blob: 4213912b91b73b887e293c525d9a4146992a84bb [file] [log] [blame]
Steve Blocka7e24c12009-10-30 11:49:00 +00001// Copyright 2009 the V8 project authors. All rights reserved.
2// Redistribution and use in source and binary forms, with or without
3// modification, are permitted provided that the following conditions are
4// met:
5//
6// * Redistributions of source code must retain the above copyright
7// notice, this list of conditions and the following disclaimer.
8// * Redistributions in binary form must reproduce the above
9// copyright notice, this list of conditions and the following
10// disclaimer in the documentation and/or other materials provided
11// with the distribution.
12// * Neither the name of Google Inc. nor the names of its
13// contributors may be used to endorse or promote products derived
14// from this software without specific prior written permission.
15//
16// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27
28#include <assert.h>
29#include <stdio.h>
30#include <stdarg.h>
31
32#include "v8.h"
Leon Clarkef7060e22010-06-03 12:02:55 +010033
34#if defined(V8_TARGET_ARCH_X64)
35
Steve Blocka7e24c12009-10-30 11:49:00 +000036#include "disasm.h"
37
38namespace disasm {
39
40enum OperandType {
41 UNSET_OP_ORDER = 0,
42 // Operand size decides between 16, 32 and 64 bit operands.
43 REG_OPER_OP_ORDER = 1, // Register destination, operand source.
44 OPER_REG_OP_ORDER = 2, // Operand destination, register source.
45 // Fixed 8-bit operands.
46 BYTE_SIZE_OPERAND_FLAG = 4,
47 BYTE_REG_OPER_OP_ORDER = REG_OPER_OP_ORDER | BYTE_SIZE_OPERAND_FLAG,
48 BYTE_OPER_REG_OP_ORDER = OPER_REG_OP_ORDER | BYTE_SIZE_OPERAND_FLAG
49};
50
51//------------------------------------------------------------------
52// Tables
53//------------------------------------------------------------------
54struct ByteMnemonic {
55 int b; // -1 terminates, otherwise must be in range (0..255)
56 OperandType op_order_;
57 const char* mnem;
58};
59
60
61static ByteMnemonic two_operands_instr[] = {
62 { 0x00, BYTE_OPER_REG_OP_ORDER, "add" },
63 { 0x01, OPER_REG_OP_ORDER, "add" },
64 { 0x02, BYTE_REG_OPER_OP_ORDER, "add" },
65 { 0x03, REG_OPER_OP_ORDER, "add" },
66 { 0x08, BYTE_OPER_REG_OP_ORDER, "or" },
67 { 0x09, OPER_REG_OP_ORDER, "or" },
68 { 0x0A, BYTE_REG_OPER_OP_ORDER, "or" },
69 { 0x0B, REG_OPER_OP_ORDER, "or" },
70 { 0x10, BYTE_OPER_REG_OP_ORDER, "adc" },
71 { 0x11, OPER_REG_OP_ORDER, "adc" },
72 { 0x12, BYTE_REG_OPER_OP_ORDER, "adc" },
73 { 0x13, REG_OPER_OP_ORDER, "adc" },
74 { 0x18, BYTE_OPER_REG_OP_ORDER, "sbb" },
75 { 0x19, OPER_REG_OP_ORDER, "sbb" },
76 { 0x1A, BYTE_REG_OPER_OP_ORDER, "sbb" },
77 { 0x1B, REG_OPER_OP_ORDER, "sbb" },
78 { 0x20, BYTE_OPER_REG_OP_ORDER, "and" },
79 { 0x21, OPER_REG_OP_ORDER, "and" },
80 { 0x22, BYTE_REG_OPER_OP_ORDER, "and" },
81 { 0x23, REG_OPER_OP_ORDER, "and" },
82 { 0x28, BYTE_OPER_REG_OP_ORDER, "sub" },
83 { 0x29, OPER_REG_OP_ORDER, "sub" },
84 { 0x2A, BYTE_REG_OPER_OP_ORDER, "sub" },
85 { 0x2B, REG_OPER_OP_ORDER, "sub" },
86 { 0x30, BYTE_OPER_REG_OP_ORDER, "xor" },
87 { 0x31, OPER_REG_OP_ORDER, "xor" },
88 { 0x32, BYTE_REG_OPER_OP_ORDER, "xor" },
89 { 0x33, REG_OPER_OP_ORDER, "xor" },
90 { 0x38, BYTE_OPER_REG_OP_ORDER, "cmp" },
91 { 0x39, OPER_REG_OP_ORDER, "cmp" },
92 { 0x3A, BYTE_REG_OPER_OP_ORDER, "cmp" },
93 { 0x3B, REG_OPER_OP_ORDER, "cmp" },
94 { 0x63, REG_OPER_OP_ORDER, "movsxlq" },
95 { 0x84, BYTE_REG_OPER_OP_ORDER, "test" },
96 { 0x85, REG_OPER_OP_ORDER, "test" },
97 { 0x86, BYTE_REG_OPER_OP_ORDER, "xchg" },
98 { 0x87, REG_OPER_OP_ORDER, "xchg" },
99 { 0x88, BYTE_OPER_REG_OP_ORDER, "mov" },
100 { 0x89, OPER_REG_OP_ORDER, "mov" },
101 { 0x8A, BYTE_REG_OPER_OP_ORDER, "mov" },
102 { 0x8B, REG_OPER_OP_ORDER, "mov" },
103 { 0x8D, REG_OPER_OP_ORDER, "lea" },
104 { -1, UNSET_OP_ORDER, "" }
105};
106
107
108static ByteMnemonic zero_operands_instr[] = {
109 { 0xC3, UNSET_OP_ORDER, "ret" },
110 { 0xC9, UNSET_OP_ORDER, "leave" },
111 { 0xF4, UNSET_OP_ORDER, "hlt" },
112 { 0xCC, UNSET_OP_ORDER, "int3" },
113 { 0x60, UNSET_OP_ORDER, "pushad" },
114 { 0x61, UNSET_OP_ORDER, "popad" },
115 { 0x9C, UNSET_OP_ORDER, "pushfd" },
116 { 0x9D, UNSET_OP_ORDER, "popfd" },
117 { 0x9E, UNSET_OP_ORDER, "sahf" },
118 { 0x99, UNSET_OP_ORDER, "cdq" },
119 { 0x9B, UNSET_OP_ORDER, "fwait" },
Leon Clarked91b9f72010-01-27 17:25:45 +0000120 { 0xA4, UNSET_OP_ORDER, "movs" },
121 { 0xA5, UNSET_OP_ORDER, "movs" },
122 { 0xA6, UNSET_OP_ORDER, "cmps" },
123 { 0xA7, UNSET_OP_ORDER, "cmps" },
Steve Blocka7e24c12009-10-30 11:49:00 +0000124 { -1, UNSET_OP_ORDER, "" }
125};
126
127
128static ByteMnemonic call_jump_instr[] = {
129 { 0xE8, UNSET_OP_ORDER, "call" },
130 { 0xE9, UNSET_OP_ORDER, "jmp" },
131 { -1, UNSET_OP_ORDER, "" }
132};
133
134
135static ByteMnemonic short_immediate_instr[] = {
136 { 0x05, UNSET_OP_ORDER, "add" },
137 { 0x0D, UNSET_OP_ORDER, "or" },
138 { 0x15, UNSET_OP_ORDER, "adc" },
139 { 0x1D, UNSET_OP_ORDER, "sbb" },
140 { 0x25, UNSET_OP_ORDER, "and" },
141 { 0x2D, UNSET_OP_ORDER, "sub" },
142 { 0x35, UNSET_OP_ORDER, "xor" },
143 { 0x3D, UNSET_OP_ORDER, "cmp" },
144 { -1, UNSET_OP_ORDER, "" }
145};
146
147
148static const char* conditional_code_suffix[] = {
149 "o", "no", "c", "nc", "z", "nz", "na", "a",
150 "s", "ns", "pe", "po", "l", "ge", "le", "g"
151};
152
153
154enum InstructionType {
155 NO_INSTR,
156 ZERO_OPERANDS_INSTR,
157 TWO_OPERANDS_INSTR,
158 JUMP_CONDITIONAL_SHORT_INSTR,
159 REGISTER_INSTR,
160 PUSHPOP_INSTR, // Has implicit 64-bit operand size.
161 MOVE_REG_INSTR,
162 CALL_JUMP_INSTR,
163 SHORT_IMMEDIATE_INSTR
164};
165
166
Leon Clarked91b9f72010-01-27 17:25:45 +0000167enum Prefixes {
168 ESCAPE_PREFIX = 0x0F,
169 OPERAND_SIZE_OVERRIDE_PREFIX = 0x66,
170 ADDRESS_SIZE_OVERRIDE_PREFIX = 0x67,
171 REPNE_PREFIX = 0xF2,
172 REP_PREFIX = 0xF3,
173 REPEQ_PREFIX = REP_PREFIX
174};
175
176
Steve Blocka7e24c12009-10-30 11:49:00 +0000177struct InstructionDesc {
178 const char* mnem;
179 InstructionType type;
180 OperandType op_order_;
181 bool byte_size_operation; // Fixed 8-bit operation.
182};
183
184
185class InstructionTable {
186 public:
187 InstructionTable();
188 const InstructionDesc& Get(byte x) const {
189 return instructions_[x];
190 }
191
192 private:
193 InstructionDesc instructions_[256];
194 void Clear();
195 void Init();
196 void CopyTable(ByteMnemonic bm[], InstructionType type);
197 void SetTableRange(InstructionType type, byte start, byte end, bool byte_size,
198 const char* mnem);
199 void AddJumpConditionalShort();
200};
201
202
203InstructionTable::InstructionTable() {
204 Clear();
205 Init();
206}
207
208
209void InstructionTable::Clear() {
210 for (int i = 0; i < 256; i++) {
211 instructions_[i].mnem = "(bad)";
212 instructions_[i].type = NO_INSTR;
213 instructions_[i].op_order_ = UNSET_OP_ORDER;
214 instructions_[i].byte_size_operation = false;
215 }
216}
217
218
219void InstructionTable::Init() {
220 CopyTable(two_operands_instr, TWO_OPERANDS_INSTR);
221 CopyTable(zero_operands_instr, ZERO_OPERANDS_INSTR);
222 CopyTable(call_jump_instr, CALL_JUMP_INSTR);
223 CopyTable(short_immediate_instr, SHORT_IMMEDIATE_INSTR);
224 AddJumpConditionalShort();
225 SetTableRange(PUSHPOP_INSTR, 0x50, 0x57, false, "push");
226 SetTableRange(PUSHPOP_INSTR, 0x58, 0x5F, false, "pop");
227 SetTableRange(MOVE_REG_INSTR, 0xB8, 0xBF, false, "mov");
228}
229
230
231void InstructionTable::CopyTable(ByteMnemonic bm[], InstructionType type) {
232 for (int i = 0; bm[i].b >= 0; i++) {
233 InstructionDesc* id = &instructions_[bm[i].b];
234 id->mnem = bm[i].mnem;
235 OperandType op_order = bm[i].op_order_;
236 id->op_order_ =
237 static_cast<OperandType>(op_order & ~BYTE_SIZE_OPERAND_FLAG);
Steve Blockd0582a62009-12-15 09:54:21 +0000238 ASSERT_EQ(NO_INSTR, id->type); // Information not already entered
Steve Blocka7e24c12009-10-30 11:49:00 +0000239 id->type = type;
240 id->byte_size_operation = ((op_order & BYTE_SIZE_OPERAND_FLAG) != 0);
241 }
242}
243
244
245void InstructionTable::SetTableRange(InstructionType type,
246 byte start,
247 byte end,
248 bool byte_size,
249 const char* mnem) {
250 for (byte b = start; b <= end; b++) {
251 InstructionDesc* id = &instructions_[b];
Steve Blockd0582a62009-12-15 09:54:21 +0000252 ASSERT_EQ(NO_INSTR, id->type); // Information not already entered
Steve Blocka7e24c12009-10-30 11:49:00 +0000253 id->mnem = mnem;
254 id->type = type;
255 id->byte_size_operation = byte_size;
256 }
257}
258
259
260void InstructionTable::AddJumpConditionalShort() {
261 for (byte b = 0x70; b <= 0x7F; b++) {
262 InstructionDesc* id = &instructions_[b];
Steve Blockd0582a62009-12-15 09:54:21 +0000263 ASSERT_EQ(NO_INSTR, id->type); // Information not already entered
Steve Blocka7e24c12009-10-30 11:49:00 +0000264 id->mnem = NULL; // Computed depending on condition code.
265 id->type = JUMP_CONDITIONAL_SHORT_INSTR;
266 }
267}
268
269
270static InstructionTable instruction_table;
271
272static InstructionDesc cmov_instructions[16] = {
273 {"cmovo", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
274 {"cmovno", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
275 {"cmovc", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
276 {"cmovnc", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
277 {"cmovz", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
278 {"cmovnz", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
279 {"cmovna", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
280 {"cmova", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
281 {"cmovs", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
282 {"cmovns", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
283 {"cmovpe", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
284 {"cmovpo", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
285 {"cmovl", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
286 {"cmovge", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
287 {"cmovle", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false},
288 {"cmovg", TWO_OPERANDS_INSTR, REG_OPER_OP_ORDER, false}
289};
290
291//------------------------------------------------------------------------------
292// DisassemblerX64 implementation.
293
294enum UnimplementedOpcodeAction {
295 CONTINUE_ON_UNIMPLEMENTED_OPCODE,
296 ABORT_ON_UNIMPLEMENTED_OPCODE
297};
298
299// A new DisassemblerX64 object is created to disassemble each instruction.
300// The object can only disassemble a single instruction.
301class DisassemblerX64 {
302 public:
303 DisassemblerX64(const NameConverter& converter,
304 UnimplementedOpcodeAction unimplemented_action =
305 ABORT_ON_UNIMPLEMENTED_OPCODE)
306 : converter_(converter),
307 tmp_buffer_pos_(0),
308 abort_on_unimplemented_(
309 unimplemented_action == ABORT_ON_UNIMPLEMENTED_OPCODE),
310 rex_(0),
311 operand_size_(0),
312 group_1_prefix_(0),
313 byte_size_operand_(false) {
314 tmp_buffer_[0] = '\0';
315 }
316
317 virtual ~DisassemblerX64() {
318 }
319
320 // Writes one disassembled instruction into 'buffer' (0-terminated).
321 // Returns the length of the disassembled machine instruction in bytes.
322 int InstructionDecode(v8::internal::Vector<char> buffer, byte* instruction);
323
324 private:
325 enum OperandSize {
326 BYTE_SIZE = 0,
327 WORD_SIZE = 1,
328 DOUBLEWORD_SIZE = 2,
329 QUADWORD_SIZE = 3
330 };
331
332 const NameConverter& converter_;
333 v8::internal::EmbeddedVector<char, 128> tmp_buffer_;
334 unsigned int tmp_buffer_pos_;
335 bool abort_on_unimplemented_;
336 // Prefixes parsed
337 byte rex_;
338 byte operand_size_; // 0x66 or (if no group 3 prefix is present) 0x0.
339 byte group_1_prefix_; // 0xF2, 0xF3, or (if no group 1 prefix is present) 0.
340 // Byte size operand override.
341 bool byte_size_operand_;
342
343 void setRex(byte rex) {
344 ASSERT_EQ(0x40, rex & 0xF0);
345 rex_ = rex;
346 }
347
348 bool rex() { return rex_ != 0; }
349
350 bool rex_b() { return (rex_ & 0x01) != 0; }
351
352 // Actual number of base register given the low bits and the rex.b state.
353 int base_reg(int low_bits) { return low_bits | ((rex_ & 0x01) << 3); }
354
355 bool rex_x() { return (rex_ & 0x02) != 0; }
356
357 bool rex_r() { return (rex_ & 0x04) != 0; }
358
359 bool rex_w() { return (rex_ & 0x08) != 0; }
360
361 OperandSize operand_size() {
362 if (byte_size_operand_) return BYTE_SIZE;
363 if (rex_w()) return QUADWORD_SIZE;
364 if (operand_size_ != 0) return WORD_SIZE;
365 return DOUBLEWORD_SIZE;
366 }
367
368 char operand_size_code() {
369 return "bwlq"[operand_size()];
370 }
371
372 const char* NameOfCPURegister(int reg) const {
373 return converter_.NameOfCPURegister(reg);
374 }
375
376 const char* NameOfByteCPURegister(int reg) const {
377 return converter_.NameOfByteCPURegister(reg);
378 }
379
380 const char* NameOfXMMRegister(int reg) const {
381 return converter_.NameOfXMMRegister(reg);
382 }
383
384 const char* NameOfAddress(byte* addr) const {
385 return converter_.NameOfAddress(addr);
386 }
387
388 // Disassembler helper functions.
389 void get_modrm(byte data,
390 int* mod,
391 int* regop,
392 int* rm) {
393 *mod = (data >> 6) & 3;
394 *regop = ((data & 0x38) >> 3) | (rex_r() ? 8 : 0);
395 *rm = (data & 7) | (rex_b() ? 8 : 0);
396 }
397
398 void get_sib(byte data,
399 int* scale,
400 int* index,
401 int* base) {
402 *scale = (data >> 6) & 3;
403 *index = ((data >> 3) & 7) | (rex_x() ? 8 : 0);
404 *base = (data & 7) | (rex_b() ? 8 : 0);
405 }
406
407 typedef const char* (DisassemblerX64::*RegisterNameMapping)(int reg) const;
408
409 int PrintRightOperandHelper(byte* modrmp,
410 RegisterNameMapping register_name);
411 int PrintRightOperand(byte* modrmp);
412 int PrintRightByteOperand(byte* modrmp);
Steve Blockd0582a62009-12-15 09:54:21 +0000413 int PrintRightXMMOperand(byte* modrmp);
Steve Blocka7e24c12009-10-30 11:49:00 +0000414 int PrintOperands(const char* mnem,
415 OperandType op_order,
416 byte* data);
417 int PrintImmediate(byte* data, OperandSize size);
418 int PrintImmediateOp(byte* data);
419 const char* TwoByteMnemonic(byte opcode);
420 int TwoByteOpcodeInstruction(byte* data);
Steve Blockd0582a62009-12-15 09:54:21 +0000421 int F6F7Instruction(byte* data);
Steve Blocka7e24c12009-10-30 11:49:00 +0000422 int ShiftInstruction(byte* data);
423 int JumpShort(byte* data);
424 int JumpConditional(byte* data);
425 int JumpConditionalShort(byte* data);
426 int SetCC(byte* data);
427 int FPUInstruction(byte* data);
Steve Blockd0582a62009-12-15 09:54:21 +0000428 int MemoryFPUInstruction(int escape_opcode, int regop, byte* modrm_start);
429 int RegisterFPUInstruction(int escape_opcode, byte modrm_byte);
Steve Blocka7e24c12009-10-30 11:49:00 +0000430 void AppendToBuffer(const char* format, ...);
431
432 void UnimplementedInstruction() {
433 if (abort_on_unimplemented_) {
434 CHECK(false);
435 } else {
436 AppendToBuffer("'Unimplemented Instruction'");
437 }
438 }
439};
440
441
442void DisassemblerX64::AppendToBuffer(const char* format, ...) {
443 v8::internal::Vector<char> buf = tmp_buffer_ + tmp_buffer_pos_;
444 va_list args;
445 va_start(args, format);
446 int result = v8::internal::OS::VSNPrintF(buf, format, args);
447 va_end(args);
448 tmp_buffer_pos_ += result;
449}
450
451
452int DisassemblerX64::PrintRightOperandHelper(
453 byte* modrmp,
454 RegisterNameMapping register_name) {
455 int mod, regop, rm;
456 get_modrm(*modrmp, &mod, &regop, &rm);
457 switch (mod) {
458 case 0:
459 if ((rm & 7) == 5) {
460 int32_t disp = *reinterpret_cast<int32_t*>(modrmp + 1);
461 AppendToBuffer("[0x%x]", disp);
462 return 5;
463 } else if ((rm & 7) == 4) {
464 // Codes for SIB byte.
465 byte sib = *(modrmp + 1);
466 int scale, index, base;
467 get_sib(sib, &scale, &index, &base);
468 if (index == 4 && (base & 7) == 4 && scale == 0 /*times_1*/) {
469 // index == rsp means no index. Only use sib byte with no index for
470 // rsp and r12 base.
Steve Block8defd9f2010-07-08 12:39:36 +0100471 AppendToBuffer("[%s]", NameOfCPURegister(base));
Steve Blocka7e24c12009-10-30 11:49:00 +0000472 return 2;
473 } else if (base == 5) {
474 // base == rbp means no base register (when mod == 0).
475 int32_t disp = *reinterpret_cast<int32_t*>(modrmp + 2);
476 AppendToBuffer("[%s*%d+0x%x]",
Steve Block8defd9f2010-07-08 12:39:36 +0100477 NameOfCPURegister(index),
Steve Blocka7e24c12009-10-30 11:49:00 +0000478 1 << scale, disp);
479 return 6;
480 } else if (index != 4 && base != 5) {
481 // [base+index*scale]
482 AppendToBuffer("[%s+%s*%d]",
Steve Block8defd9f2010-07-08 12:39:36 +0100483 NameOfCPURegister(base),
484 NameOfCPURegister(index),
Steve Blocka7e24c12009-10-30 11:49:00 +0000485 1 << scale);
486 return 2;
487 } else {
488 UnimplementedInstruction();
489 return 1;
490 }
491 } else {
Steve Block8defd9f2010-07-08 12:39:36 +0100492 AppendToBuffer("[%s]", NameOfCPURegister(rm));
Steve Blocka7e24c12009-10-30 11:49:00 +0000493 return 1;
494 }
495 break;
496 case 1: // fall through
497 case 2:
498 if ((rm & 7) == 4) {
499 byte sib = *(modrmp + 1);
500 int scale, index, base;
501 get_sib(sib, &scale, &index, &base);
502 int disp = (mod == 2) ? *reinterpret_cast<int32_t*>(modrmp + 2)
503 : *reinterpret_cast<char*>(modrmp + 2);
504 if (index == 4 && (base & 7) == 4 && scale == 0 /*times_1*/) {
505 if (-disp > 0) {
Steve Block8defd9f2010-07-08 12:39:36 +0100506 AppendToBuffer("[%s-0x%x]", NameOfCPURegister(base), -disp);
Steve Blocka7e24c12009-10-30 11:49:00 +0000507 } else {
Steve Block8defd9f2010-07-08 12:39:36 +0100508 AppendToBuffer("[%s+0x%x]", NameOfCPURegister(base), disp);
Steve Blocka7e24c12009-10-30 11:49:00 +0000509 }
510 } else {
511 if (-disp > 0) {
512 AppendToBuffer("[%s+%s*%d-0x%x]",
Steve Block8defd9f2010-07-08 12:39:36 +0100513 NameOfCPURegister(base),
514 NameOfCPURegister(index),
Steve Blocka7e24c12009-10-30 11:49:00 +0000515 1 << scale,
516 -disp);
517 } else {
518 AppendToBuffer("[%s+%s*%d+0x%x]",
Steve Block8defd9f2010-07-08 12:39:36 +0100519 NameOfCPURegister(base),
520 NameOfCPURegister(index),
Steve Blocka7e24c12009-10-30 11:49:00 +0000521 1 << scale,
522 disp);
523 }
524 }
525 return mod == 2 ? 6 : 3;
526 } else {
527 // No sib.
528 int disp = (mod == 2) ? *reinterpret_cast<int32_t*>(modrmp + 1)
529 : *reinterpret_cast<char*>(modrmp + 1);
530 if (-disp > 0) {
Steve Block8defd9f2010-07-08 12:39:36 +0100531 AppendToBuffer("[%s-0x%x]", NameOfCPURegister(rm), -disp);
Steve Blocka7e24c12009-10-30 11:49:00 +0000532 } else {
Steve Block8defd9f2010-07-08 12:39:36 +0100533 AppendToBuffer("[%s+0x%x]", NameOfCPURegister(rm), disp);
Steve Blocka7e24c12009-10-30 11:49:00 +0000534 }
535 return (mod == 2) ? 5 : 2;
536 }
537 break;
538 case 3:
539 AppendToBuffer("%s", (this->*register_name)(rm));
540 return 1;
541 default:
542 UnimplementedInstruction();
543 return 1;
544 }
545 UNREACHABLE();
546}
547
548
549int DisassemblerX64::PrintImmediate(byte* data, OperandSize size) {
550 int64_t value;
551 int count;
552 switch (size) {
553 case BYTE_SIZE:
554 value = *data;
555 count = 1;
556 break;
557 case WORD_SIZE:
558 value = *reinterpret_cast<int16_t*>(data);
559 count = 2;
560 break;
561 case DOUBLEWORD_SIZE:
562 value = *reinterpret_cast<uint32_t*>(data);
563 count = 4;
564 break;
565 case QUADWORD_SIZE:
566 value = *reinterpret_cast<int32_t*>(data);
567 count = 4;
568 break;
569 default:
570 UNREACHABLE();
571 value = 0; // Initialize variables on all paths to satisfy the compiler.
572 count = 0;
573 }
574 AppendToBuffer("%" V8_PTR_PREFIX "x", value);
575 return count;
576}
577
578
579int DisassemblerX64::PrintRightOperand(byte* modrmp) {
580 return PrintRightOperandHelper(modrmp,
581 &DisassemblerX64::NameOfCPURegister);
582}
583
584
585int DisassemblerX64::PrintRightByteOperand(byte* modrmp) {
586 return PrintRightOperandHelper(modrmp,
587 &DisassemblerX64::NameOfByteCPURegister);
588}
589
590
Steve Blockd0582a62009-12-15 09:54:21 +0000591int DisassemblerX64::PrintRightXMMOperand(byte* modrmp) {
592 return PrintRightOperandHelper(modrmp,
593 &DisassemblerX64::NameOfXMMRegister);
594}
595
596
Steve Blocka7e24c12009-10-30 11:49:00 +0000597// Returns number of bytes used including the current *data.
598// Writes instruction's mnemonic, left and right operands to 'tmp_buffer_'.
599int DisassemblerX64::PrintOperands(const char* mnem,
600 OperandType op_order,
601 byte* data) {
602 byte modrm = *data;
603 int mod, regop, rm;
604 get_modrm(modrm, &mod, &regop, &rm);
605 int advance = 0;
606 const char* register_name =
607 byte_size_operand_ ? NameOfByteCPURegister(regop)
608 : NameOfCPURegister(regop);
609 switch (op_order) {
610 case REG_OPER_OP_ORDER: {
611 AppendToBuffer("%s%c %s,",
612 mnem,
613 operand_size_code(),
614 register_name);
615 advance = byte_size_operand_ ? PrintRightByteOperand(data)
616 : PrintRightOperand(data);
617 break;
618 }
619 case OPER_REG_OP_ORDER: {
620 AppendToBuffer("%s%c ", mnem, operand_size_code());
621 advance = byte_size_operand_ ? PrintRightByteOperand(data)
622 : PrintRightOperand(data);
623 AppendToBuffer(",%s", register_name);
624 break;
625 }
626 default:
627 UNREACHABLE();
628 break;
629 }
630 return advance;
631}
632
633
634// Returns number of bytes used by machine instruction, including *data byte.
635// Writes immediate instructions to 'tmp_buffer_'.
636int DisassemblerX64::PrintImmediateOp(byte* data) {
637 bool byte_size_immediate = (*data & 0x02) != 0;
638 byte modrm = *(data + 1);
639 int mod, regop, rm;
640 get_modrm(modrm, &mod, &regop, &rm);
641 const char* mnem = "Imm???";
642 switch (regop) {
643 case 0:
644 mnem = "add";
645 break;
646 case 1:
647 mnem = "or";
648 break;
649 case 2:
650 mnem = "adc";
651 break;
652 case 4:
653 mnem = "and";
654 break;
655 case 5:
656 mnem = "sub";
657 break;
658 case 6:
659 mnem = "xor";
660 break;
661 case 7:
662 mnem = "cmp";
663 break;
664 default:
665 UnimplementedInstruction();
666 }
667 AppendToBuffer("%s%c ", mnem, operand_size_code());
668 int count = PrintRightOperand(data + 1);
669 AppendToBuffer(",0x");
670 OperandSize immediate_size = byte_size_immediate ? BYTE_SIZE : operand_size();
671 count += PrintImmediate(data + 1 + count, immediate_size);
672 return 1 + count;
673}
674
675
676// Returns number of bytes used, including *data.
Steve Blockd0582a62009-12-15 09:54:21 +0000677int DisassemblerX64::F6F7Instruction(byte* data) {
678 ASSERT(*data == 0xF7 || *data == 0xF6);
Steve Blocka7e24c12009-10-30 11:49:00 +0000679 byte modrm = *(data + 1);
680 int mod, regop, rm;
681 get_modrm(modrm, &mod, &regop, &rm);
682 if (mod == 3 && regop != 0) {
683 const char* mnem = NULL;
684 switch (regop) {
685 case 2:
686 mnem = "not";
687 break;
688 case 3:
689 mnem = "neg";
690 break;
691 case 4:
692 mnem = "mul";
693 break;
694 case 7:
695 mnem = "idiv";
696 break;
697 default:
698 UnimplementedInstruction();
699 }
700 AppendToBuffer("%s%c %s",
701 mnem,
702 operand_size_code(),
703 NameOfCPURegister(rm));
704 return 2;
Steve Blocka7e24c12009-10-30 11:49:00 +0000705 } else if (regop == 0) {
706 AppendToBuffer("test%c ", operand_size_code());
Steve Blockd0582a62009-12-15 09:54:21 +0000707 int count = PrintRightOperand(data + 1); // Use name of 64-bit register.
708 AppendToBuffer(",0x");
709 count += PrintImmediate(data + 1 + count, operand_size());
710 return 1 + count;
Steve Blocka7e24c12009-10-30 11:49:00 +0000711 } else {
712 UnimplementedInstruction();
713 return 2;
714 }
715}
716
717
718int DisassemblerX64::ShiftInstruction(byte* data) {
719 byte op = *data & (~1);
720 if (op != 0xD0 && op != 0xD2 && op != 0xC0) {
721 UnimplementedInstruction();
722 return 1;
723 }
724 byte modrm = *(data + 1);
725 int mod, regop, rm;
726 get_modrm(modrm, &mod, &regop, &rm);
727 regop &= 0x7; // The REX.R bit does not affect the operation.
728 int imm8 = -1;
729 int num_bytes = 2;
730 if (mod != 3) {
731 UnimplementedInstruction();
732 return num_bytes;
733 }
734 const char* mnem = NULL;
735 switch (regop) {
736 case 0:
737 mnem = "rol";
738 break;
739 case 1:
740 mnem = "ror";
741 break;
742 case 2:
743 mnem = "rcl";
744 break;
745 case 3:
746 mnem = "rcr";
747 break;
748 case 4:
749 mnem = "shl";
750 break;
751 case 5:
752 mnem = "shr";
753 break;
754 case 7:
755 mnem = "sar";
756 break;
757 default:
758 UnimplementedInstruction();
759 return num_bytes;
760 }
Steve Blockd0582a62009-12-15 09:54:21 +0000761 ASSERT_NE(NULL, mnem);
Steve Blocka7e24c12009-10-30 11:49:00 +0000762 if (op == 0xD0) {
763 imm8 = 1;
764 } else if (op == 0xC0) {
765 imm8 = *(data + 2);
766 num_bytes = 3;
767 }
768 AppendToBuffer("%s%c %s,",
769 mnem,
770 operand_size_code(),
771 byte_size_operand_ ? NameOfByteCPURegister(rm)
772 : NameOfCPURegister(rm));
773 if (op == 0xD2) {
774 AppendToBuffer("cl");
775 } else {
776 AppendToBuffer("%d", imm8);
777 }
778 return num_bytes;
779}
780
781
782// Returns number of bytes used, including *data.
783int DisassemblerX64::JumpShort(byte* data) {
Steve Blockd0582a62009-12-15 09:54:21 +0000784 ASSERT_EQ(0xEB, *data);
Steve Blocka7e24c12009-10-30 11:49:00 +0000785 byte b = *(data + 1);
786 byte* dest = data + static_cast<int8_t>(b) + 2;
787 AppendToBuffer("jmp %s", NameOfAddress(dest));
788 return 2;
789}
790
791
792// Returns number of bytes used, including *data.
793int DisassemblerX64::JumpConditional(byte* data) {
Steve Blockd0582a62009-12-15 09:54:21 +0000794 ASSERT_EQ(0x0F, *data);
Steve Blocka7e24c12009-10-30 11:49:00 +0000795 byte cond = *(data + 1) & 0x0F;
796 byte* dest = data + *reinterpret_cast<int32_t*>(data + 2) + 6;
797 const char* mnem = conditional_code_suffix[cond];
798 AppendToBuffer("j%s %s", mnem, NameOfAddress(dest));
799 return 6; // includes 0x0F
800}
801
802
803// Returns number of bytes used, including *data.
804int DisassemblerX64::JumpConditionalShort(byte* data) {
805 byte cond = *data & 0x0F;
806 byte b = *(data + 1);
807 byte* dest = data + static_cast<int8_t>(b) + 2;
808 const char* mnem = conditional_code_suffix[cond];
809 AppendToBuffer("j%s %s", mnem, NameOfAddress(dest));
810 return 2;
811}
812
813
814// Returns number of bytes used, including *data.
815int DisassemblerX64::SetCC(byte* data) {
Steve Blockd0582a62009-12-15 09:54:21 +0000816 ASSERT_EQ(0x0F, *data);
Steve Blocka7e24c12009-10-30 11:49:00 +0000817 byte cond = *(data + 1) & 0x0F;
818 const char* mnem = conditional_code_suffix[cond];
819 AppendToBuffer("set%s%c ", mnem, operand_size_code());
820 PrintRightByteOperand(data + 2);
821 return 3; // includes 0x0F
822}
823
824
825// Returns number of bytes used, including *data.
826int DisassemblerX64::FPUInstruction(byte* data) {
Steve Blockd0582a62009-12-15 09:54:21 +0000827 byte escape_opcode = *data;
828 ASSERT_EQ(0xD8, escape_opcode & 0xF8);
829 byte modrm_byte = *(data+1);
830
831 if (modrm_byte >= 0xC0) {
832 return RegisterFPUInstruction(escape_opcode, modrm_byte);
833 } else {
834 return MemoryFPUInstruction(escape_opcode, modrm_byte, data+1);
Steve Blocka7e24c12009-10-30 11:49:00 +0000835 }
Steve Blockd0582a62009-12-15 09:54:21 +0000836}
837
838int DisassemblerX64::MemoryFPUInstruction(int escape_opcode,
839 int modrm_byte,
840 byte* modrm_start) {
841 const char* mnem = "?";
842 int regop = (modrm_byte >> 3) & 0x7; // reg/op field of modrm byte.
843 switch (escape_opcode) {
844 case 0xD9: switch (regop) {
845 case 0: mnem = "fld_s"; break;
846 case 3: mnem = "fstp_s"; break;
847 case 7: mnem = "fstcw"; break;
848 default: UnimplementedInstruction();
849 }
850 break;
851
852 case 0xDB: switch (regop) {
853 case 0: mnem = "fild_s"; break;
854 case 1: mnem = "fisttp_s"; break;
855 case 2: mnem = "fist_s"; break;
856 case 3: mnem = "fistp_s"; break;
857 default: UnimplementedInstruction();
858 }
859 break;
860
861 case 0xDD: switch (regop) {
862 case 0: mnem = "fld_d"; break;
863 case 3: mnem = "fstp_d"; break;
864 default: UnimplementedInstruction();
865 }
866 break;
867
868 case 0xDF: switch (regop) {
869 case 5: mnem = "fild_d"; break;
870 case 7: mnem = "fistp_d"; break;
871 default: UnimplementedInstruction();
872 }
873 break;
874
875 default: UnimplementedInstruction();
876 }
877 AppendToBuffer("%s ", mnem);
878 int count = PrintRightOperand(modrm_start);
879 return count + 1;
880}
881
882int DisassemblerX64::RegisterFPUInstruction(int escape_opcode,
883 byte modrm_byte) {
884 bool has_register = false; // Is the FPU register encoded in modrm_byte?
885 const char* mnem = "?";
886
887 switch (escape_opcode) {
888 case 0xD8:
889 UnimplementedInstruction();
890 break;
891
892 case 0xD9:
893 switch (modrm_byte & 0xF8) {
Kristian Monsen0d5e1162010-09-30 15:31:59 +0100894 case 0xC0:
895 mnem = "fld";
896 has_register = true;
897 break;
Steve Blockd0582a62009-12-15 09:54:21 +0000898 case 0xC8:
899 mnem = "fxch";
900 has_register = true;
901 break;
902 default:
903 switch (modrm_byte) {
904 case 0xE0: mnem = "fchs"; break;
905 case 0xE1: mnem = "fabs"; break;
906 case 0xE4: mnem = "ftst"; break;
907 case 0xE8: mnem = "fld1"; break;
Kristian Monsen0d5e1162010-09-30 15:31:59 +0100908 case 0xEB: mnem = "fldpi"; break;
Steve Blockd0582a62009-12-15 09:54:21 +0000909 case 0xEE: mnem = "fldz"; break;
910 case 0xF5: mnem = "fprem1"; break;
911 case 0xF7: mnem = "fincstp"; break;
912 case 0xF8: mnem = "fprem"; break;
913 case 0xFE: mnem = "fsin"; break;
914 case 0xFF: mnem = "fcos"; break;
915 default: UnimplementedInstruction();
916 }
917 }
918 break;
919
920 case 0xDA:
921 if (modrm_byte == 0xE9) {
922 mnem = "fucompp";
923 } else {
924 UnimplementedInstruction();
925 }
926 break;
927
928 case 0xDB:
929 if ((modrm_byte & 0xF8) == 0xE8) {
930 mnem = "fucomi";
931 has_register = true;
932 } else if (modrm_byte == 0xE2) {
933 mnem = "fclex";
934 } else {
935 UnimplementedInstruction();
936 }
937 break;
938
939 case 0xDC:
940 has_register = true;
941 switch (modrm_byte & 0xF8) {
942 case 0xC0: mnem = "fadd"; break;
943 case 0xE8: mnem = "fsub"; break;
944 case 0xC8: mnem = "fmul"; break;
945 case 0xF8: mnem = "fdiv"; break;
946 default: UnimplementedInstruction();
947 }
948 break;
949
950 case 0xDD:
951 has_register = true;
952 switch (modrm_byte & 0xF8) {
953 case 0xC0: mnem = "ffree"; break;
954 case 0xD8: mnem = "fstp"; break;
955 default: UnimplementedInstruction();
956 }
957 break;
958
959 case 0xDE:
960 if (modrm_byte == 0xD9) {
961 mnem = "fcompp";
962 } else {
963 has_register = true;
964 switch (modrm_byte & 0xF8) {
965 case 0xC0: mnem = "faddp"; break;
966 case 0xE8: mnem = "fsubp"; break;
967 case 0xC8: mnem = "fmulp"; break;
968 case 0xF8: mnem = "fdivp"; break;
969 default: UnimplementedInstruction();
970 }
971 }
972 break;
973
974 case 0xDF:
975 if (modrm_byte == 0xE0) {
976 mnem = "fnstsw_ax";
977 } else if ((modrm_byte & 0xF8) == 0xE8) {
978 mnem = "fucomip";
979 has_register = true;
980 }
981 break;
982
983 default: UnimplementedInstruction();
984 }
985
986 if (has_register) {
987 AppendToBuffer("%s st%d", mnem, modrm_byte & 0x7);
988 } else {
989 AppendToBuffer("%s", mnem);
990 }
Steve Blocka7e24c12009-10-30 11:49:00 +0000991 return 2;
992}
993
994
Steve Blockd0582a62009-12-15 09:54:21 +0000995
Steve Blocka7e24c12009-10-30 11:49:00 +0000996// Handle all two-byte opcodes, which start with 0x0F.
997// These instructions may be affected by an 0x66, 0xF2, or 0xF3 prefix.
998// We do not use any three-byte opcodes, which start with 0x0F38 or 0x0F3A.
999int DisassemblerX64::TwoByteOpcodeInstruction(byte* data) {
1000 byte opcode = *(data + 1);
1001 byte* current = data + 2;
1002 // At return, "current" points to the start of the next instruction.
1003 const char* mnemonic = TwoByteMnemonic(opcode);
Andrei Popescu402d9372010-02-26 13:31:12 +00001004 if (operand_size_ == 0x66) {
1005 // 0x66 0x0F prefix.
Steve Blocka7e24c12009-10-30 11:49:00 +00001006 int mod, regop, rm;
Steve Block6ded16b2010-05-10 14:33:55 +01001007 if (opcode == 0x3A) {
1008 byte third_byte = *current;
1009 current = data + 3;
1010 if (third_byte == 0x17) {
1011 get_modrm(*current, &mod, &regop, &rm);
1012 AppendToBuffer("extractps "); // reg/m32, xmm, imm8
1013 current += PrintRightOperand(current);
1014 AppendToBuffer(", %s, %d", NameOfCPURegister(regop), (*current) & 3);
1015 current += 1;
1016 } else {
1017 UnimplementedInstruction();
1018 }
Steve Blocka7e24c12009-10-30 11:49:00 +00001019 } else {
Steve Block6ded16b2010-05-10 14:33:55 +01001020 get_modrm(*current, &mod, &regop, &rm);
1021 if (opcode == 0x6E) {
1022 AppendToBuffer("mov%c %s,",
1023 rex_w() ? 'q' : 'd',
1024 NameOfXMMRegister(regop));
1025 current += PrintRightOperand(current);
1026 } else if (opcode == 0x7E) {
Ben Murdochbb769b22010-08-11 14:56:33 +01001027 AppendToBuffer("mov%c ",
1028 rex_w() ? 'q' : 'd');
1029 current += PrintRightOperand(current);
1030 AppendToBuffer(", %s", NameOfXMMRegister(regop));
Steve Block6ded16b2010-05-10 14:33:55 +01001031 } else {
1032 const char* mnemonic = "?";
1033 if (opcode == 0x57) {
1034 mnemonic = "xorpd";
1035 } else if (opcode == 0x2E) {
Steve Block6ded16b2010-05-10 14:33:55 +01001036 mnemonic = "ucomisd";
Steve Block8defd9f2010-07-08 12:39:36 +01001037 } else if (opcode == 0x2F) {
1038 mnemonic = "comisd";
Steve Block6ded16b2010-05-10 14:33:55 +01001039 } else {
1040 UnimplementedInstruction();
1041 }
1042 AppendToBuffer("%s %s,", mnemonic, NameOfXMMRegister(regop));
1043 current += PrintRightXMMOperand(current);
1044 }
Steve Blocka7e24c12009-10-30 11:49:00 +00001045 }
1046 } else if (group_1_prefix_ == 0xF2) {
1047 // Beginning of instructions with prefix 0xF2.
1048
1049 if (opcode == 0x11 || opcode == 0x10) {
1050 // MOVSD: Move scalar double-precision fp to/from/between XMM registers.
1051 AppendToBuffer("movsd ");
1052 int mod, regop, rm;
1053 get_modrm(*current, &mod, &regop, &rm);
1054 if (opcode == 0x11) {
1055 current += PrintRightOperand(current);
1056 AppendToBuffer(",%s", NameOfXMMRegister(regop));
1057 } else {
1058 AppendToBuffer("%s,", NameOfXMMRegister(regop));
1059 current += PrintRightOperand(current);
1060 }
1061 } else if (opcode == 0x2A) {
1062 // CVTSI2SD: integer to XMM double conversion.
1063 int mod, regop, rm;
1064 get_modrm(*current, &mod, &regop, &rm);
Steve Block8defd9f2010-07-08 12:39:36 +01001065 AppendToBuffer("%sd %s,", mnemonic, NameOfXMMRegister(regop));
Steve Blockd0582a62009-12-15 09:54:21 +00001066 current += PrintRightOperand(current);
Kristian Monsen0d5e1162010-09-30 15:31:59 +01001067 } else if (opcode == 0x2C) {
1068 // CVTTSD2SI:
1069 // Convert with truncation scalar double-precision FP to integer.
1070 int mod, regop, rm;
1071 get_modrm(*current, &mod, &regop, &rm);
1072 AppendToBuffer("cvttsd2si%c %s,",
1073 operand_size_code(), NameOfCPURegister(regop));
1074 current += PrintRightXMMOperand(current);
1075 } else if (opcode == 0x2D) {
1076 // CVTSD2SI: Convert scalar double-precision FP to integer.
1077 int mod, regop, rm;
1078 get_modrm(*current, &mod, &regop, &rm);
1079 AppendToBuffer("cvtsd2si%c %s,",
1080 operand_size_code(), NameOfCPURegister(regop));
1081 current += PrintRightXMMOperand(current);
Steve Block6ded16b2010-05-10 14:33:55 +01001082 } else if ((opcode & 0xF8) == 0x58 || opcode == 0x51) {
Steve Blocka7e24c12009-10-30 11:49:00 +00001083 // XMM arithmetic. Mnemonic was retrieved at the start of this function.
1084 int mod, regop, rm;
1085 get_modrm(*current, &mod, &regop, &rm);
Steve Blockd0582a62009-12-15 09:54:21 +00001086 AppendToBuffer("%s %s,", mnemonic, NameOfXMMRegister(regop));
1087 current += PrintRightXMMOperand(current);
Steve Blocka7e24c12009-10-30 11:49:00 +00001088 } else {
1089 UnimplementedInstruction();
1090 }
Steve Block6ded16b2010-05-10 14:33:55 +01001091 } else if (group_1_prefix_ == 0xF3) {
1092 // Instructions with prefix 0xF3.
Steve Block8defd9f2010-07-08 12:39:36 +01001093 if (opcode == 0x11 || opcode == 0x10) {
1094 // MOVSS: Move scalar double-precision fp to/from/between XMM registers.
1095 AppendToBuffer("movss ");
1096 int mod, regop, rm;
1097 get_modrm(*current, &mod, &regop, &rm);
1098 if (opcode == 0x11) {
1099 current += PrintRightOperand(current);
1100 AppendToBuffer(",%s", NameOfXMMRegister(regop));
1101 } else {
1102 AppendToBuffer("%s,", NameOfXMMRegister(regop));
1103 current += PrintRightOperand(current);
1104 }
1105 } else if (opcode == 0x2A) {
1106 // CVTSI2SS: integer to XMM single conversion.
1107 int mod, regop, rm;
1108 get_modrm(*current, &mod, &regop, &rm);
1109 AppendToBuffer("%ss %s,", mnemonic, NameOfXMMRegister(regop));
1110 current += PrintRightOperand(current);
1111 } else if (opcode == 0x2C) {
Kristian Monsen0d5e1162010-09-30 15:31:59 +01001112 // CVTTSS2SI:
1113 // Convert with truncation scalar single-precision FP to dword integer.
Steve Block6ded16b2010-05-10 14:33:55 +01001114 // Assert that mod is not 3, so source is memory, not an XMM register.
1115 ASSERT_NE(0xC0, *current & 0xC0);
1116 current += PrintOperands("cvttss2si", REG_OPER_OP_ORDER, current);
1117 } else if (opcode == 0x5A) {
Kristian Monsen0d5e1162010-09-30 15:31:59 +01001118 // CVTSS2SD:
1119 // Convert scalar single-precision FP to scalar double-precision FP.
Steve Block6ded16b2010-05-10 14:33:55 +01001120 int mod, regop, rm;
1121 get_modrm(*current, &mod, &regop, &rm);
1122 AppendToBuffer("cvtss2sd %s,", NameOfXMMRegister(regop));
1123 current += PrintRightXMMOperand(current);
1124 } else {
1125 UnimplementedInstruction();
1126 }
Andrei Popescu402d9372010-02-26 13:31:12 +00001127 } else if (opcode == 0x1F) {
1128 // NOP
1129 int mod, regop, rm;
1130 get_modrm(*current, &mod, &regop, &rm);
1131 current++;
1132 if (regop == 4) { // SIB byte present.
1133 current++;
1134 }
1135 if (mod == 1) { // Byte displacement.
1136 current += 1;
1137 } else if (mod == 2) { // 32-bit displacement.
1138 current += 4;
1139 } // else no immediate displacement.
1140 AppendToBuffer("nop");
1141 } else if (opcode == 0xA2 || opcode == 0x31) {
1142 // RDTSC or CPUID
1143 AppendToBuffer("%s", mnemonic);
1144
1145 } else if ((opcode & 0xF0) == 0x40) {
1146 // CMOVcc: conditional move.
1147 int condition = opcode & 0x0F;
1148 const InstructionDesc& idesc = cmov_instructions[condition];
1149 byte_size_operand_ = idesc.byte_size_operation;
1150 current += PrintOperands(idesc.mnem, idesc.op_order_, current);
1151
1152 } else if ((opcode & 0xF0) == 0x80) {
1153 // Jcc: Conditional jump (branch).
1154 current = data + JumpConditional(data);
1155
1156 } else if (opcode == 0xBE || opcode == 0xBF || opcode == 0xB6 ||
1157 opcode == 0xB7 || opcode == 0xAF) {
1158 // Size-extending moves, IMUL.
1159 current += PrintOperands(mnemonic, REG_OPER_OP_ORDER, current);
1160
1161 } else if ((opcode & 0xF0) == 0x90) {
1162 // SETcc: Set byte on condition. Needs pointer to beginning of instruction.
1163 current = data + SetCC(data);
1164
1165 } else if (opcode == 0xAB || opcode == 0xA5 || opcode == 0xAD) {
1166 // SHLD, SHRD (double-precision shift), BTS (bit set).
1167 AppendToBuffer("%s ", mnemonic);
1168 int mod, regop, rm;
1169 get_modrm(*current, &mod, &regop, &rm);
1170 current += PrintRightOperand(current);
1171 if (opcode == 0xAB) {
1172 AppendToBuffer(",%s", NameOfCPURegister(regop));
1173 } else {
1174 AppendToBuffer(",%s,cl", NameOfCPURegister(regop));
1175 }
Steve Blocka7e24c12009-10-30 11:49:00 +00001176 } else {
1177 UnimplementedInstruction();
1178 }
Steve Blockd0582a62009-12-15 09:54:21 +00001179 return static_cast<int>(current - data);
Steve Blocka7e24c12009-10-30 11:49:00 +00001180}
1181
1182
1183// Mnemonics for two-byte opcode instructions starting with 0x0F.
1184// The argument is the second byte of the two-byte opcode.
1185// Returns NULL if the instruction is not handled here.
1186const char* DisassemblerX64::TwoByteMnemonic(byte opcode) {
1187 switch (opcode) {
1188 case 0x1F:
1189 return "nop";
Steve Block8defd9f2010-07-08 12:39:36 +01001190 case 0x2A: // F2/F3 prefix.
1191 return "cvtsi2s";
Steve Blocka7e24c12009-10-30 11:49:00 +00001192 case 0x31:
1193 return "rdtsc";
Steve Block6ded16b2010-05-10 14:33:55 +01001194 case 0x51: // F2 prefix.
1195 return "sqrtsd";
Steve Blocka7e24c12009-10-30 11:49:00 +00001196 case 0x58: // F2 prefix.
1197 return "addsd";
1198 case 0x59: // F2 prefix.
1199 return "mulsd";
1200 case 0x5C: // F2 prefix.
1201 return "subsd";
1202 case 0x5E: // F2 prefix.
1203 return "divsd";
1204 case 0xA2:
1205 return "cpuid";
1206 case 0xA5:
1207 return "shld";
1208 case 0xAB:
1209 return "bts";
1210 case 0xAD:
1211 return "shrd";
1212 case 0xAF:
1213 return "imul";
1214 case 0xB6:
1215 return "movzxb";
1216 case 0xB7:
1217 return "movzxw";
1218 case 0xBE:
1219 return "movsxb";
1220 case 0xBF:
1221 return "movsxw";
1222 default:
1223 return NULL;
1224 }
1225}
1226
1227
1228// Disassembles the instruction at instr, and writes it into out_buffer.
1229int DisassemblerX64::InstructionDecode(v8::internal::Vector<char> out_buffer,
1230 byte* instr) {
1231 tmp_buffer_pos_ = 0; // starting to write as position 0
1232 byte* data = instr;
1233 bool processed = true; // Will be set to false if the current instruction
1234 // is not in 'instructions' table.
1235 byte current;
1236
1237 // Scan for prefixes.
1238 while (true) {
1239 current = *data;
Leon Clarked91b9f72010-01-27 17:25:45 +00001240 if (current == OPERAND_SIZE_OVERRIDE_PREFIX) { // Group 3 prefix.
Steve Blocka7e24c12009-10-30 11:49:00 +00001241 operand_size_ = current;
1242 } else if ((current & 0xF0) == 0x40) { // REX prefix.
1243 setRex(current);
1244 if (rex_w()) AppendToBuffer("REX.W ");
Leon Clarked91b9f72010-01-27 17:25:45 +00001245 } else if ((current & 0xFE) == 0xF2) { // Group 1 prefix (0xF2 or 0xF3).
Steve Blocka7e24c12009-10-30 11:49:00 +00001246 group_1_prefix_ = current;
1247 } else { // Not a prefix - an opcode.
1248 break;
1249 }
1250 data++;
1251 }
1252
1253 const InstructionDesc& idesc = instruction_table.Get(current);
1254 byte_size_operand_ = idesc.byte_size_operation;
1255 switch (idesc.type) {
1256 case ZERO_OPERANDS_INSTR:
Leon Clarked91b9f72010-01-27 17:25:45 +00001257 if (current >= 0xA4 && current <= 0xA7) {
1258 // String move or compare operations.
1259 if (group_1_prefix_ == REP_PREFIX) {
1260 // REP.
1261 AppendToBuffer("rep ");
1262 }
1263 if (rex_w()) AppendToBuffer("REX.W ");
1264 AppendToBuffer("%s%c", idesc.mnem, operand_size_code());
1265 } else {
1266 AppendToBuffer("%s", idesc.mnem, operand_size_code());
1267 }
Steve Blocka7e24c12009-10-30 11:49:00 +00001268 data++;
1269 break;
1270
1271 case TWO_OPERANDS_INSTR:
1272 data++;
1273 data += PrintOperands(idesc.mnem, idesc.op_order_, data);
1274 break;
1275
1276 case JUMP_CONDITIONAL_SHORT_INSTR:
1277 data += JumpConditionalShort(data);
1278 break;
1279
1280 case REGISTER_INSTR:
1281 AppendToBuffer("%s%c %s",
1282 idesc.mnem,
1283 operand_size_code(),
1284 NameOfCPURegister(base_reg(current & 0x07)));
1285 data++;
1286 break;
1287 case PUSHPOP_INSTR:
1288 AppendToBuffer("%s %s",
1289 idesc.mnem,
1290 NameOfCPURegister(base_reg(current & 0x07)));
1291 data++;
1292 break;
1293 case MOVE_REG_INSTR: {
1294 byte* addr = NULL;
1295 switch (operand_size()) {
1296 case WORD_SIZE:
1297 addr = reinterpret_cast<byte*>(*reinterpret_cast<int16_t*>(data + 1));
1298 data += 3;
1299 break;
1300 case DOUBLEWORD_SIZE:
1301 addr = reinterpret_cast<byte*>(*reinterpret_cast<int32_t*>(data + 1));
1302 data += 5;
1303 break;
1304 case QUADWORD_SIZE:
1305 addr = reinterpret_cast<byte*>(*reinterpret_cast<int64_t*>(data + 1));
1306 data += 9;
1307 break;
1308 default:
1309 UNREACHABLE();
1310 }
1311 AppendToBuffer("mov%c %s,%s",
1312 operand_size_code(),
1313 NameOfCPURegister(base_reg(current & 0x07)),
1314 NameOfAddress(addr));
1315 break;
1316 }
1317
1318 case CALL_JUMP_INSTR: {
1319 byte* addr = data + *reinterpret_cast<int32_t*>(data + 1) + 5;
1320 AppendToBuffer("%s %s", idesc.mnem, NameOfAddress(addr));
1321 data += 5;
1322 break;
1323 }
1324
1325 case SHORT_IMMEDIATE_INSTR: {
1326 byte* addr =
1327 reinterpret_cast<byte*>(*reinterpret_cast<int32_t*>(data + 1));
1328 AppendToBuffer("%s rax, %s", idesc.mnem, NameOfAddress(addr));
1329 data += 5;
1330 break;
1331 }
1332
1333 case NO_INSTR:
1334 processed = false;
1335 break;
1336
1337 default:
1338 UNIMPLEMENTED(); // This type is not implemented.
1339 }
1340
1341 // The first byte didn't match any of the simple opcodes, so we
1342 // need to do special processing on it.
1343 if (!processed) {
1344 switch (*data) {
1345 case 0xC2:
1346 AppendToBuffer("ret 0x%x", *reinterpret_cast<uint16_t*>(data + 1));
1347 data += 3;
1348 break;
1349
1350 case 0x69: // fall through
1351 case 0x6B: {
1352 int mod, regop, rm;
1353 get_modrm(*(data + 1), &mod, &regop, &rm);
1354 int32_t imm = *data == 0x6B ? *(data + 2)
1355 : *reinterpret_cast<int32_t*>(data + 2);
Steve Block6ded16b2010-05-10 14:33:55 +01001356 AppendToBuffer("imul%c %s,%s,0x%x",
1357 operand_size_code(),
1358 NameOfCPURegister(regop),
Steve Blocka7e24c12009-10-30 11:49:00 +00001359 NameOfCPURegister(rm), imm);
1360 data += 2 + (*data == 0x6B ? 1 : 4);
1361 break;
1362 }
1363
Steve Blocka7e24c12009-10-30 11:49:00 +00001364 case 0x81: // fall through
1365 case 0x83: // 0x81 with sign extension bit set
1366 data += PrintImmediateOp(data);
1367 break;
1368
1369 case 0x0F:
1370 data += TwoByteOpcodeInstruction(data);
1371 break;
1372
1373 case 0x8F: {
1374 data++;
1375 int mod, regop, rm;
1376 get_modrm(*data, &mod, &regop, &rm);
1377 if (regop == 0) {
1378 AppendToBuffer("pop ");
1379 data += PrintRightOperand(data);
1380 }
1381 }
1382 break;
1383
1384 case 0xFF: {
1385 data++;
1386 int mod, regop, rm;
1387 get_modrm(*data, &mod, &regop, &rm);
1388 const char* mnem = NULL;
1389 switch (regop) {
1390 case 0:
1391 mnem = "inc";
1392 break;
1393 case 1:
1394 mnem = "dec";
1395 break;
1396 case 2:
1397 mnem = "call";
1398 break;
1399 case 4:
1400 mnem = "jmp";
1401 break;
1402 case 6:
1403 mnem = "push";
1404 break;
1405 default:
1406 mnem = "???";
1407 }
1408 AppendToBuffer(((regop <= 1) ? "%s%c " : "%s "),
1409 mnem,
1410 operand_size_code());
1411 data += PrintRightOperand(data);
1412 }
1413 break;
1414
1415 case 0xC7: // imm32, fall through
1416 case 0xC6: // imm8
1417 {
1418 bool is_byte = *data == 0xC6;
1419 data++;
1420
1421 AppendToBuffer("mov%c ", is_byte ? 'b' : operand_size_code());
1422 data += PrintRightOperand(data);
1423 int32_t imm = is_byte ? *data : *reinterpret_cast<int32_t*>(data);
1424 AppendToBuffer(",0x%x", imm);
1425 data += is_byte ? 1 : 4;
1426 }
1427 break;
1428
1429 case 0x80: {
1430 data++;
1431 AppendToBuffer("cmpb ");
1432 data += PrintRightOperand(data);
1433 int32_t imm = *data;
1434 AppendToBuffer(",0x%x", imm);
1435 data++;
1436 }
1437 break;
1438
1439 case 0x88: // 8bit, fall through
1440 case 0x89: // 32bit
1441 {
1442 bool is_byte = *data == 0x88;
1443 int mod, regop, rm;
1444 data++;
1445 get_modrm(*data, &mod, &regop, &rm);
1446 AppendToBuffer("mov%c ", is_byte ? 'b' : operand_size_code());
1447 data += PrintRightOperand(data);
1448 AppendToBuffer(",%s", NameOfCPURegister(regop));
1449 }
1450 break;
1451
1452 case 0x90:
1453 case 0x91:
1454 case 0x92:
1455 case 0x93:
1456 case 0x94:
1457 case 0x95:
1458 case 0x96:
1459 case 0x97: {
Steve Blockd0582a62009-12-15 09:54:21 +00001460 int reg = (*data & 0x7) | (rex_b() ? 8 : 0);
Steve Blocka7e24c12009-10-30 11:49:00 +00001461 if (reg == 0) {
1462 AppendToBuffer("nop"); // Common name for xchg rax,rax.
1463 } else {
1464 AppendToBuffer("xchg%c rax, %s",
1465 operand_size_code(),
1466 NameOfCPURegister(reg));
1467 }
Steve Blockd0582a62009-12-15 09:54:21 +00001468 data++;
Steve Blocka7e24c12009-10-30 11:49:00 +00001469 }
Steve Blockd0582a62009-12-15 09:54:21 +00001470 break;
Steve Blocka7e24c12009-10-30 11:49:00 +00001471
1472 case 0xFE: {
1473 data++;
1474 int mod, regop, rm;
1475 get_modrm(*data, &mod, &regop, &rm);
Kristian Monsen0d5e1162010-09-30 15:31:59 +01001476 if (regop == 1) {
1477 AppendToBuffer("decb ");
1478 data += PrintRightOperand(data);
Steve Blocka7e24c12009-10-30 11:49:00 +00001479 } else {
1480 UnimplementedInstruction();
1481 }
Steve Blocka7e24c12009-10-30 11:49:00 +00001482 }
1483 break;
1484
1485 case 0x68:
1486 AppendToBuffer("push 0x%x", *reinterpret_cast<int32_t*>(data + 1));
1487 data += 5;
1488 break;
1489
1490 case 0x6A:
1491 AppendToBuffer("push 0x%x", *reinterpret_cast<int8_t*>(data + 1));
1492 data += 2;
1493 break;
1494
1495 case 0xA1: // Fall through.
1496 case 0xA3:
1497 switch (operand_size()) {
1498 case DOUBLEWORD_SIZE: {
1499 const char* memory_location = NameOfAddress(
1500 reinterpret_cast<byte*>(
1501 *reinterpret_cast<int32_t*>(data + 1)));
1502 if (*data == 0xA1) { // Opcode 0xA1
1503 AppendToBuffer("movzxlq rax,(%s)", memory_location);
1504 } else { // Opcode 0xA3
1505 AppendToBuffer("movzxlq (%s),rax", memory_location);
1506 }
1507 data += 5;
1508 break;
1509 }
1510 case QUADWORD_SIZE: {
1511 // New x64 instruction mov rax,(imm_64).
1512 const char* memory_location = NameOfAddress(
1513 *reinterpret_cast<byte**>(data + 1));
1514 if (*data == 0xA1) { // Opcode 0xA1
1515 AppendToBuffer("movq rax,(%s)", memory_location);
1516 } else { // Opcode 0xA3
1517 AppendToBuffer("movq (%s),rax", memory_location);
1518 }
1519 data += 9;
1520 break;
1521 }
1522 default:
1523 UnimplementedInstruction();
1524 data += 2;
1525 }
1526 break;
1527
1528 case 0xA8:
1529 AppendToBuffer("test al,0x%x", *reinterpret_cast<uint8_t*>(data + 1));
1530 data += 2;
1531 break;
1532
1533 case 0xA9: {
1534 int64_t value = 0;
1535 switch (operand_size()) {
1536 case WORD_SIZE:
1537 value = *reinterpret_cast<uint16_t*>(data + 1);
1538 data += 3;
1539 break;
1540 case DOUBLEWORD_SIZE:
1541 value = *reinterpret_cast<uint32_t*>(data + 1);
1542 data += 5;
1543 break;
1544 case QUADWORD_SIZE:
1545 value = *reinterpret_cast<int32_t*>(data + 1);
1546 data += 5;
1547 break;
1548 default:
1549 UNREACHABLE();
1550 }
1551 AppendToBuffer("test%c rax,0x%"V8_PTR_PREFIX"x",
1552 operand_size_code(),
1553 value);
1554 break;
1555 }
1556 case 0xD1: // fall through
1557 case 0xD3: // fall through
1558 case 0xC1:
1559 data += ShiftInstruction(data);
1560 break;
1561 case 0xD0: // fall through
1562 case 0xD2: // fall through
1563 case 0xC0:
1564 byte_size_operand_ = true;
1565 data += ShiftInstruction(data);
1566 break;
1567
1568 case 0xD9: // fall through
1569 case 0xDA: // fall through
1570 case 0xDB: // fall through
1571 case 0xDC: // fall through
1572 case 0xDD: // fall through
1573 case 0xDE: // fall through
1574 case 0xDF:
1575 data += FPUInstruction(data);
1576 break;
1577
1578 case 0xEB:
1579 data += JumpShort(data);
1580 break;
1581
Steve Blockd0582a62009-12-15 09:54:21 +00001582 case 0xF6:
1583 byte_size_operand_ = true; // fall through
Steve Blocka7e24c12009-10-30 11:49:00 +00001584 case 0xF7:
Steve Blockd0582a62009-12-15 09:54:21 +00001585 data += F6F7Instruction(data);
Steve Blocka7e24c12009-10-30 11:49:00 +00001586 break;
1587
1588 default:
1589 UnimplementedInstruction();
1590 data += 1;
1591 }
1592 } // !processed
1593
1594 if (tmp_buffer_pos_ < sizeof tmp_buffer_) {
1595 tmp_buffer_[tmp_buffer_pos_] = '\0';
1596 }
1597
Steve Blockd0582a62009-12-15 09:54:21 +00001598 int instr_len = static_cast<int>(data - instr);
Steve Blocka7e24c12009-10-30 11:49:00 +00001599 ASSERT(instr_len > 0); // Ensure progress.
1600
1601 int outp = 0;
1602 // Instruction bytes.
1603 for (byte* bp = instr; bp < data; bp++) {
1604 outp += v8::internal::OS::SNPrintF(out_buffer + outp, "%02x", *bp);
1605 }
1606 for (int i = 6 - instr_len; i >= 0; i--) {
1607 outp += v8::internal::OS::SNPrintF(out_buffer + outp, " ");
1608 }
1609
1610 outp += v8::internal::OS::SNPrintF(out_buffer + outp, " %s",
1611 tmp_buffer_.start());
1612 return instr_len;
1613}
1614
1615//------------------------------------------------------------------------------
1616
1617
1618static const char* cpu_regs[16] = {
1619 "rax", "rcx", "rdx", "rbx", "rsp", "rbp", "rsi", "rdi",
1620 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
1621};
1622
1623
1624static const char* byte_cpu_regs[16] = {
1625 "al", "cl", "dl", "bl", "spl", "bpl", "sil", "dil",
1626 "r8l", "r9l", "r10l", "r11l", "r12l", "r13l", "r14l", "r15l"
1627};
1628
1629
1630static const char* xmm_regs[16] = {
1631 "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
1632 "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15"
1633};
1634
1635
1636const char* NameConverter::NameOfAddress(byte* addr) const {
1637 static v8::internal::EmbeddedVector<char, 32> tmp_buffer;
1638 v8::internal::OS::SNPrintF(tmp_buffer, "%p", addr);
1639 return tmp_buffer.start();
1640}
1641
1642
1643const char* NameConverter::NameOfConstant(byte* addr) const {
1644 return NameOfAddress(addr);
1645}
1646
1647
1648const char* NameConverter::NameOfCPURegister(int reg) const {
1649 if (0 <= reg && reg < 16)
1650 return cpu_regs[reg];
1651 return "noreg";
1652}
1653
1654
1655const char* NameConverter::NameOfByteCPURegister(int reg) const {
1656 if (0 <= reg && reg < 16)
1657 return byte_cpu_regs[reg];
1658 return "noreg";
1659}
1660
1661
1662const char* NameConverter::NameOfXMMRegister(int reg) const {
1663 if (0 <= reg && reg < 16)
1664 return xmm_regs[reg];
1665 return "noxmmreg";
1666}
1667
1668
1669const char* NameConverter::NameInCode(byte* addr) const {
1670 // X64 does not embed debug strings at the moment.
1671 UNREACHABLE();
1672 return "";
1673}
1674
1675//------------------------------------------------------------------------------
1676
1677Disassembler::Disassembler(const NameConverter& converter)
1678 : converter_(converter) { }
1679
1680Disassembler::~Disassembler() { }
1681
1682
1683int Disassembler::InstructionDecode(v8::internal::Vector<char> buffer,
1684 byte* instruction) {
1685 DisassemblerX64 d(converter_, CONTINUE_ON_UNIMPLEMENTED_OPCODE);
1686 return d.InstructionDecode(buffer, instruction);
1687}
1688
1689
1690// The X64 assembler does not use constant pools.
1691int Disassembler::ConstantPoolSizeAt(byte* instruction) {
1692 return -1;
1693}
1694
1695
1696void Disassembler::Disassemble(FILE* f, byte* begin, byte* end) {
1697 NameConverter converter;
1698 Disassembler d(converter);
1699 for (byte* pc = begin; pc < end;) {
1700 v8::internal::EmbeddedVector<char, 128> buffer;
1701 buffer[0] = '\0';
1702 byte* prev_pc = pc;
1703 pc += d.InstructionDecode(buffer, pc);
1704 fprintf(f, "%p", prev_pc);
1705 fprintf(f, " ");
1706
1707 for (byte* bp = prev_pc; bp < pc; bp++) {
1708 fprintf(f, "%02x", *bp);
1709 }
Steve Blockd0582a62009-12-15 09:54:21 +00001710 for (int i = 6 - static_cast<int>(pc - prev_pc); i >= 0; i--) {
Steve Blocka7e24c12009-10-30 11:49:00 +00001711 fprintf(f, " ");
1712 }
1713 fprintf(f, " %s\n", buffer.start());
1714 }
1715}
1716
1717} // namespace disasm
Leon Clarkef7060e22010-06-03 12:02:55 +01001718
1719#endif // V8_TARGET_ARCH_X64