch = true;
298 pcrel = insn.startLocation + insn.immediateOffset + insn.immediateSize;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000299 if(immediate & 0x80)
300 immediate |= ~(0xffull);
301 break;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000302 case TYPE_REL32:
303 case TYPE_REL64:
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000304 isBranch = true;
305 pcrel = insn.startLocation + insn.immediateOffset + insn.immediateSize;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000306 if(immediate & 0x80000000)
307 immediate |= ~(0xffffffffull);
308 break;
Sean Callanan4cd930f2010-05-05 22:47:27 +0000309 default:
310 // operand is 64 bits wide. Do nothing.
311 break;
312 }
Craig Topper092e2fe2013-08-24 19:50:11 +0000313
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000314 if(!tryAddingSymbolicOperand(immediate + pcrel, isBranch, insn.startLocation,
315 insn.immediateOffset, insn.immediateSize,
316 mcInst, Dis))
317 mcInst.addOperand(MCOperand::CreateImm(immediate));
Sean Callanan04cc3072009-12-19 02:59:52 +0000318}
319
320/// translateRMRegister - Translates a register stored in the R/M field of the
321/// ModR/M byte to its LLVM equivalent and appends it to an MCInst.
322/// @param mcInst - The MCInst to append to.
323/// @param insn - The internal instruction to extract the R/M field
324/// from.
Sean Callanan010b3732010-04-02 21:23:51 +0000325/// @return - 0 on success; -1 otherwise
326static bool translateRMRegister(MCInst &mcInst,
Sean Callanan04cc3072009-12-19 02:59:52 +0000327 InternalInstruction &insn) {
Sean Callanan010b3732010-04-02 21:23:51 +0000328 if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
329 debug("A R/M register operand may not have a SIB byte");
330 return true;
331 }
Sean Callanan04cc3072009-12-19 02:59:52 +0000332
333 switch (insn.eaBase) {
Sean Callanan010b3732010-04-02 21:23:51 +0000334 default:
335 debug("Unexpected EA base register");
336 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000337 case EA_BASE_NONE:
Sean Callanan010b3732010-04-02 21:23:51 +0000338 debug("EA_BASE_NONE for ModR/M base");
339 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000340#define ENTRY(x) case EA_BASE_##x:
341 ALL_EA_BASES
342#undef ENTRY
Sean Callanan010b3732010-04-02 21:23:51 +0000343 debug("A R/M register operand may not have a base; "
344 "the operand must be a register.");
345 return true;
346#define ENTRY(x) \
Sean Callanan04cc3072009-12-19 02:59:52 +0000347 case EA_REG_##x: \
348 mcInst.addOperand(MCOperand::CreateReg(X86::x)); break;
349 ALL_REGS
350#undef ENTRY
Sean Callanan04cc3072009-12-19 02:59:52 +0000351 }
Sean Callanan010b3732010-04-02 21:23:51 +0000352
353 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000354}
355
356/// translateRMMemory - Translates a memory operand stored in the Mod and R/M
357/// fields of an internal instruction (and possibly its SIB byte) to a memory
358/// operand in LLVM's format, and appends it to an MCInst.
359///
360/// @param mcInst - The MCInst to append to.
361/// @param insn - The instruction to extract Mod, R/M, and SIB fields
362/// from.
Sean Callanan010b3732010-04-02 21:23:51 +0000363/// @return - 0 on success; nonzero otherwise
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000364static bool translateRMMemory(MCInst &mcInst, InternalInstruction &insn,
365 const MCDisassembler *Dis) {
Sean Callanan04cc3072009-12-19 02:59:52 +0000366 // Addresses in an MCInst are represented as five operands:
367 // 1. basereg (register) The R/M base, or (if there is a SIB) the
368 // SIB base
369 // 2. scaleamount (immediate) 1, or (if there is a SIB) the specified
370 // scale amount
371 // 3. indexreg (register) x86_registerNONE, or (if there is a SIB)
372 // the index (which is multiplied by the
373 // scale amount)
374 // 4. displacement (immediate) 0, or the displacement if there is one
375 // 5. segmentreg (register) x86_registerNONE for now, but could be set
376 // if we have segment overrides
377
378 MCOperand baseReg;
379 MCOperand scaleAmount;
380 MCOperand indexReg;
381 MCOperand displacement;
382 MCOperand segmentReg;
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000383 uint64_t pcrel = 0;
Sean Callanan04cc3072009-12-19 02:59:52 +0000384
385 if (insn.eaBase == EA_BASE_sib || insn.eaBase == EA_BASE_sib64) {
386 if (insn.sibBase != SIB_BASE_NONE) {
387 switch (insn.sibBase) {
388 default:
Sean Callanan010b3732010-04-02 21:23:51 +0000389 debug("Unexpected sibBase");
390 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000391#define ENTRY(x) \
Sean Callanan36eab802009-12-22 21:12:55 +0000392 case SIB_BASE_##x: \
Sean Callanan04cc3072009-12-19 02:59:52 +0000393 baseReg = MCOperand::CreateReg(X86::x); break;
394 ALL_SIB_BASES
395#undef ENTRY
396 }
397 } else {
398 baseReg = MCOperand::CreateReg(0);
399 }
Manman Rena0982042012-06-26 19:47:59 +0000400
401 // Check whether we are handling VSIB addressing mode for GATHER.
402 // If sibIndex was set to SIB_INDEX_NONE, index offset is 4 and
403 // we should use SIB_INDEX_XMM4|YMM4 for VSIB.
404 // I don't see a way to get the correct IndexReg in readSIB:
405 // We can tell whether it is VSIB or SIB after instruction ID is decoded,
406 // but instruction ID may not be decoded yet when calling readSIB.
407 uint32_t Opcode = mcInst.getOpcode();
Manman Ren98a5bf22012-06-29 00:54:20 +0000408 bool IndexIs128 = (Opcode == X86::VGATHERDPDrm ||
409 Opcode == X86::VGATHERDPDYrm ||
410 Opcode == X86::VGATHERQPDrm ||
411 Opcode == X86::VGATHERDPSrm ||
412 Opcode == X86::VGATHERQPSrm ||
413 Opcode == X86::VPGATHERDQrm ||
414 Opcode == X86::VPGATHERDQYrm ||
415 Opcode == X86::VPGATHERQQrm ||
416 Opcode == X86::VPGATHERDDrm ||
417 Opcode == X86::VPGATHERQDrm);
418 bool IndexIs256 = (Opcode == X86::VGATHERQPDYrm ||
419 Opcode == X86::VGATHERDPSYrm ||
420 Opcode == X86::VGATHERQPSYrm ||
421 Opcode == X86::VPGATHERQQYrm ||
422 Opcode == X86::VPGATHERDDYrm ||
423 Opcode == X86::VPGATHERQDYrm);
424 if (IndexIs128 || IndexIs256) {
Manman Rena0982042012-06-26 19:47:59 +0000425 unsigned IndexOffset = insn.sibIndex -
426 (insn.addressSize == 8 ? SIB_INDEX_RAX:SIB_INDEX_EAX);
Manman Ren98a5bf22012-06-29 00:54:20 +0000427 SIBIndex IndexBase = IndexIs256 ? SIB_INDEX_YMM0 : SIB_INDEX_XMM0;
Manman Rena0982042012-06-26 19:47:59 +0000428 insn.sibIndex = (SIBIndex)(IndexBase +
429 (insn.sibIndex == SIB_INDEX_NONE ? 4 : IndexOffset));
430 }
431
Sean Callanan04cc3072009-12-19 02:59:52 +0000432 if (insn.sibIndex != SIB_INDEX_NONE) {
433 switch (insn.sibIndex) {
434 default:
Sean Callanan010b3732010-04-02 21:23:51 +0000435 debug("Unexpected sibIndex");
436 return true;
Sean Callanan36eab802009-12-22 21:12:55 +0000437#define ENTRY(x) \
Sean Callanan04cc3072009-12-19 02:59:52 +0000438 case SIB_INDEX_##x: \
439 indexReg = MCOperand::CreateReg(X86::x); break;
440 EA_BASES_32BIT
441 EA_BASES_64BIT
Manman Rena0982042012-06-26 19:47:59 +0000442 REGS_XMM
443 REGS_YMM
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000444 REGS_ZMM
Sean Callanan04cc3072009-12-19 02:59:52 +0000445#undef ENTRY
446 }
447 } else {
448 indexReg = MCOperand::CreateReg(0);
449 }
450
451 scaleAmount = MCOperand::CreateImm(insn.sibScale);
452 } else {
453 switch (insn.eaBase) {
454 case EA_BASE_NONE:
Sean Callanan010b3732010-04-02 21:23:51 +0000455 if (insn.eaDisplacement == EA_DISP_NONE) {
456 debug("EA_BASE_NONE and EA_DISP_NONE for ModR/M base");
457 return true;
458 }
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000459 if (insn.mode == MODE_64BIT){
460 pcrel = insn.startLocation +
461 insn.displacementOffset + insn.displacementSize;
Kevin Enderbyb119c082012-02-29 22:58:34 +0000462 tryAddingPcLoadReferenceComment(insn.startLocation +
463 insn.displacementOffset,
464 insn.displacement + pcrel, Dis);
Sean Callanan04cc3072009-12-19 02:59:52 +0000465 baseReg = MCOperand::CreateReg(X86::RIP); // Section 2.2.1.6
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000466 }
Sean Callanan04cc3072009-12-19 02:59:52 +0000467 else
468 baseReg = MCOperand::CreateReg(0);
469
470 indexReg = MCOperand::CreateReg(0);
471 break;
472 case EA_BASE_BX_SI:
473 baseReg = MCOperand::CreateReg(X86::BX);
474 indexReg = MCOperand::CreateReg(X86::SI);
475 break;
476 case EA_BASE_BX_DI:
477 baseReg = MCOperand::CreateReg(X86::BX);
478 indexReg = MCOperand::CreateReg(X86::DI);
479 break;
480 case EA_BASE_BP_SI:
481 baseReg = MCOperand::CreateReg(X86::BP);
482 indexReg = MCOperand::CreateReg(X86::SI);
483 break;
484 case EA_BASE_BP_DI:
485 baseReg = MCOperand::CreateReg(X86::BP);
486 indexReg = MCOperand::CreateReg(X86::DI);
487 break;
488 default:
489 indexReg = MCOperand::CreateReg(0);
490 switch (insn.eaBase) {
491 default:
Sean Callanan010b3732010-04-02 21:23:51 +0000492 debug("Unexpected eaBase");
493 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000494 // Here, we will use the fill-ins defined above. However,
495 // BX_SI, BX_DI, BP_SI, and BP_DI are all handled above and
496 // sib and sib64 were handled in the top-level if, so they're only
497 // placeholders to keep the compiler happy.
498#define ENTRY(x) \
499 case EA_BASE_##x: \
500 baseReg = MCOperand::CreateReg(X86::x); break;
501 ALL_EA_BASES
502#undef ENTRY
503#define ENTRY(x) case EA_REG_##x:
504 ALL_REGS
505#undef ENTRY
Sean Callanan010b3732010-04-02 21:23:51 +0000506 debug("A R/M memory operand may not be a register; "
507 "the base field must be a base.");
508 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000509 }
510 }
Sean Callanan36eab802009-12-22 21:12:55 +0000511
512 scaleAmount = MCOperand::CreateImm(1);
Sean Callanan04cc3072009-12-19 02:59:52 +0000513 }
514
515 displacement = MCOperand::CreateImm(insn.displacement);
516
517 static const uint8_t segmentRegnums[SEG_OVERRIDE_max] = {
518 0, // SEG_OVERRIDE_NONE
519 X86::CS,
520 X86::SS,
521 X86::DS,
522 X86::ES,
523 X86::FS,
524 X86::GS
525 };
526
527 segmentReg = MCOperand::CreateReg(segmentRegnums[insn.segmentOverride]);
528
529 mcInst.addOperand(baseReg);
530 mcInst.addOperand(scaleAmount);
531 mcInst.addOperand(indexReg);
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000532 if(!tryAddingSymbolicOperand(insn.displacement + pcrel, false,
533 insn.startLocation, insn.displacementOffset,
534 insn.displacementSize, mcInst, Dis))
535 mcInst.addOperand(displacement);
Chris Lattner55595fb2010-07-13 04:23:55 +0000536 mcInst.addOperand(segmentReg);
Sean Callanan010b3732010-04-02 21:23:51 +0000537 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000538}
539
540/// translateRM - Translates an operand stored in the R/M (and possibly SIB)
541/// byte of an instruction to LLVM form, and appends it to an MCInst.
542///
543/// @param mcInst - The MCInst to append to.
544/// @param operand - The operand, as stored in the descriptor table.
545/// @param insn - The instruction to extract Mod, R/M, and SIB fields
546/// from.
Sean Callanan010b3732010-04-02 21:23:51 +0000547/// @return - 0 on success; nonzero otherwise
Benjamin Kramerde0a4fb2010-10-23 09:10:44 +0000548static bool translateRM(MCInst &mcInst, const OperandSpecifier &operand,
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000549 InternalInstruction &insn, const MCDisassembler *Dis) {
Sean Callanan04cc3072009-12-19 02:59:52 +0000550 switch (operand.type) {
551 default:
Sean Callanan010b3732010-04-02 21:23:51 +0000552 debug("Unexpected type for a R/M operand");
553 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000554 case TYPE_R8:
555 case TYPE_R16:
556 case TYPE_R32:
557 case TYPE_R64:
558 case TYPE_Rv:
559 case TYPE_MM:
560 case TYPE_MM32:
561 case TYPE_MM64:
562 case TYPE_XMM:
563 case TYPE_XMM32:
564 case TYPE_XMM64:
565 case TYPE_XMM128:
Sean Callananc3fd5232011-03-15 01:23:15 +0000566 case TYPE_XMM256:
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000567 case TYPE_XMM512:
Sean Callanan04cc3072009-12-19 02:59:52 +0000568 case TYPE_DEBUGREG:
Sean Callanane7e1cf92010-05-06 20:59:00 +0000569 case TYPE_CONTROLREG:
Sean Callanan010b3732010-04-02 21:23:51 +0000570 return translateRMRegister(mcInst, insn);
Sean Callanan04cc3072009-12-19 02:59:52 +0000571 case TYPE_M:
572 case TYPE_M8:
573 case TYPE_M16:
574 case TYPE_M32:
575 case TYPE_M64:
576 case TYPE_M128:
Sean Callananc3fd5232011-03-15 01:23:15 +0000577 case TYPE_M256:
Sean Callanan04cc3072009-12-19 02:59:52 +0000578 case TYPE_M512:
579 case TYPE_Mv:
580 case TYPE_M32FP:
581 case TYPE_M64FP:
582 case TYPE_M80FP:
583 case TYPE_M16INT:
584 case TYPE_M32INT:
585 case TYPE_M64INT:
586 case TYPE_M1616:
587 case TYPE_M1632:
588 case TYPE_M1664:
Sean Callanan36eab802009-12-22 21:12:55 +0000589 case TYPE_LEA:
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000590 return translateRMMemory(mcInst, insn, Dis);
Sean Callanan04cc3072009-12-19 02:59:52 +0000591 }
592}
593
594/// translateFPRegister - Translates a stack position on the FPU stack to its
595/// LLVM form, and appends it to an MCInst.
596///
597/// @param mcInst - The MCInst to append to.
598/// @param stackPos - The stack position to translate.
Sean Callanan010b3732010-04-02 21:23:51 +0000599/// @return - 0 on success; nonzero otherwise.
600static bool translateFPRegister(MCInst &mcInst,
601 uint8_t stackPos) {
602 if (stackPos >= 8) {
603 debug("Invalid FP stack position");
604 return true;
605 }
Sean Callanan04cc3072009-12-19 02:59:52 +0000606
607 mcInst.addOperand(MCOperand::CreateReg(X86::ST0 + stackPos));
Sean Callanan010b3732010-04-02 21:23:51 +0000608
609 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000610}
611
612/// translateOperand - Translates an operand stored in an internal instruction
613/// to LLVM's format and appends it to an MCInst.
614///
615/// @param mcInst - The MCInst to append to.
616/// @param operand - The operand, as stored in the descriptor table.
617/// @param insn - The internal instruction.
Sean Callanan010b3732010-04-02 21:23:51 +0000618/// @return - false on success; true otherwise.
Benjamin Kramerde0a4fb2010-10-23 09:10:44 +0000619static bool translateOperand(MCInst &mcInst, const OperandSpecifier &operand,
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000620 InternalInstruction &insn,
621 const MCDisassembler *Dis) {
Sean Callanan04cc3072009-12-19 02:59:52 +0000622 switch (operand.encoding) {
623 default:
Sean Callanan010b3732010-04-02 21:23:51 +0000624 debug("Unhandled operand encoding during translation");
625 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000626 case ENCODING_REG:
627 translateRegister(mcInst, insn.reg);
Sean Callanan010b3732010-04-02 21:23:51 +0000628 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000629 case ENCODING_RM:
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000630 return translateRM(mcInst, operand, insn, Dis);
Sean Callanan04cc3072009-12-19 02:59:52 +0000631 case ENCODING_CB:
632 case ENCODING_CW:
633 case ENCODING_CD:
634 case ENCODING_CP:
635 case ENCODING_CO:
636 case ENCODING_CT:
Sean Callanan010b3732010-04-02 21:23:51 +0000637 debug("Translation of code offsets isn't supported.");
638 return true;
Sean Callanan04cc3072009-12-19 02:59:52 +0000639 case ENCODING_IB:
640 case ENCODING_IW:
641 case ENCODING_ID:
642 case ENCODING_IO:
643 case ENCODING_Iv:
644 case ENCODING_Ia:
Sean Callanan4cd930f2010-05-05 22:47:27 +0000645 translateImmediate(mcInst,
646 insn.immediates[insn.numImmediatesTranslated++],
647 operand,
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000648 insn,
649 Dis);
Sean Callanan010b3732010-04-02 21:23:51 +0000650 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000651 case ENCODING_RB:
652 case ENCODING_RW:
653 case ENCODING_RD:
654 case ENCODING_RO:
655 translateRegister(mcInst, insn.opcodeRegister);
Sean Callanan010b3732010-04-02 21:23:51 +0000656 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000657 case ENCODING_I:
Sean Callanan010b3732010-04-02 21:23:51 +0000658 return translateFPRegister(mcInst, insn.opcodeModifier);
Sean Callanan04cc3072009-12-19 02:59:52 +0000659 case ENCODING_Rv:
660 translateRegister(mcInst, insn.opcodeRegister);
Sean Callanan010b3732010-04-02 21:23:51 +0000661 return false;
Sean Callananc3fd5232011-03-15 01:23:15 +0000662 case ENCODING_VVVV:
663 translateRegister(mcInst, insn.vvvv);
664 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000665 case ENCODING_DUP:
Craig Topperb8aec082012-08-01 07:39:18 +0000666 return translateOperand(mcInst, insn.operands[operand.type - TYPE_DUP0],
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000667 insn, Dis);
Sean Callanan04cc3072009-12-19 02:59:52 +0000668 }
669}
670
671/// translateInstruction - Translates an internal instruction and all its
672/// operands to an MCInst.
673///
674/// @param mcInst - The MCInst to populate with the instruction's data.
675/// @param insn - The internal instruction.
Sean Callanan010b3732010-04-02 21:23:51 +0000676/// @return - false on success; true otherwise.
677static bool translateInstruction(MCInst &mcInst,
Kevin Enderby6fbcd8d2012-02-23 18:18:17 +0000678 InternalInstruction &insn,
679 const MCDisassembler *Dis) {
Sean Callanan010b3732010-04-02 21:23:51 +0000680 if (!insn.spec) {
681 debug("Instruction has no specification");
682 return true;
683 }
Sean Callanan04cc3072009-12-19 02:59:52 +0000684
685 mcInst.setOpcode(insn.instructionID);
Kevin Enderby35fd7922013-06-20 22:32:18 +0000686 // If when reading the prefix bytes we determined the overlapping 0xf2 or 0xf3
687 // prefix bytes should be disassembled as xrelease and xacquire then set the
688 // opcode to those instead of the rep and repne opcodes.
689 if (insn.xAcquireRelease) {
690 if(mcInst.getOpcode() == X86::REP_PREFIX)
691 mcInst.setOpcode(X86::XRELEASE_PREFIX);
692 else if(mcInst.getOpcode() == X86::REPNE_PREFIX)
693 mcInst.setOpcode(X86::XACQUIRE_PREFIX);
694 }
Sean Callanan04cc3072009-12-19 02:59:52 +0000695
696 int index;
697
698 insn.numImmediatesTranslated = 0;
699
700 for (index = 0; index < X86_MAX_OPERANDS; ++index) {
Craig Topperb8aec082012-08-01 07:39:18 +0000701 if (insn.operands[index].encoding != ENCODING_NONE) {
702 if (translateOperand(mcInst, insn.operands[index], insn, Dis)) {
Sean Callanan010b3732010-04-02 21:23:51 +0000703 return true;
704 }
705 }
Sean Callanan04cc3072009-12-19 02:59:52 +0000706 }
Sean Callanan010b3732010-04-02 21:23:51 +0000707
708 return false;
Sean Callanan04cc3072009-12-19 02:59:52 +0000709}
Daniel Dunbar900f2ce2009-11-25 06:53:08 +0000710
Benjamin Kramer478e8de2012-02-11 14:50:54 +0000711static MCDisassembler *createX86_32Disassembler(const Target &T,
712 const MCSubtargetInfo &STI) {
713 return new X86Disassembler::X86GenericDisassembler(STI, MODE_32BIT,
714 T.createMCInstrInfo());
Daniel Dunbar900f2ce2009-11-25 06:53:08 +0000715}
716
Benjamin Kramer478e8de2012-02-11 14:50:54 +0000717static MCDisassembler *createX86_64Disassembler(const Target &T,
718 const MCSubtargetInfo &STI) {
719 return new X86Disassembler::X86GenericDisassembler(STI, MODE_64BIT,
720 T.createMCInstrInfo());
Daniel Dunbar900f2ce2009-11-25 06:53:08 +0000721}
722
723extern "C" void LLVMInitializeX86Disassembler() {
724 // Register the disassembler.
725 TargetRegistry::RegisterMCDisassembler(TheX86_32Target,
726 createX86_32Disassembler);
727 TargetRegistry::RegisterMCDisassembler(TheX86_64Target,
728 createX86_64Disassembler);
729}