1/*
2 * Copyright (C) 2012 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 *      http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17#include "disassembler_arm.h"
18
19#include <inttypes.h>
20
21#include <ostream>
22#include <sstream>
23
24#include "arch/arm/registers_arm.h"
25#include "base/logging.h"
26#include "base/stringprintf.h"
27#include "thread.h"
28
29namespace art {
30namespace arm {
31
32size_t DisassemblerArm::Dump(std::ostream& os, const uint8_t* begin) {
33  if ((reinterpret_cast<intptr_t>(begin) & 1) == 0) {
34    DumpArm(os, begin);
35    return 4;
36  } else {
37    // remove thumb specifier bits
38    begin = reinterpret_cast<const uint8_t*>(reinterpret_cast<uintptr_t>(begin) & ~1);
39    return DumpThumb16(os, begin);
40  }
41}
42
43void DisassemblerArm::Dump(std::ostream& os, const uint8_t* begin, const uint8_t* end) {
44  if ((reinterpret_cast<intptr_t>(begin) & 1) == 0) {
45    for (const uint8_t* cur = begin; cur < end; cur += 4) {
46      DumpArm(os, cur);
47    }
48  } else {
49    // remove thumb specifier bits
50    begin = reinterpret_cast<const uint8_t*>(reinterpret_cast<uintptr_t>(begin) & ~1);
51    end = reinterpret_cast<const uint8_t*>(reinterpret_cast<uintptr_t>(end) & ~1);
52    for (const uint8_t* cur = begin; cur < end;) {
53      cur += DumpThumb16(os, cur);
54    }
55  }
56}
57
58static const char* kConditionCodeNames[] = {
59  "eq",  // 0000 - equal
60  "ne",  // 0001 - not-equal
61  "cs",  // 0010 - carry-set, greater than, equal or unordered
62  "cc",  // 0011 - carry-clear, less than
63  "mi",  // 0100 - minus, negative
64  "pl",  // 0101 - plus, positive or zero
65  "vs",  // 0110 - overflow
66  "vc",  // 0111 - no overflow
67  "hi",  // 1000 - unsigned higher
68  "ls",  // 1001 - unsigned lower or same
69  "ge",  // 1010 - signed greater than or equal
70  "lt",  // 1011 - signed less than
71  "gt",  // 1100 - signed greater than
72  "le",  // 1101 - signed less than or equal
73  "",    // 1110 - always
74  "nv",  // 1111 - never (mostly obsolete, but might be a clue that we're mistranslating)
75};
76
77void DisassemblerArm::DumpCond(std::ostream& os, uint32_t cond) {
78  if (cond < 15) {
79    os << kConditionCodeNames[cond];
80  } else {
81    os << "Unexpected condition: " << cond;
82  }
83}
84
85void DisassemblerArm::DumpMemoryDomain(std::ostream& os, uint32_t domain) {
86  switch (domain) {
87    case 15U /* 0b1111 */: os << "sy"; break;
88    case 14U /* 0b1110 */: os << "st"; break;
89    case 11U /* 0b1011 */: os << "ish"; break;
90    case 10U /* 0b1010 */: os << "ishst"; break;
91    case  7U /* 0b0111 */: os << "nsh"; break;
92    case  6U /* 0b0110 */: os << "nshst"; break;
93    case  3U /* 0b0011 */: os << "osh"; break;
94    case  2U /* 0b0010 */: os << "oshst"; break;
95  }
96}
97
98void DisassemblerArm::DumpBranchTarget(std::ostream& os, const uint8_t* instr_ptr, int32_t imm32) {
99  os << StringPrintf("%+d (", imm32) << FormatInstructionPointer(instr_ptr + imm32) << ")";
100}
101
102static uint32_t ReadU16(const uint8_t* ptr) {
103  return ptr[0] | (ptr[1] << 8);
104}
105
106static uint32_t ReadU32(const uint8_t* ptr) {
107  return ptr[0] | (ptr[1] << 8) | (ptr[2] << 16) | (ptr[3] << 24);
108}
109
110static const char* kDataProcessingOperations[] = {
111  "and", "eor", "sub", "rsb", "add", "adc", "sbc", "rsc",
112  "tst", "teq", "cmp", "cmn", "orr", "mov", "bic", "mvn",
113};
114
115static const char* kThumbDataProcessingOperations[] = {
116  "and", "eor", "lsl", "lsr", "asr", "adc", "sbc", "ror",
117  "tst", "rsb", "cmp", "cmn", "orr", "mul", "bic", "mvn",
118};
119
120static const char* const kThumb2ShiftOperations[] = {
121    "lsl", "lsr", "asr", "ror"
122};
123
124static const char* kThumbReverseOperations[] = {
125    "rev", "rev16", "rbit", "revsh"
126};
127
128struct ArmRegister {
129  explicit ArmRegister(uint32_t r_in) : r(r_in) { CHECK_LE(r_in, 15U); }
130  ArmRegister(uint32_t instruction, uint32_t at_bit) : r((instruction >> at_bit) & 0xf) {
131    CHECK_LE(r, 15U);
132  }
133  uint32_t r;
134};
135std::ostream& operator<<(std::ostream& os, const ArmRegister& r) {
136  if (r.r == 13) {
137    os << "sp";
138  } else if (r.r == 14) {
139    os << "lr";
140  } else if (r.r == 15) {
141    os << "pc";
142  } else {
143    os << "r" << r.r;
144  }
145  return os;
146}
147
148struct ThumbRegister : ArmRegister {
149  ThumbRegister(uint16_t instruction, uint16_t at_bit) : ArmRegister((instruction >> at_bit) & 0x7) {}
150};
151
152struct RmLslImm2 {
153  explicit RmLslImm2(uint32_t instr) : imm2((instr >> 4) & 0x3), rm(instr & 0xf) {}
154  uint32_t imm2;
155  ArmRegister rm;
156};
157std::ostream& operator<<(std::ostream& os, const RmLslImm2& r) {
158  os << r.rm;
159  if (r.imm2 != 0) {
160    os << ", lsl #" << r.imm2;
161  }
162  return os;
163}
164
165struct ShiftedImmediate {
166  explicit ShiftedImmediate(uint32_t instruction) {
167    uint32_t rotate = ((instruction >> 8) & 0xf);
168    uint32_t imm = (instruction & 0xff);
169    value = (imm >> (2 * rotate)) | (imm << (32 - (2 * rotate)));
170  }
171  uint32_t value;
172};
173std::ostream& operator<<(std::ostream& os, const ShiftedImmediate& rhs) {
174  os << "#" << rhs.value;
175  return os;
176}
177
178struct RegisterList {
179  explicit RegisterList(uint32_t instruction) : register_list(instruction & 0xffff) {}
180  uint32_t register_list;
181};
182std::ostream& operator<<(std::ostream& os, const RegisterList& rhs) {
183  if (rhs.register_list == 0) {
184    os << "<no register list?>";
185    return os;
186  }
187  os << "{";
188  bool first = true;
189  for (size_t i = 0; i < 16; i++) {
190    if ((rhs.register_list & (1 << i)) != 0) {
191      if (first) {
192        first = false;
193      } else {
194        os << ", ";
195      }
196      os << ArmRegister(i);
197    }
198  }
199  os << "}";
200  return os;
201}
202
203struct FpRegister {
204  explicit FpRegister(uint32_t instr, uint16_t at_bit, uint16_t extra_at_bit) {
205    size = (instr >> 8) & 1;
206    uint32_t Vn = (instr >> at_bit) & 0xF;
207    uint32_t N = (instr >> extra_at_bit) & 1;
208    r = (size != 0 ? ((N << 4) | Vn) : ((Vn << 1) | N));
209  }
210  explicit FpRegister(uint32_t instr, uint16_t at_bit, uint16_t extra_at_bit,
211                      uint32_t forced_size) {
212    size = forced_size;
213    uint32_t Vn = (instr >> at_bit) & 0xF;
214    uint32_t N = (instr >> extra_at_bit) & 1;
215    r = (size != 0 ? ((N << 4) | Vn) : ((Vn << 1) | N));
216  }
217  FpRegister(const FpRegister& other, uint32_t offset)
218      : size(other.size), r(other.r + offset) {}
219
220  uint32_t size;  // 0 = f32, 1 = f64
221  uint32_t r;
222};
223std::ostream& operator<<(std::ostream& os, const FpRegister& rhs) {
224  return os << ((rhs.size != 0) ? "d" : "s") << rhs.r;
225}
226
227struct FpRegisterRange {
228  explicit FpRegisterRange(uint32_t instr)
229      : first(instr, 12, 22), imm8(instr & 0xFF) {}
230  FpRegister first;
231  uint32_t imm8;
232};
233std::ostream& operator<<(std::ostream& os, const FpRegisterRange& rhs) {
234  os << "{" << rhs.first;
235  int count = (rhs.first.size != 0 ? ((rhs.imm8 + 1u) >> 1) : rhs.imm8);
236  if (count > 1) {
237    os << "-" << FpRegister(rhs.first, count - 1);
238  }
239  if (rhs.imm8 == 0) {
240    os << " (EMPTY)";
241  } else if (rhs.first.size != 0 && (rhs.imm8 & 1) != 0) {
242    os << rhs.first << " (HALF)";
243  }
244  os << "}";
245  return os;
246}
247
248void DisassemblerArm::DumpArm(std::ostream& os, const uint8_t* instr_ptr) {
249  uint32_t instruction = ReadU32(instr_ptr);
250  uint32_t cond = (instruction >> 28) & 0xf;
251  uint32_t op1 = (instruction >> 25) & 0x7;
252  std::string opcode;
253  std::string suffixes;
254  std::ostringstream args;
255  switch (op1) {
256    case 0:
257    case 1:  // Data processing instructions.
258      {
259        if ((instruction & 0x0ff000f0) == 0x01200070) {  // BKPT
260          opcode = "bkpt";
261          uint32_t imm12 = (instruction >> 8) & 0xfff;
262          uint32_t imm4 = (instruction & 0xf);
263          args << '#' << ((imm12 << 4) | imm4);
264          break;
265        }
266        if ((instruction & 0x0fffffd0) == 0x012fff10) {  // BX and BLX (register)
267          opcode = (((instruction >> 5) & 1) ? "blx" : "bx");
268          args << ArmRegister(instruction & 0xf);
269          break;
270        }
271        bool i = (instruction & (1 << 25)) != 0;
272        bool s = (instruction & (1 << 20)) != 0;
273        uint32_t op = (instruction >> 21) & 0xf;
274        opcode = kDataProcessingOperations[op];
275        bool implicit_s = ((op & ~3) == 8);  // TST, TEQ, CMP, and CMN.
276        bool is_mov = op == 13U /* 0b1101 */ || op == 15U /* 0b1111 */;
277        if (is_mov) {
278          // Show only Rd and Rm.
279          if (s) {
280             suffixes += 's';
281           }
282           args << ArmRegister(instruction, 12) << ", ";
283           if (i) {
284              args << ShiftedImmediate(instruction);
285            } else {
286              // TODO: Shifted register.
287              args << ArmRegister(instruction, 16) << ", " << ArmRegister(instruction, 0);
288            }
289        } else {
290          if (implicit_s) {
291            // Rd is unused (and not shown), and we don't show the 's' suffix either.
292          } else {
293            if (s) {
294              suffixes += 's';
295            }
296            args << ArmRegister(instruction, 12) << ", ";
297          }
298          if (i) {
299            args << ArmRegister(instruction, 16) << ", " << ShiftedImmediate(instruction);
300          } else {
301            // TODO: Shifted register.
302            args << ArmRegister(instruction, 16) << ", " << ArmRegister(instruction, 0);
303          }
304        }
305      }
306      break;
307    case 2:  // Load/store word and unsigned byte.
308      {
309        bool p = (instruction & (1 << 24)) != 0;
310        bool b = (instruction & (1 << 22)) != 0;
311        bool w = (instruction & (1 << 21)) != 0;
312        bool l = (instruction & (1 << 20)) != 0;
313        opcode = StringPrintf("%s%s", (l ? "ldr" : "str"), (b ? "b" : ""));
314        args << ArmRegister(instruction, 12) << ", ";
315        ArmRegister rn(instruction, 16);
316        if (rn.r == 0xf) {
317          UNIMPLEMENTED(FATAL) << "literals";
318        } else {
319          bool wback = !p || w;
320          uint32_t offset = (instruction & 0xfff);
321          if (p && !wback) {
322            args << "[" << rn << ", #" << offset << "]";
323          } else if (p && wback) {
324            args << "[" << rn << ", #" << offset << "]!";
325          } else if (!p && wback) {
326            args << "[" << rn << "], #" << offset;
327          } else {
328            LOG(FATAL) << p << " " << w;
329          }
330          if (rn.r == 9) {
331            args << "  ; ";
332            Thread::DumpThreadOffset<4>(args, offset);
333          }
334        }
335      }
336      break;
337    case 4:  // Load/store multiple.
338      {
339        bool p = (instruction & (1 << 24)) != 0;
340        bool u = (instruction & (1 << 23)) != 0;
341        bool w = (instruction & (1 << 21)) != 0;
342        bool l = (instruction & (1 << 20)) != 0;
343        opcode = StringPrintf("%s%c%c", (l ? "ldm" : "stm"), (u ? 'i' : 'd'), (p ? 'b' : 'a'));
344        args << ArmRegister(instruction, 16) << (w ? "!" : "") << ", " << RegisterList(instruction);
345      }
346      break;
347    case 5:  // Branch/branch with link.
348      {
349        bool bl = (instruction & (1 << 24)) != 0;
350        opcode = (bl ? "bl" : "b");
351        int32_t imm26 = (instruction & 0xffffff) << 2;
352        int32_t imm32 = (imm26 << 6) >> 6;  // Sign extend.
353        DumpBranchTarget(args, instr_ptr + 8, imm32);
354      }
355      break;
356    default:
357      opcode = "???";
358      break;
359    }
360    opcode += kConditionCodeNames[cond];
361    opcode += suffixes;
362    // TODO: a more complete ARM disassembler could generate wider opcodes.
363    os << FormatInstructionPointer(instr_ptr)
364       << StringPrintf(": %08x\t%-7s ", instruction, opcode.c_str())
365       << args.str() << '\n';
366}
367
368int32_t ThumbExpand(int32_t imm12) {
369  if ((imm12 & 0xC00) == 0) {
370    switch ((imm12 >> 8) & 3) {
371      case 0:
372        return imm12 & 0xFF;
373      case 1:
374        return ((imm12 & 0xFF) << 16) | (imm12 & 0xFF);
375      case 2:
376        return ((imm12 & 0xFF) << 24) | ((imm12 & 0xFF) << 8);
377      default:  // 3
378        return ((imm12 & 0xFF) << 24) | ((imm12 & 0xFF) << 16) | ((imm12 & 0xFF) << 8) |
379            (imm12 & 0xFF);
380    }
381  } else {
382    uint32_t val = 0x80 | (imm12 & 0x7F);
383    int32_t rotate = (imm12 >> 7) & 0x1F;
384    return (val >> rotate) | (val << (32 - rotate));
385  }
386}
387
388uint32_t VFPExpand32(uint32_t imm8) {
389  CHECK_EQ(imm8 & 0xffu, imm8);
390  uint32_t bit_a = (imm8 >> 7) & 1;
391  uint32_t bit_b = (imm8 >> 6) & 1;
392  uint32_t slice = imm8 & 0x3f;
393  return (bit_a << 31) | ((1 << 30) - (bit_b << 25)) | (slice << 19);
394}
395
396static uint64_t VFPExpand64(uint32_t imm8) {
397  CHECK_EQ(imm8 & 0xffu, imm8);
398  uint64_t bit_a = (imm8 >> 7) & 1;
399  uint64_t bit_b = (imm8 >> 6) & 1;
400  uint64_t slice = imm8 & 0x3f;
401  return (bit_a << 63) | ((UINT64_C(1) << 62) - (bit_b << 54)) | (slice << 48);
402}
403
404enum T2LitType {
405  kT2LitInvalid,
406  kT2LitUByte,
407  kT2LitSByte,
408  kT2LitUHalf,
409  kT2LitSHalf,
410  kT2LitUWord,
411  kT2LitSWord,
412  kT2LitHexWord,
413  kT2LitULong,
414  kT2LitSLong,
415  kT2LitHexLong,
416};
417std::ostream& operator<<(std::ostream& os, T2LitType type) {
418  return os << static_cast<int>(type);
419}
420
421void DumpThumb2Literal(std::ostream& args, const uint8_t* instr_ptr, uint32_t U, uint32_t imm32,
422                       T2LitType type) {
423  // Literal offsets (imm32) are not required to be aligned so we may need unaligned access.
424  typedef const int16_t unaligned_int16_t __attribute__ ((aligned (1)));
425  typedef const uint16_t unaligned_uint16_t __attribute__ ((aligned (1)));
426  typedef const int32_t unaligned_int32_t __attribute__ ((aligned (1)));
427  typedef const uint32_t unaligned_uint32_t __attribute__ ((aligned (1)));
428  typedef const int64_t unaligned_int64_t __attribute__ ((aligned (1)));
429  typedef const uint64_t unaligned_uint64_t __attribute__ ((aligned (1)));
430
431  uintptr_t pc = RoundDown(reinterpret_cast<intptr_t>(instr_ptr) + 4, 4);
432  uintptr_t lit_adr = U ? pc + imm32 : pc - imm32;
433  args << "  ; ";
434  switch (type) {
435    case kT2LitUByte:
436      args << *reinterpret_cast<const uint8_t*>(lit_adr);
437      break;
438    case kT2LitSByte:
439      args << *reinterpret_cast<const int8_t*>(lit_adr);
440      break;
441    case kT2LitUHalf:
442      args << *reinterpret_cast<const unaligned_uint16_t*>(lit_adr);
443      break;
444    case kT2LitSHalf:
445      args << *reinterpret_cast<const unaligned_int16_t*>(lit_adr);
446      break;
447    case kT2LitUWord:
448      args << *reinterpret_cast<const unaligned_uint32_t*>(lit_adr);
449      break;
450    case kT2LitSWord:
451      args << *reinterpret_cast<const unaligned_int32_t*>(lit_adr);
452      break;
453    case kT2LitHexWord:
454      args << StringPrintf("0x%08x", *reinterpret_cast<const unaligned_uint32_t*>(lit_adr));
455      break;
456    case kT2LitULong:
457      args << *reinterpret_cast<const unaligned_uint64_t*>(lit_adr);
458      break;
459    case kT2LitSLong:
460      args << *reinterpret_cast<const unaligned_int64_t*>(lit_adr);
461      break;
462    case kT2LitHexLong:
463      args << StringPrintf("0x%" PRIx64, *reinterpret_cast<unaligned_int64_t*>(lit_adr));
464      break;
465    default:
466      LOG(FATAL) << "Invalid type: " << type;
467      break;
468  }
469}
470
471size_t DisassemblerArm::DumpThumb32(std::ostream& os, const uint8_t* instr_ptr) {
472  uint32_t instr = (ReadU16(instr_ptr) << 16) | ReadU16(instr_ptr + 2);
473  // |111|1 1|1000000|0000|1111110000000000|
474  // |5 3|2 1|0987654|3  0|5    0    5    0|
475  // |---|---|-------|----|----------------|
476  // |332|2 2|2222222|1111|1111110000000000|
477  // |1 9|8 7|6543210|9  6|5    0    5    0|
478  // |---|---|-------|----|----------------|
479  // |111|op1| op2   |    |                |
480  uint32_t op1 = (instr >> 27) & 3;
481  if (op1 == 0) {
482    return DumpThumb16(os, instr_ptr);
483  }
484
485  uint32_t op2 = (instr >> 20) & 0x7F;
486  std::ostringstream opcode;
487  std::ostringstream args;
488  switch (op1) {
489    case 0:
490      break;
491    case 1:
492      if ((op2 & 0x64) == 0) {  // 00x x0xx
493        // |111|11|10|00|0|00|0000|1111110000000000|
494        // |5 3|21|09|87|6|54|3  0|5    0    5    0|
495        // |---|--|--|--|-|--|----|----------------|
496        // |332|22|22|22|2|22|1111|1111110000000000|
497        // |1 9|87|65|43|2|10|9  6|5    0    5    0|
498        // |---|--|--|--|-|--|----|----------------|
499        // |111|01|00|op|0|WL| Rn |                |
500        // |111|01| op2      |    |                |
501        // STM - 111 01 00-01-0-W0 nnnn rrrrrrrrrrrrrrrr
502        // LDM - 111 01 00-01-0-W1 nnnn rrrrrrrrrrrrrrrr
503        // PUSH- 111 01 00-01-0-10 1101 0M0rrrrrrrrrrrrr
504        // POP - 111 01 00-01-0-11 1101 PM0rrrrrrrrrrrrr
505        uint32_t op = (instr >> 23) & 3;
506        uint32_t W = (instr >> 21) & 1;
507        uint32_t L = (instr >> 20) & 1;
508        ArmRegister Rn(instr, 16);
509        if (op == 1 || op == 2) {
510          if (op == 1) {
511            if (L == 0) {
512              opcode << "stm";
513              args << Rn << (W == 0 ? "" : "!") << ", ";
514            } else {
515              if (Rn.r != 13) {
516                opcode << "ldm";
517                args << Rn << (W == 0 ? "" : "!") << ", ";
518              } else {
519                opcode << "pop";
520              }
521            }
522          } else {
523            if (L == 0) {
524              if (Rn.r != 13) {
525                opcode << "stmdb";
526                args << Rn << (W == 0 ? "" : "!") << ", ";
527              } else {
528                opcode << "push";
529              }
530            } else {
531              opcode << "ldmdb";
532              args << Rn << (W == 0 ? "" : "!") << ", ";
533            }
534          }
535          args << RegisterList(instr);
536        }
537      } else if ((op2 & 0x64) == 4) {  // 00x x1xx
538        uint32_t op3 = (instr >> 23) & 3;
539        uint32_t op4 = (instr >> 20) & 3;
540        // uint32_t op5 = (instr >> 4) & 0xF;
541        ArmRegister Rn(instr, 16);
542        ArmRegister Rt(instr, 12);
543        ArmRegister Rd(instr, 8);
544        uint32_t imm8 = instr & 0xFF;
545        if ((op3 & 2) == 2) {     // 1x
546          int W = (instr >> 21) & 1;
547          int U = (instr >> 23) & 1;
548          int P = (instr >> 24) & 1;
549
550          if ((op4 & 1) == 1) {
551            opcode << "ldrd";
552          } else {
553            opcode << "strd";
554          }
555          args << Rt << "," << Rd << ", [" << Rn;
556          const char *sign = U ? "+" : "-";
557          if (P == 0 && W == 1) {
558            args << "], #" << sign << (imm8 << 2);
559          } else {
560            args << ", #" << sign << (imm8 << 2) << "]";
561            if (W == 1) {
562              args << "!";
563            }
564          }
565        } else {                  // 0x
566          switch (op4) {
567            case 0:
568              if (op3 == 0) {   // op3 is 00, op4 is 00
569                opcode << "strex";
570                args << Rd << ", " << Rt << ", [" << Rn << ", #" << (imm8 << 2) << "]";
571                if (Rd.r == 13 || Rd.r == 15 || Rt.r == 13 || Rt.r == 15 || Rn.r == 15 ||
572                    Rd.r == Rn.r || Rd.r == Rt.r) {
573                  args << " (UNPREDICTABLE)";
574                }
575              } else {          // op3 is 01, op4 is 00
576                // this is one of strexb, strexh or strexd
577                int op5 = (instr >> 4) & 0xf;
578                switch (op5) {
579                  case 4:
580                  case 5:
581                    opcode << ((op5 == 4) ? "strexb" : "strexh");
582                    Rd = ArmRegister(instr, 0);
583                    args << Rd << ", " << Rt << ", [" << Rn << "]";
584                    if (Rd.r == 13 || Rd.r == 15 || Rt.r == 13 || Rt.r == 15 || Rn.r == 15 ||
585                        Rd.r == Rn.r || Rd.r == Rt.r || (instr & 0xf00) != 0xf00) {
586                      args << " (UNPREDICTABLE)";
587                    }
588                    break;
589                  case 7:
590                    opcode << "strexd";
591                    ArmRegister Rt2 = Rd;
592                    Rd = ArmRegister(instr, 0);
593                    args << Rd << ", " << Rt << ", " << Rt2 << ", [" << Rn << "]";
594                    if (Rd.r == 13 || Rd.r == 15 || Rt.r == 13 || Rt.r == 15 ||
595                        Rt2.r == 13 || Rt2.r == 15 || Rn.r == 15 ||
596                        Rd.r == Rn.r || Rd.r == Rt.r || Rd.r == Rt2.r) {
597                      args << " (UNPREDICTABLE)";
598                    }
599                    break;
600                }
601              }
602              break;
603            case 1:
604              if (op3 == 0) {   // op3 is 00, op4 is 01
605                opcode << "ldrex";
606                args << Rt << ", [" << Rn << ", #" << (imm8 << 2) << "]";
607                if (Rt.r == 13 || Rt.r == 15 || Rn.r == 15 || (instr & 0xf00) != 0xf00) {
608                  args << " (UNPREDICTABLE)";
609                }
610              } else {          // op3 is 01, op4 is 01
611                // this is one of strexb, strexh or strexd
612                int op5 = (instr >> 4) & 0xf;
613                switch (op5) {
614                  case 0:
615                    opcode << "tbb";
616                    break;
617                  case 1:
618                    opcode << "tbh";
619                    break;
620                  case 4:
621                  case 5:
622                    opcode << ((op5 == 4) ? "ldrexb" : "ldrexh");
623                    args << Rt << ", [" << Rn << "]";
624                    if (Rt.r == 13 || Rt.r == 15 || Rn.r == 15 || (instr & 0xf0f) != 0xf0f) {
625                      args << " (UNPREDICTABLE)";
626                    }
627                    break;
628                  case 7:
629                    opcode << "ldrexd";
630                    args << Rt << ", " << Rd /* Rt2 */ << ", [" << Rn << "]";
631                    if (Rt.r == 13 || Rt.r == 15 || Rd.r == 13 /* Rt2 */ || Rd.r == 15 /* Rt2 */ ||
632                        Rn.r == 15 || (instr & 0x00f) != 0x00f) {
633                      args << " (UNPREDICTABLE)";
634                    }
635                    break;
636                }
637              }
638              break;
639            case 2:     // op3 is 0x, op4 is 10
640            case 3:   // op3 is 0x, op4 is 11
641              if (op4 == 2) {
642                opcode << "strd";
643              } else {
644                opcode << "ldrd";
645              }
646              int W = (instr >> 21) & 1;
647              int U = (instr >> 23) & 1;
648              int P = (instr >> 24) & 1;
649
650              args << Rt << "," << Rd << ", [" << Rn;
651              const char *sign = U ? "+" : "-";
652              if (P == 0 && W == 1) {
653                args << "], #" << sign << imm8;
654              } else {
655                args << ", #" << sign << imm8 << "]";
656                if (W == 1) {
657                  args << "!";
658                }
659              }
660              break;
661          }
662        }
663
664      } else if ((op2 & 0x60) == 0x20) {  // 01x xxxx
665        // Data-processing (shifted register)
666        // |111|1110|0000|0|0000|1111|1100|00|00|0000|
667        // |5 3|2109|8765|4|3  0|5   |10 8|7 |5 |3  0|
668        // |---|----|----|-|----|----|----|--|--|----|
669        // |332|2222|2222|2|1111|1111|1100|00|00|0000|
670        // |1 9|8765|4321|0|9  6|5   |10 8|7 |5 |3  0|
671        // |---|----|----|-|----|----|----|--|--|----|
672        // |111|0101| op3|S| Rn |imm3| Rd |i2|ty| Rm |
673        uint32_t op3 = (instr >> 21) & 0xF;
674        uint32_t S = (instr >> 20) & 1;
675        uint32_t imm3 = ((instr >> 12) & 0x7);
676        uint32_t imm2 = ((instr >> 6) & 0x3);
677        uint32_t imm5 = ((imm3 << 2) | imm2);
678        uint32_t shift_type = ((instr >> 4) & 0x3);
679        ArmRegister Rd(instr, 8);
680        ArmRegister Rn(instr, 16);
681        ArmRegister Rm(instr, 0);
682        switch (op3) {
683          case 0x0:
684            if (Rd.r != 0xF) {
685              opcode << "and";
686            } else {
687              if (S != 1U) {
688                opcode << "UNKNOWN TST-" << S;
689                break;
690              }
691              opcode << "tst";
692              S = 0;  // don't print 's'
693            }
694            break;
695          case 0x1: opcode << "bic"; break;
696          case 0x2:
697            if (Rn.r != 0xF) {
698              opcode << "orr";
699            } else {
700              // TODO: use canonical form if there is a shift (lsl, ...).
701              opcode << "mov";
702            }
703            break;
704          case 0x3:
705            if (Rn.r != 0xF) {
706              opcode << "orn";
707            } else {
708              opcode << "mvn";
709            }
710            break;
711          case 0x4:
712            if (Rd.r != 0xF) {
713              opcode << "eor";
714            } else {
715              if (S != 1U) {
716                opcode << "UNKNOWN TEQ-" << S;
717                break;
718              }
719              opcode << "teq";
720              S = 0;  // don't print 's'
721            }
722            break;
723          case 0x6: opcode << "pkh"; break;
724          case 0x8:
725            if (Rd.r != 0xF) {
726              opcode << "add";
727            } else {
728              if (S != 1U) {
729                opcode << "UNKNOWN CMN-" << S;
730                break;
731              }
732              opcode << "cmn";
733              S = 0;  // don't print 's'
734            }
735            break;
736          case 0xA: opcode << "adc"; break;
737          case 0xB: opcode << "sbc"; break;
738          case 0xD:
739            if (Rd.r != 0xF) {
740              opcode << "sub";
741            } else {
742              if (S != 1U) {
743                opcode << "UNKNOWN CMP-" << S;
744                break;
745              }
746              opcode << "cmp";
747              S = 0;  // don't print 's'
748            }
749            break;
750          case 0xE: opcode << "rsb"; break;
751          default: opcode << "UNKNOWN DPSR-" << op3; break;
752        }
753
754        if (S == 1) {
755          opcode << "s";
756        }
757        opcode << ".w";
758
759        if (Rd.r != 0xF) {
760          args << Rd << ", ";
761        }
762        if (Rn.r != 0xF) {
763          args << Rn << ", ";
764        }
765        args << Rm;
766
767        // Shift operand.
768        bool noShift = (imm5 == 0 && shift_type != 0x3);
769        if (!noShift) {
770          args << ", ";
771          switch (shift_type) {
772            case 0x0: args << "lsl"; break;
773            case 0x1: args << "lsr"; break;
774            case 0x2: args << "asr"; break;
775            case 0x3:
776              if (imm5 == 0) {
777                args << "rrx";
778              } else {
779                args << "ror";
780              }
781              break;
782          }
783          if (shift_type != 0x3 /* rrx */) {
784            args << StringPrintf(" #%d", (0 != imm5 || 0 == shift_type) ? imm5 : 32);
785          }
786        }
787
788      } else if ((op2 & 0x40) == 0x40) {  // 1xx xxxx
789        // Co-processor instructions
790        // |111|1|11|000000|0000|1111|1100|000|0  |0000|
791        // |5 3|2|10|987654|3  0|54 2|10 8|7 5|4  |   0|
792        // |---|-|--|------|----|----|----|---|---|----|
793        // |332|2|22|222222|1111|1111|1100|000|0  |0000|
794        // |1 9|8|76|543210|9  6|54 2|10 8|7 5|4  |   0|
795        // |---|-|--|------|----|----|----|---|---|----|
796        // |111| |11| op3  | Rn |    |copr|   |op4|    |
797        uint32_t op3 = (instr >> 20) & 0x3F;
798        uint32_t coproc = (instr >> 8) & 0xF;
799        uint32_t op4 = (instr >> 4) & 0x1;
800
801        if (coproc == 0xA || coproc == 0xB) {   // 101x
802          if (op3 < 0x20 && (op3 & ~5) != 0) {     // 0xxxxx and not 000x0x
803            // Extension register load/store instructions
804            // |1111|110|00000|0000|1111|110|0|00000000|
805            // |5  2|1 9|87654|3  0|5  2|1 9|8|7      0|
806            // |----|---|-----|----|----|---|-|--------|
807            // |3322|222|22222|1111|1111|110|0|00000000|
808            // |1  8|7 5|4   0|9  6|5  2|1 9|8|7      0|
809            // |----|---|-----|----|----|---|-|--------|
810            // |1110|110|PUDWL| Rn | Vd |101|S|  imm8  |
811            uint32_t P = (instr >> 24) & 1;
812            uint32_t U = (instr >> 23) & 1;
813            uint32_t W = (instr >> 21) & 1;
814            if (P == U && W == 1) {
815              opcode << "UNDEFINED";
816            } else {
817              uint32_t L = (instr >> 20) & 1;
818              uint32_t S = (instr >> 8) & 1;
819              ArmRegister Rn(instr, 16);
820              if (P == 1 && W == 0) {  // VLDR
821                FpRegister d(instr, 12, 22);
822                uint32_t imm8 = instr & 0xFF;
823                opcode << (L == 1 ? "vldr" : "vstr");
824                args << d << ", [" << Rn << ", #" << ((U == 1) ? "" : "-")
825                     << (imm8 << 2) << "]";
826                if (Rn.r == 15 && U == 1) {
827                  DumpThumb2Literal(args, instr_ptr, U, imm8 << 2, kT2LitHexLong);
828                }
829              } else if (Rn.r == 13 && W == 1 && U == L) {  // VPUSH/VPOP
830                opcode << (L == 1 ? "vpop" : "vpush");
831                args << FpRegisterRange(instr);
832              } else {  // VLDM
833                opcode << (L == 1 ? "vldm" : "vstm");
834                args << Rn << ((W == 1) ? "!" : "") << ", "
835                     << FpRegisterRange(instr);
836              }
837              opcode << (S == 1 ? ".f64" : ".f32");
838            }
839          } else if ((op3 >> 1) == 2) {      // 00010x
840            if ((instr & 0xD0) == 0x10) {
841              // 64bit transfers between ARM core and extension registers.
842              uint32_t L = (instr >> 20) & 1;
843              uint32_t S = (instr >> 8) & 1;
844              ArmRegister Rt2(instr, 16);
845              ArmRegister Rt(instr, 12);
846              FpRegister m(instr, 0, 5);
847              opcode << "vmov" << (S ? ".f64" : ".f32");
848              if (L == 1) {
849                args << Rt << ", " << Rt2 << ", ";
850              }
851              if (S) {
852                args << m;
853              } else {
854                args << m << ", " << FpRegister(m, 1);
855              }
856              if (L == 0) {
857                args << ", " << Rt << ", " << Rt2;
858              }
859              if (Rt.r == 15 || Rt.r == 13 || Rt2.r == 15 || Rt2.r == 13 ||
860                  (S == 0 && m.r == 31) || (L == 1 && Rt.r == Rt2.r)) {
861                args << " (UNPREDICTABLE)";
862              }
863            }
864          } else if ((op3 >> 4) == 2 && op4 == 0) {     // 10xxxx, op = 0
865            // fp data processing
866            // VMLA, VMLS, VMUL, VNMUL, VADD, VSUB, VDIV, VMOV, ...
867            // |1111|1100|0|0|00|0000|1111|110|0|0|0|0|0|0000|
868            // |5  2|1  8|7|6|54|3  0|5  2|1 9|8|7|6|5|4|3  0|
869            // |----|----|-|-|--|----|----|---|-|-|-|-|-|----|
870            // |3322|2222|2|2|22|1111|1111|110|0|0|0|0|0|0000|
871            // |1  8|7  4|3|2|10|9  6|5  2|1 9|8|7|6|5|4|3  0|
872            // |----|----|-|-|--|----|----|---|-|-|-|-|-|----|
873            // |1110|1110|  op3 | Vn | Vd |101|S|N|Q|M|0| Vm |
874            // |1110|1110|0|D|00| Vn | Vd |101|S|N|0|M|0| Vm | VMLA
875            // |1110|1110|0|D|00| Vn | Vd |101|S|N|1|M|0| Vm | VMLS
876            // |1110|1110|0|D|10| Vn | Vd |101|S|N|0|M|0| Vm | VMUL
877            // |1110|1110|0|D|10| Vn | Vd |101|S|N|1|M|0| Vm | VNMUL
878            // |1110|1110|0|D|11| Vn | Vd |101|S|N|0|M|0| Vm | VADD
879            // |1110|1110|0|D|11| Vn | Vd |101|S|N|1|M|0| Vm | VSUB
880            // |1110|1110|1|D|00| Vn | Vd |101|S|N|0|M|0| Vm | VDIV
881            // |1110|1110|1|D|11| iH | Vd |101|S|0|0|0|0| iL | VMOV (imm)
882            // |1110|1110|1|D|11|op5 | Vd |101|S|.|1|M|0| Vm | ... (see below)
883            uint32_t S = (instr >> 8) & 1;
884            uint32_t Q = (instr >> 6) & 1;
885            FpRegister d(instr, 12, 22);
886            FpRegister n(instr, 16, 7);
887            FpRegister m(instr, 0, 5);
888            if ((op3 & 0xB) == 0) {  // 100x00
889              opcode << (Q == 0 ? "vmla" : "vmls") << (S != 0 ? ".f64" : ".f32");
890              args << d << ", " << n << ", " << m;
891            } else if ((op3 & 0xB) == 0x2) {  // 100x10
892              opcode << (Q == 0 ? "vmul" : "vnmul") << (S != 0 ? ".f64" : ".f32");
893              args << d << ", " << n << ", " << m;
894            } else if ((op3 & 0xB) == 0x3) {  // 100x11
895              opcode << (Q == 0 ? "vadd" : "vsub") << (S != 0 ? ".f64" : ".f32");
896              args << d << ", " << n << ", " << m;
897            } else if ((op3 & 0xB) == 0x8 && Q == 0) {  // 101x00, Q == 0
898              opcode << "vdiv" << (S != 0 ? ".f64" : ".f32");
899              args << d << ", " << n << ", " << m;
900            } else if ((op3 & 0xB) == 0xB && Q == 0) {  // 101x11, Q == 0
901              uint32_t imm8 = ((instr & 0xf0000u) >> 12) | (instr & 0xfu);
902              opcode << "vmov" << (S != 0 ? ".f64" : ".f32");
903              args << d << ", " << (S != 0 ? StringPrintf("0x%016" PRIx64, VFPExpand64(imm8))
904                                           : StringPrintf("0x%08x", VFPExpand32(imm8)));
905              if ((instr & 0xa0) != 0) {
906                args << " (UNPREDICTABLE)";
907              }
908            } else if ((op3 & 0xB) == 0xB && Q == 1) {  // 101x11, Q == 1
909              // VNEG, VSQRT, VCMP, VCMPE, VCVT (floating-point conversion)
910              // |1111|1100|0|0|00|0000|1111|110|0|0 |0|0|0|0000|
911              // |5  2|1  8|7|6|54|3  0|5  2|1 9|8|7 |6|5|4|3  0|
912              // |----|----|-|-|--|----|----|---|-|- |-|-|-|----|
913              // |3322|2222|2|2|22|1111|1111|110|0|0 |0|0|0|0000|
914              // |1  8|7  4|3|2|10|9  6|5  2|1 9|8|7 |6|5|4|3  0|
915              // |----|----|-|-|--|----|----|---|-|- |-|-|-|----|
916              // |1110|1110|1|D|11|0000| Vd |101|S|0 |1|M|0| Vm | VMOV (reg)
917              // |1110|1110|1|D|11|0000| Vd |101|S|1 |1|M|0| Vm | VABS
918              // |1110|1110|1|D|11|0001| Vd |101|S|0 |1|M|0| Vm | VNEG
919              // |1110|1110|1|D|11|0001| Vd |101|S|1 |1|M|0| Vm | VSQRT
920              // |1110|1110|1|D|11|0100| Vd |101|S|op|1|M|0| Vm | VCMP
921              // |1110|1110|1|D|11|0101| Vd |101|S|op|1|0|0|0000| VCMPE
922              // |1110|1110|1|D|11|op5 | Vd |101|S|op|1|M|0| Vm | VCVT
923              uint32_t op5 = (instr >> 16) & 0xF;
924              uint32_t op = (instr >> 7) & 1;
925              // Register types in VCVT instructions rely on the combination of op5 and S.
926              FpRegister Dd(instr, 12, 22, 1);
927              FpRegister Sd(instr, 12, 22, 0);
928              FpRegister Dm(instr, 0, 5, 1);
929              FpRegister Sm(instr, 0, 5, 0);
930              if (op5 == 0) {
931                opcode << (op == 0 ? "vmov" : "vabs") << (S != 0 ? ".f64" : ".f32");
932                args << d << ", " << m;
933              } else if (op5 == 1) {
934                opcode << (op != 0 ? "vsqrt" : "vneg") << (S != 0 ? ".f64" : ".f32");
935                args << d << ", " << m;
936              } else if (op5 == 4) {
937                opcode << "vcmp" << (S != 0 ? ".f64" : ".f32");
938                args << d << ", " << m;
939                if (op != 0) {
940                  args << " (quiet nan)";
941                }
942              } else if (op5 == 5) {
943                opcode << "vcmpe" << (S != 0 ? ".f64" : ".f32");
944                args << d << ", #0.0";
945                if (op != 0) {
946                  args << " (quiet nan)";
947                }
948                if ((instr & 0x2f) != 0) {
949                  args << " (UNPREDICTABLE)";
950                }
951              } else if (op5 == 0xD) {
952                if (S == 1) {
953                  // vcvt{r}.s32.f64
954                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".s32.f64";
955                  args << Sd << ", " << Dm;
956                } else {
957                  // vcvt{r}.s32.f32
958                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".s32.f32";
959                  args << Sd << ", " << Sm;
960                }
961              } else if (op5 == 0xC) {
962                if (S == 1) {
963                  // vcvt{r}.u32.f64
964                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".u32.f64";
965                  args << Sd << ", " << Dm;
966                } else {
967                  // vcvt{r}.u32.f32
968                  opcode << "vcvt" << (op == 0 ? "r" : "") << ".u32.f32";
969                  args << Sd << ", " << Sm;
970                }
971              } else if (op5 == 0x8) {
972                if (S == 1) {
973                  // vcvt.f64.<Tm>
974                  opcode << "vcvt.f64." << (op == 0 ? "u" : "s") << "32";
975                  args << Dd << ", " << Sm;
976                } else {
977                  // vcvt.f32.<Tm>
978                  opcode << "vcvt.f32." << (op == 0 ? "u" : "s") << "32";
979                  args << Sd << ", " << Sm;
980                }
981              } else if (op5 == 0x7) {
982                if (op == 1) {
983                  if (S == 1) {
984                    // vcvt.f64.f32
985                    opcode << "vcvt.f64.f32";
986                    args << Dd << ", " << Sm;
987                  } else {
988                    // vcvt.f32.f64
989                    opcode << "vcvt.f32.f64";
990                    args << Sd << ", " << Dm;
991                  }
992                }
993              } else if ((op5 & 0xa) == 0xa) {
994                opcode << "vcvt";
995                args << "[undecoded: floating <-> fixed]";
996              }
997            }
998          } else if ((op3 >> 4) == 2 && op4 == 1) {     // 10xxxx, op = 1
999            if (coproc == 10 && (op3 & 0xE) == 0) {
1000              // VMOV (between ARM core register and single-precision register)
1001              // |1111|1100|000|0 |0000|1111|1100|0|00|0|0000|
1002              // |5   |1  8|7 5|4 |3  0|5  2|1  8|7|65|4|3  0|
1003              // |----|----|---|- |----|----|----|-|--|-|----|
1004              // |3322|2222|222|2 |1111|1111|1100|0|00|0|0000|
1005              // |1  8|7  4|3 1|0 |9  6|5  2|1  8|7|65|4|3  0|
1006              // |----|----|---|- |----|----|----|-|--|-|----|
1007              // |1110|1110|000|op| Vn | Rt |1010|N|00|1|0000|
1008              uint32_t op = op3 & 1;
1009              ArmRegister Rt(instr, 12);
1010              FpRegister n(instr, 16, 7);
1011              opcode << "vmov.f32";
1012              if (op) {
1013                args << Rt << ", " << n;
1014              } else {
1015                args << n << ", " << Rt;
1016              }
1017              if (Rt.r == 13 || Rt.r == 15 || (instr & 0x6F) != 0) {
1018                args << " (UNPREDICTABLE)";
1019              }
1020            } else if (coproc == 10 && op3 == 0x2F) {
1021              // VMRS
1022              // |1111|11000000|0000|1111|1100|000|0|0000|
1023              // |5   |1      4|3  0|5  2|1  8|7 5|4|3  0|
1024              // |----|--------|----|----|----|---|-|----|
1025              // |3322|22222222|1111|1111|1100|000|0|0000|
1026              // |1  8|7      0|9  6|5  2|1  8|7 5|4|3  0|
1027              // |----|--------|----|----|----|---|-|----|
1028              // |1110|11101111|reg | Rt |1010|000|1|0000| - last 7 0s are (0)
1029              uint32_t spec_reg = (instr >> 16) & 0xF;
1030              ArmRegister Rt(instr, 12);
1031              opcode << "vmrs";
1032              if (spec_reg == 1) {
1033                if (Rt.r == 15) {
1034                  args << "APSR_nzcv, FPSCR";
1035                } else if (Rt.r == 13) {
1036                  args << Rt << ", FPSCR (UNPREDICTABLE)";
1037                } else {
1038                  args << Rt << ", FPSCR";
1039                }
1040              } else {
1041                args << "(PRIVILEGED)";
1042              }
1043            } else if (coproc == 11 && (op3 & 0x9) != 8) {
1044              // VMOV (ARM core register to scalar or vice versa; 8/16/32-bit)
1045            }
1046          }
1047        }
1048      }
1049      break;
1050    case 2:
1051      if ((instr & 0x8000) == 0 && (op2 & 0x20) == 0) {
1052        // Data-processing (modified immediate)
1053        // |111|11|10|0000|0|0000|1|111|1100|00000000|
1054        // |5 3|21|09|8765|4|3  0|5|4 2|10 8|7 5    0|
1055        // |---|--|--|----|-|----|-|---|----|--------|
1056        // |332|22|22|2222|2|1111|1|111|1100|00000000|
1057        // |1 9|87|65|4321|0|9  6|5|4 2|10 8|7 5    0|
1058        // |---|--|--|----|-|----|-|---|----|--------|
1059        // |111|10|i0| op3|S| Rn |0|iii| Rd |iiiiiiii|
1060        //  111 10 x0 xxxx x xxxx opxxx xxxx xxxxxxxx
1061        uint32_t i = (instr >> 26) & 1;
1062        uint32_t op3 = (instr >> 21) & 0xF;
1063        uint32_t S = (instr >> 20) & 1;
1064        ArmRegister Rn(instr, 16);
1065        uint32_t imm3 = (instr >> 12) & 7;
1066        ArmRegister Rd(instr, 8);
1067        uint32_t imm8 = instr & 0xFF;
1068        int32_t imm32 = (i << 11) | (imm3 << 8) | imm8;
1069        if (Rn.r == 0xF && (op3 == 0x2 || op3 == 0x3)) {
1070          if (op3 == 0x2) {
1071            opcode << "mov";
1072            if (S == 1) {
1073              opcode << "s";
1074            }
1075            opcode << ".w";
1076          } else {
1077            opcode << "mvn";
1078            if (S == 1) {
1079              opcode << "s";
1080            }
1081          }
1082          args << Rd << ", #" << ThumbExpand(imm32);
1083        } else if (Rd.r == 0xF && S == 1 &&
1084                   (op3 == 0x0 || op3 == 0x4 || op3 == 0x8 || op3 == 0xD)) {
1085          if (op3 == 0x0) {
1086            opcode << "tst";
1087          } else if (op3 == 0x4) {
1088            opcode << "teq";
1089          } else if (op3 == 0x8) {
1090            opcode << "cmn.w";
1091          } else {
1092            opcode << "cmp.w";
1093          }
1094          args << Rn << ", #" << ThumbExpand(imm32);
1095        } else {
1096          switch (op3) {
1097            case 0x0: opcode << "and"; break;
1098            case 0x1: opcode << "bic"; break;
1099            case 0x2: opcode << "orr"; break;
1100            case 0x3: opcode << "orn"; break;
1101            case 0x4: opcode << "eor"; break;
1102            case 0x8: opcode << "add"; break;
1103            case 0xA: opcode << "adc"; break;
1104            case 0xB: opcode << "sbc"; break;
1105            case 0xD: opcode << "sub"; break;
1106            case 0xE: opcode << "rsb"; break;
1107            default: opcode << "UNKNOWN DPMI-" << op3; break;
1108          }
1109          if (S == 1) {
1110            opcode << "s";
1111          }
1112          args << Rd << ", " << Rn << ", #" << ThumbExpand(imm32);
1113        }
1114      } else if ((instr & 0x8000) == 0 && (op2 & 0x20) != 0) {
1115        // Data-processing (plain binary immediate)
1116        // |111|11|10|00000|0000|1|111110000000000|
1117        // |5 3|21|09|87654|3  0|5|4   0    5    0|
1118        // |---|--|--|-----|----|-|---------------|
1119        // |332|22|22|22222|1111|1|111110000000000|
1120        // |1 9|87|65|43210|9  6|5|4   0    5    0|
1121        // |---|--|--|-----|----|-|---------------|
1122        // |111|10|x1| op3 | Rn |0|xxxxxxxxxxxxxxx|
1123        uint32_t op3 = (instr >> 20) & 0x1F;
1124        switch (op3) {
1125          case 0x00: case 0x0A: {
1126            // ADD/SUB.W Rd, Rn #imm12 - 111 10 i1 0101 0 nnnn 0 iii dddd iiiiiiii
1127            ArmRegister Rd(instr, 8);
1128            ArmRegister Rn(instr, 16);
1129            uint32_t i = (instr >> 26) & 1;
1130            uint32_t imm3 = (instr >> 12) & 0x7;
1131            uint32_t imm8 = instr & 0xFF;
1132            uint32_t imm12 = (i << 11) | (imm3 << 8) | imm8;
1133            if (Rn.r != 0xF) {
1134              opcode << (op3 == 0 ? "addw" : "subw");
1135              args << Rd << ", " << Rn << ", #" << imm12;
1136            } else {
1137              opcode << "adr";
1138              args << Rd << ", ";
1139              DumpBranchTarget(args, instr_ptr + 4, (op3 == 0) ? imm12 : -imm12);
1140            }
1141            break;
1142          }
1143          case 0x04: case 0x0C: {
1144            // MOVW/T Rd, #imm16     - 111 10 i0 0010 0 iiii 0 iii dddd iiiiiiii
1145            ArmRegister Rd(instr, 8);
1146            uint32_t i = (instr >> 26) & 1;
1147            uint32_t imm3 = (instr >> 12) & 0x7;
1148            uint32_t imm8 = instr & 0xFF;
1149            uint32_t Rn = (instr >> 16) & 0xF;
1150            uint32_t imm16 = (Rn << 12) | (i << 11) | (imm3 << 8) | imm8;
1151            opcode << (op3 == 0x04 ? "movw" : "movt");
1152            args << Rd << ", #" << imm16;
1153            break;
1154          }
1155          case 0x16: {
1156            // BFI Rd, Rn, #lsb, #width - 111 10 0 11 011 0 nnnn 0 iii dddd ii 0 iiiii
1157            ArmRegister Rd(instr, 8);
1158            ArmRegister Rn(instr, 16);
1159            uint32_t msb = instr & 0x1F;
1160            uint32_t imm2 = (instr >> 6) & 0x3;
1161            uint32_t imm3 = (instr >> 12) & 0x7;
1162            uint32_t lsb = (imm3 << 2) | imm2;
1163            uint32_t width = msb - lsb + 1;
1164            if (Rn.r != 0xF) {
1165              opcode << "bfi";
1166              args << Rd << ", " << Rn << ", #" << lsb << ", #" << width;
1167            } else {
1168              opcode << "bfc";
1169              args << Rd << ", #" << lsb << ", #" << width;
1170            }
1171            break;
1172          }
1173          default:
1174            break;
1175        }
1176      } else {
1177        // Branches and miscellaneous control
1178        // |111|11|1000000|0000|1|111|1100|00000000|
1179        // |5 3|21|0987654|3  0|5|4 2|10 8|7 5    0|
1180        // |---|--|-------|----|-|---|----|--------|
1181        // |332|22|2222222|1111|1|111|1100|00000000|
1182        // |1 9|87|6543210|9  6|5|4 2|10 8|7 5    0|
1183        // |---|--|-------|----|-|---|----|--------|
1184        // |111|10| op2   |    |1|op3|op4 |        |
1185
1186        uint32_t op3 = (instr >> 12) & 7;
1187        // uint32_t op4 = (instr >> 8) & 0xF;
1188        switch (op3) {
1189          case 0:
1190            if ((op2 & 0x38) != 0x38) {
1191              // Conditional branch
1192              // |111|11|1|0000|000000|1|1|1 |1|1 |10000000000|
1193              // |5 3|21|0|9876|543  0|5|4|3 |2|1 |0    5    0|
1194              // |---|--|-|----|------|-|-|--|-|--|-----------|
1195              // |332|22|2|2222|221111|1|1|1 |1|1 |10000000000|
1196              // |1 9|87|6|5432|109  6|5|4|3 |2|1 |0    5    0|
1197              // |---|--|-|----|------|-|-|--|-|--|-----------|
1198              // |111|10|S|cond| imm6 |1|0|J1|0|J2| imm11     |
1199              uint32_t S = (instr >> 26) & 1;
1200              uint32_t J2 = (instr >> 11) & 1;
1201              uint32_t J1 = (instr >> 13) & 1;
1202              uint32_t imm6 = (instr >> 16) & 0x3F;
1203              uint32_t imm11 = instr & 0x7FF;
1204              uint32_t cond = (instr >> 22) & 0xF;
1205              int32_t imm32 = (S << 20) |  (J2 << 19) | (J1 << 18) | (imm6 << 12) | (imm11 << 1);
1206              imm32 = (imm32 << 11) >> 11;  // sign extend 21bit immediate
1207              opcode << "b";
1208              DumpCond(opcode, cond);
1209              opcode << ".w";
1210              DumpBranchTarget(args, instr_ptr + 4, imm32);
1211            } else if (op2 == 0x3B) {
1212              // Miscellaneous control instructions
1213              uint32_t op5 = (instr >> 4) & 0xF;
1214              switch (op5) {
1215                case 4: opcode << "dsb"; DumpMemoryDomain(args, instr & 0xF); break;
1216                case 5: opcode << "dmb"; DumpMemoryDomain(args, instr & 0xF); break;
1217                case 6: opcode << "isb"; DumpMemoryDomain(args, instr & 0xF); break;
1218              }
1219            }
1220            break;
1221          case 2:
1222            if ((op2 & 0x38) == 0x38) {
1223              if (op2 == 0x7F) {
1224                opcode << "udf";
1225              }
1226              break;
1227            }
1228            FALLTHROUGH_INTENDED;  // Else deliberate fall-through to B.
1229          case 1: case 3: {
1230            // B
1231            // |111|11|1|0000|000000|11|1 |1|1 |10000000000|
1232            // |5 3|21|0|9876|543  0|54|3 |2|1 |0    5    0|
1233            // |---|--|-|----|------|--|--|-|--|-----------|
1234            // |332|22|2|2222|221111|11|1 |1|1 |10000000000|
1235            // |1 9|87|6|5  2|10   6|54|3 |2|1 |0    5    0|
1236            // |---|--|-|----|------|--|--|-|--|-----------|
1237            // |111|10|S|cond| imm6 |10|J1|0|J2| imm11     |
1238            // |111|10|S| imm10     |10|J1|1|J2| imm11     |
1239            uint32_t S = (instr >> 26) & 1;
1240            uint32_t cond = (instr >> 22) & 0xF;
1241            uint32_t J2 = (instr >> 11) & 1;
1242            uint32_t form = (instr >> 12) & 1;
1243            uint32_t J1 = (instr >> 13) & 1;
1244            uint32_t imm10 = (instr >> 16) & 0x3FF;
1245            uint32_t imm6  = (instr >> 16) & 0x3F;
1246            uint32_t imm11 = instr & 0x7FF;
1247            opcode << "b";
1248            int32_t imm32;
1249            if (form == 0) {
1250              DumpCond(opcode, cond);
1251              imm32 = (S << 20) | (J2 << 19) | (J1 << 18) | (imm6 << 12) | (imm11 << 1);
1252              imm32 = (imm32 << 11) >> 11;  // sign extend 21 bit immediate.
1253            } else {
1254              uint32_t I1 = ~(J1 ^ S);
1255              uint32_t I2 = ~(J2 ^ S);
1256              imm32 = (S << 24) | (I1 << 23) | (I2 << 22) | (imm10 << 12) | (imm11 << 1);
1257              imm32 = (imm32 << 8) >> 8;  // sign extend 24 bit immediate.
1258            }
1259            opcode << ".w";
1260            DumpBranchTarget(args, instr_ptr + 4, imm32);
1261            break;
1262          }
1263          case 4: case 6: case 5: case 7: {
1264            // BL, BLX (immediate)
1265            // |111|11|1|0000000000|11|1 |1|1 |10000000000|
1266            // |5 3|21|0|9876543  0|54|3 |2|1 |0    5    0|
1267            // |---|--|-|----------|--|--|-|--|-----------|
1268            // |332|22|2|2222221111|11|1 |1|1 |10000000000|
1269            // |1 9|87|6|5    0   6|54|3 |2|1 |0    5    0|
1270            // |---|--|-|----------|--|--|-|--|-----------|
1271            // |111|10|S| imm10    |11|J1|L|J2| imm11     |
1272            uint32_t S = (instr >> 26) & 1;
1273            uint32_t J2 = (instr >> 11) & 1;
1274            uint32_t L = (instr >> 12) & 1;
1275            uint32_t J1 = (instr >> 13) & 1;
1276            uint32_t imm10 = (instr >> 16) & 0x3FF;
1277            uint32_t imm11 = instr & 0x7FF;
1278            if (L == 0) {
1279              opcode << "bx";
1280            } else {
1281              opcode << "blx";
1282            }
1283            uint32_t I1 = ~(J1 ^ S);
1284            uint32_t I2 = ~(J2 ^ S);
1285            int32_t imm32 = (S << 24) | (I1 << 23) | (I2 << 22) | (imm10 << 12) | (imm11 << 1);
1286            imm32 = (imm32 << 8) >> 8;  // sign extend 24 bit immediate.
1287            DumpBranchTarget(args, instr_ptr + 4, imm32);
1288            break;
1289          }
1290        }
1291      }
1292      break;
1293    case 3:
1294      switch (op2) {
1295        case 0x07: case 0x0F: case 0x17: case 0x1F: {  // Explicitly UNDEFINED, A6.3.
1296          opcode << "UNDEFINED";
1297          break;
1298        }
1299        case 0x06: case 0x0E: {  // "Store single data item" undefined opcodes, A6.3.10.
1300          opcode << "UNDEFINED [store]";
1301          break;
1302        }
1303        case 0x15: case 0x1D: {  // "Load word" undefined opcodes, A6.3.7.
1304          opcode << "UNDEFINED [load]";
1305          break;
1306        }
1307        case 0x10: case 0x12: case 0x14: case 0x16: case 0x18: case 0x1A: case 0x1C: case 0x1E: {
1308          opcode << "UNKNOWN " << op2 << " [SIMD]";
1309          break;
1310        }
1311        case 0x01: case 0x00: case 0x09: case 0x08:   // {LD,ST}RB{,T}
1312        case 0x03: case 0x02: case 0x0B: case 0x0A:   // {LD,ST}RH{,T}
1313        case 0x05: case 0x04: case 0x0D: case 0x0C:   // {LD,ST}R{,T}
1314        case 0x11:            case 0x19:              // LDRSB{,T} (no signed store)
1315        case 0x13:            case 0x1B: {            // LDRSH{,T} (no signed store)
1316          // Load:
1317          // (Store is the same except that l==0 and always s==0 below.)
1318          //                       00s.whl (sign, word, half, load)
1319          // LDR{S}B  imm12: 11111|00s1001| Rn | Rt |imm12             (0x09)
1320          // LDR{S}B   imm8: 11111|00s0001| Rn | Rt |1PUW|imm8         (0x01)
1321          // LDR{S}BT  imm8: 11111|00s0001| Rn | Rt |1110|imm8         (0x01)
1322          // LDR{S}B    lit: 11111|00sU001|1111| Rt |imm12             (0x01/0x09)
1323          // LDR{S}B    reg: 11111|00s0001| Rn | Rt |000000|imm2| Rm   (0x01)
1324          // LDR{S}H  imm12: 11111|00s1011| Rn | Rt |imm12             (0x0B)
1325          // LDR{S}H   imm8: 11111|00s0011| Rn | Rt |1PUW|imm8         (0x03)
1326          // LDR{S}HT  imm8: 11111|00s0011| Rn | Rt |1110|imm8         (0x03)
1327          // LDR{S}H    lit: 11111|00sU011|1111| Rt |imm12             (0x03/0x0B)
1328          // LDR{S}H    reg: 11111|00s0011| Rn | Rt |000000|imm2| Rm   (0x03)
1329          // LDR      imm12: 11111|0001101| Rn | Rt |imm12             (0x0D)
1330          // LDR       imm8: 11111|0000101| Rn | Rt |1PUW|imm8         (0x05)
1331          // LDRT      imm8: 11111|0000101| Rn | Rt |1110|imm8         (0x05)
1332          // LDR        lit: 11111|000U101|1111| Rt |imm12             (0x05/0x0D)
1333          // LDR        reg: 11111|0000101| Rn | Rt |000000|imm2| Rm   (0x05)
1334          //
1335          // If Rt == 15, instead of load we have preload:
1336          // PLD{W}   imm12: 11111|00010W1| Rn |1111|imm12             (0x09/0x0B)
1337          // PLD{W}    imm8: 11111|00000W1| Rn |1111|1100|imm8         (0x01/0x03); -imm8
1338          // PLD        lit: 11111|000U001|1111|1111|imm12             (0x01/0x09)
1339          // PLD{W}     reg: 11111|00000W1| Rn |1111|000000|imm2| Rm   (0x01/0x03)
1340          // PLI      imm12: 11111|0011001| Rn |1111|imm12             (0x19)
1341          // PLI       imm8: 11111|0010001| Rn |1111|1100|imm8         (0x11); -imm8
1342          // PLI        lit: 11111|001U001|1111|1111|imm12             (0x01/0x09)
1343          // PLI        reg: 11111|0010001| Rn |1111|000000|imm2| Rm   (0x01/0x03)
1344
1345          bool is_load = HasBitSet(instr, 20);
1346          bool is_half = HasBitSet(instr, 21);  // W for PLD/PLDW.
1347          bool is_word = HasBitSet(instr, 22);
1348          bool is_signed = HasBitSet(instr, 24);
1349          ArmRegister Rn(instr, 16);
1350          ArmRegister Rt(instr, 12);
1351          uint32_t imm12 = instr & 0xFFF;
1352          uint32_t U = (instr >> 23) & 1;  // U for imm12
1353          uint32_t imm8 = instr & 0xFF;
1354          uint32_t op4 = (instr >> 8) & 0xF;  // 1PUW for imm8
1355          if (Rt.r == PC && is_load && !is_word) {
1356            // PLD, PLDW, PLI
1357            const char* pld_pli = (is_signed ? "pli" : "pld");
1358            const char* w = (is_half ? "w" : "");
1359            if (is_signed && !is_half) {
1360              opcode << "UNDEFINED [PLI+W]";
1361            } else if (Rn.r == PC || U != 0u) {
1362              opcode << pld_pli << w;
1363              args << "[" << Rn << ", #" << (U != 0u ? "" : "-") << imm12 << "]";
1364              if (Rn.r == PC && is_half) {
1365                args << " (UNPREDICTABLE)";
1366              }
1367            } else if ((instr & 0xFC0) == 0) {
1368              opcode << pld_pli << w;
1369              RmLslImm2 Rm(instr);
1370              args << "[" << Rn << ", " << Rm << "]";
1371            } else if (op4 == 0xC) {
1372              opcode << pld_pli << w;
1373              args << "[" << Rn << ", #-" << imm8 << "]";
1374            } else {
1375              opcode << "UNDEFINED [~" << pld_pli << "]";
1376            }
1377            break;
1378          }
1379          const char* ldr_str = is_load ? "ldr" : "str";
1380          const char* sign = is_signed ? "s" : "";
1381          const char* type = is_word ? "" : is_half ? "h" : "b";
1382          bool unpred = (Rt.r == SP && !is_word) || (Rt.r == PC && !is_load);
1383          if (Rn.r == PC && !is_load) {
1384            opcode << "UNDEFINED [STR-lit]";
1385            unpred = false;
1386          } else if (Rn.r == PC || U != 0u) {
1387            // Load/store with imm12 (load literal if Rn.r == PC; there's no store literal).
1388            opcode << ldr_str << sign << type << ".w";
1389            args << Rt << ", [" << Rn << ", #" << (U != 0u ? "" : "-") << imm12 << "]";
1390            if (Rn.r == TR && is_load) {
1391              args << "  ; ";
1392              Thread::DumpThreadOffset<4>(args, imm12);
1393            } else if (Rn.r == PC) {
1394              T2LitType lit_type[] = {
1395                  kT2LitUByte, kT2LitUHalf, kT2LitHexWord, kT2LitInvalid,
1396                  kT2LitUByte, kT2LitUHalf, kT2LitHexWord, kT2LitInvalid,
1397                  kT2LitSByte, kT2LitSHalf, kT2LitInvalid, kT2LitInvalid,
1398                  kT2LitSByte, kT2LitSHalf, kT2LitInvalid, kT2LitInvalid,
1399              };
1400              DCHECK_LT(op2 >> 1, arraysize(lit_type));
1401              DCHECK_NE(lit_type[op2 >> 1], kT2LitInvalid);
1402              DumpThumb2Literal(args, instr_ptr, U, imm12, lit_type[op2 >> 1]);
1403            }
1404          } else if ((instr & 0xFC0) == 0) {
1405            opcode << ldr_str << sign << type << ".w";
1406            RmLslImm2 Rm(instr);
1407            args << Rt << ", [" << Rn << ", " << Rm << "]";
1408            unpred = unpred || (Rm.rm.r == SP) || (Rm.rm.r == PC);
1409          } else if (is_word && Rn.r == SP && imm8 == 4 && op4 == (is_load ? 0xB : 0xD)) {
1410            opcode << (is_load ? "pop" : "push") << ".w";
1411            args << Rn;
1412            unpred = unpred || (Rn.r == SP);
1413          } else if ((op4 & 5) == 0) {
1414            opcode << "UNDEFINED [P = W = 0 for " << ldr_str << "]";
1415            unpred = false;
1416          } else {
1417            uint32_t P = (instr >> 10) & 1;
1418            U = (instr >> 9) & 1;
1419            uint32_t W = (instr >> 8) & 1;
1420            bool pre_index = (P != 0 && W == 1);
1421            bool post_index = (P == 0 && W == 1);
1422            const char* t = (P != 0 && U != 0 && W == 0) ? "t" : "";  // Unprivileged load/store?
1423            opcode << ldr_str << sign << type << t << ".w";
1424            args << Rt << ", [" << Rn << (post_index ? "]" : "") << ", #" << (U != 0 ? "" : "-")
1425                << imm8 << (post_index ? "" : "]") << (pre_index ? "!" : "");
1426            unpred = (W != 0 && Rn.r == Rt.r);
1427          }
1428          if (unpred) {
1429            args << " (UNPREDICTABLE)";
1430          }
1431          break;
1432        }
1433        case 0x29: {  // 0101001
1434          // |111|11|1000000|0000|1111|1100|00|0 0|0000|
1435          // |5 3|21|0     4|3  0|5  2|1  8|76|5 4|3  0|
1436          // |---|--|-------|----|----|----|--|---|----|
1437          // |332|22|2222222|1111|1111|1100|00|0 0|0000|
1438          // |1 9|87|6     0|9  6|5  2|1  8|76|5 4|3  0|
1439          // |---|--|-------|----|----|----|--|---|----|
1440          // |111|11|0101001| Rm |1111| Rd |11|op3| Rm |
1441          // REV   - 111 11 0101001 mmmm 1111 dddd 1000 mmmm
1442          // REV16 - 111 11 0101001 mmmm 1111 dddd 1001 mmmm
1443          // RBIT  - 111 11 0101001 mmmm 1111 dddd 1010 mmmm
1444          // REVSH - 111 11 0101001 mmmm 1111 dddd 1011 mmmm
1445          if ((instr & 0xf0c0) == 0xf080) {
1446            uint32_t op3 = (instr >> 4) & 3;
1447            opcode << kThumbReverseOperations[op3];
1448            ArmRegister Rm(instr, 0);
1449            ArmRegister Rd(instr, 8);
1450            args << Rd << ", " << Rm;
1451            ArmRegister Rm2(instr, 16);
1452            if (Rm.r != Rm2.r || Rm.r == 13 || Rm.r == 15 || Rd.r == 13 || Rd.r == 15) {
1453              args << " (UNPREDICTABLE)";
1454            }
1455          }  // else unknown instruction
1456          break;
1457        }
1458      default:      // more formats
1459        if ((op2 >> 4) == 2) {      // 010xxxx
1460          // data processing (register)
1461          if ((instr & 0x0080f0f0) == 0x0000f000) {
1462            // LSL, LSR, ASR, ROR
1463            uint32_t shift_op = (instr >> 21) & 3;
1464            uint32_t S = (instr >> 20) & 1;
1465            ArmRegister Rd(instr, 8);
1466            ArmRegister Rn(instr, 16);
1467            ArmRegister Rm(instr, 0);
1468            opcode << kThumb2ShiftOperations[shift_op] << (S != 0 ? "s" : "");
1469            args << Rd << ", " << Rn << ", " << Rm;
1470          }
1471        } else if ((op2 >> 3) == 6) {       // 0110xxx
1472          // Multiply, multiply accumulate, and absolute difference
1473          op1 = (instr >> 20) & 0x7;
1474          op2 = (instr >> 4) & 0x1;
1475          ArmRegister Ra(instr, 12);
1476          ArmRegister Rn(instr, 16);
1477          ArmRegister Rm(instr, 0);
1478          ArmRegister Rd(instr, 8);
1479          switch (op1) {
1480          case 0:
1481            if (op2 == 0) {
1482              if (Ra.r == 0xf) {
1483                opcode << "mul";
1484                args << Rd << ", " << Rn << ", " << Rm;
1485              } else {
1486                opcode << "mla";
1487                args << Rd << ", " << Rn << ", " << Rm << ", " << Ra;
1488              }
1489            } else {
1490              opcode << "mls";
1491              args << Rd << ", " << Rn << ", " << Rm << ", " << Ra;
1492            }
1493            break;
1494          case 1:
1495          case 2:
1496          case 3:
1497          case 4:
1498          case 5:
1499          case 6:
1500              break;        // do these sometime
1501          }
1502        } else if ((op2 >> 3) == 7) {       // 0111xxx
1503          // Long multiply, long multiply accumulate, and divide
1504          op1 = (instr >> 20) & 0x7;
1505          op2 = (instr >> 4) & 0xf;
1506          ArmRegister Rn(instr, 16);
1507          ArmRegister Rm(instr, 0);
1508          ArmRegister Rd(instr, 8);
1509          ArmRegister RdHi(instr, 8);
1510          ArmRegister RdLo(instr, 12);
1511          switch (op1) {
1512          case 0:
1513            opcode << "smull";
1514            args << RdLo << ", " << RdHi << ", " << Rn << ", " << Rm;
1515            break;
1516          case 1:
1517            opcode << "sdiv";
1518            args << Rd << ", " << Rn << ", " << Rm;
1519            break;
1520          case 2:
1521            opcode << "umull";
1522            args << RdLo << ", " << RdHi << ", " << Rn << ", " << Rm;
1523            break;
1524          case 3:
1525            opcode << "udiv";
1526            args << Rd << ", " << Rn << ", " << Rm;
1527            break;
1528          case 4:
1529          case 5:
1530          case 6:
1531            break;      // TODO: when we generate these...
1532          }
1533        }
1534      }
1535      break;
1536    default:
1537      break;
1538  }
1539
1540  // Apply any IT-block conditions to the opcode if necessary.
1541  if (!it_conditions_.empty()) {
1542    opcode << it_conditions_.back();
1543    it_conditions_.pop_back();
1544  }
1545  if (opcode.str().size() == 0) {
1546    opcode << "UNKNOWN " << op2;
1547  }
1548
1549  os << FormatInstructionPointer(instr_ptr)
1550     << StringPrintf(": %08x\t%-7s ", instr, opcode.str().c_str())
1551     << args.str() << '\n';
1552  return 4;
1553}  // NOLINT(readability/fn_size)
1554
1555size_t DisassemblerArm::DumpThumb16(std::ostream& os, const uint8_t* instr_ptr) {
1556  uint16_t instr = ReadU16(instr_ptr);
1557  bool is_32bit = ((instr & 0xF000) == 0xF000) || ((instr & 0xF800) == 0xE800);
1558  if (is_32bit) {
1559    return DumpThumb32(os, instr_ptr);
1560  } else {
1561    std::ostringstream opcode;
1562    std::ostringstream args;
1563    uint16_t opcode1 = instr >> 10;
1564    if (opcode1 < 0x10) {
1565      // shift (immediate), add, subtract, move, and compare
1566      uint16_t opcode2 = instr >> 9;
1567      switch (opcode2) {
1568        case 0x0: case 0x1: case 0x2: case 0x3: case 0x4: case 0x5: case 0x6: case 0x7:
1569        case 0x8: case 0x9: case 0xA: case 0xB: {
1570          // Logical shift left     - 00 000xx iii mmm ddd
1571          // Logical shift right    - 00 001xx iii mmm ddd
1572          // Arithmetic shift right - 00 010xx iii mmm ddd
1573          uint16_t imm5 = (instr >> 6) & 0x1F;
1574          ThumbRegister rm(instr, 3);
1575          ThumbRegister Rd(instr, 0);
1576          if (opcode2 <= 3) {
1577            opcode << "lsls";
1578          } else if (opcode2 <= 7) {
1579            opcode << "lsrs";
1580          } else {
1581            opcode << "asrs";
1582          }
1583          args << Rd << ", " << rm << ", #" << imm5;
1584          break;
1585        }
1586        case 0xC: case 0xD: case 0xE: case 0xF: {
1587          // Add register        - 00 01100 mmm nnn ddd
1588          // Sub register        - 00 01101 mmm nnn ddd
1589          // Add 3-bit immediate - 00 01110 iii nnn ddd
1590          // Sub 3-bit immediate - 00 01111 iii nnn ddd
1591          uint16_t imm3_or_Rm = (instr >> 6) & 7;
1592          ThumbRegister Rn(instr, 3);
1593          ThumbRegister Rd(instr, 0);
1594          if ((opcode2 & 2) != 0 && imm3_or_Rm == 0) {
1595            opcode << "mov";
1596          } else {
1597            if ((opcode2 & 1) == 0) {
1598              opcode << "adds";
1599            } else {
1600              opcode << "subs";
1601            }
1602          }
1603          args << Rd << ", " << Rn;
1604          if ((opcode2 & 2) == 0) {
1605            ArmRegister Rm(imm3_or_Rm);
1606            args << ", " << Rm;
1607          } else if (imm3_or_Rm != 0) {
1608            args << ", #" << imm3_or_Rm;
1609          }
1610          break;
1611        }
1612        case 0x10: case 0x11: case 0x12: case 0x13:
1613        case 0x14: case 0x15: case 0x16: case 0x17:
1614        case 0x18: case 0x19: case 0x1A: case 0x1B:
1615        case 0x1C: case 0x1D: case 0x1E: case 0x1F: {
1616          // MOVS Rd, #imm8 - 00100 ddd iiiiiiii
1617          // CMP  Rn, #imm8 - 00101 nnn iiiiiiii
1618          // ADDS Rn, #imm8 - 00110 nnn iiiiiiii
1619          // SUBS Rn, #imm8 - 00111 nnn iiiiiiii
1620          ThumbRegister Rn(instr, 8);
1621          uint16_t imm8 = instr & 0xFF;
1622          switch (opcode2 >> 2) {
1623            case 4: opcode << "movs"; break;
1624            case 5: opcode << "cmp"; break;
1625            case 6: opcode << "adds"; break;
1626            case 7: opcode << "subs"; break;
1627          }
1628          args << Rn << ", #" << imm8;
1629          break;
1630        }
1631        default:
1632          break;
1633      }
1634    } else if (opcode1 == 0x10) {
1635      // Data-processing
1636      uint16_t opcode2 = (instr >> 6) & 0xF;
1637      ThumbRegister rm(instr, 3);
1638      ThumbRegister rdn(instr, 0);
1639      opcode << kThumbDataProcessingOperations[opcode2];
1640      args << rdn << ", " << rm;
1641    } else if (opcode1 == 0x11) {
1642      // Special data instructions and branch and exchange
1643      uint16_t opcode2 = (instr >> 6) & 0x0F;
1644      switch (opcode2) {
1645        case 0x0: case 0x1: case 0x2: case 0x3: {
1646          // Add low registers  - 010001 0000 xxxxxx
1647          // Add high registers - 010001 0001/001x xxxxxx
1648          uint16_t DN = (instr >> 7) & 1;
1649          ArmRegister rm(instr, 3);
1650          uint16_t Rdn = instr & 7;
1651          ArmRegister DN_Rdn((DN << 3) | Rdn);
1652          opcode << "add";
1653          args << DN_Rdn << ", " << rm;
1654          break;
1655        }
1656        case 0x8: case 0x9: case 0xA: case 0xB: {
1657          // Move low registers  - 010001 1000 xxxxxx
1658          // Move high registers - 010001 1001/101x xxxxxx
1659          uint16_t DN = (instr >> 7) & 1;
1660          ArmRegister rm(instr, 3);
1661          uint16_t Rdn = instr & 7;
1662          ArmRegister DN_Rdn((DN << 3) | Rdn);
1663          opcode << "mov";
1664          args << DN_Rdn << ", " << rm;
1665          break;
1666        }
1667        case 0x5: case 0x6: case 0x7: {
1668          // Compare high registers - 010001 0101/011x xxxxxx
1669          uint16_t N = (instr >> 7) & 1;
1670          ArmRegister rm(instr, 3);
1671          uint16_t Rn = instr & 7;
1672          ArmRegister N_Rn((N << 3) | Rn);
1673          opcode << "cmp";
1674          args << N_Rn << ", " << rm;
1675          break;
1676        }
1677        case 0xC: case 0xD: case 0xE: case 0xF: {
1678          // Branch and exchange           - 010001 110x xxxxxx
1679          // Branch with link and exchange - 010001 111x xxxxxx
1680          ArmRegister rm(instr, 3);
1681          opcode << ((opcode2 & 0x2) == 0 ? "bx" : "blx");
1682          args << rm;
1683          break;
1684        }
1685        default:
1686          break;
1687      }
1688    } else if (opcode1 == 0x12 || opcode1 == 0x13) {  // 01001x
1689      ThumbRegister Rt(instr, 8);
1690      uint16_t imm8 = instr & 0xFF;
1691      opcode << "ldr";
1692      args << Rt << ", [pc, #" << (imm8 << 2) << "]";
1693    } else if ((opcode1 >= 0x14 && opcode1 <= 0x17) ||  // 0101xx
1694               (opcode1 >= 0x18 && opcode1 <= 0x1f) ||  // 011xxx
1695               (opcode1 >= 0x20 && opcode1 <= 0x27)) {  // 100xxx
1696      // Load/store single data item
1697      uint16_t opA = (instr >> 12) & 0xF;
1698      if (opA == 0x5) {
1699        uint16_t opB = (instr >> 9) & 0x7;
1700        ThumbRegister Rm(instr, 6);
1701        ThumbRegister Rn(instr, 3);
1702        ThumbRegister Rt(instr, 0);
1703        switch (opB) {
1704          case 0: opcode << "str"; break;
1705          case 1: opcode << "strh"; break;
1706          case 2: opcode << "strb"; break;
1707          case 3: opcode << "ldrsb"; break;
1708          case 4: opcode << "ldr"; break;
1709          case 5: opcode << "ldrh"; break;
1710          case 6: opcode << "ldrb"; break;
1711          case 7: opcode << "ldrsh"; break;
1712        }
1713        args << Rt << ", [" << Rn << ", " << Rm << "]";
1714      } else if (opA == 9) {
1715        uint16_t opB = (instr >> 11) & 1;
1716        ThumbRegister Rt(instr, 8);
1717        uint16_t imm8 = instr & 0xFF;
1718        opcode << (opB == 0 ? "str" : "ldr");
1719        args << Rt << ", [sp, #" << (imm8 << 2) << "]";
1720      } else {
1721        uint16_t imm5 = (instr >> 6) & 0x1F;
1722        uint16_t opB = (instr >> 11) & 1;
1723        ThumbRegister Rn(instr, 3);
1724        ThumbRegister Rt(instr, 0);
1725        switch (opA) {
1726          case 6:
1727            imm5 <<= 2;
1728            opcode << (opB == 0 ? "str" : "ldr");
1729            break;
1730          case 7:
1731            imm5 <<= 0;
1732            opcode << (opB == 0 ? "strb" : "ldrb");
1733            break;
1734          case 8:
1735            imm5 <<= 1;
1736            opcode << (opB == 0 ? "strh" : "ldrh");
1737            break;
1738        }
1739        args << Rt << ", [" << Rn << ", #" << imm5 << "]";
1740      }
1741    } else if (opcode1 >= 0x34 && opcode1 <= 0x37) {  // 1101xx
1742      int8_t imm8 = instr & 0xFF;
1743      uint32_t cond = (instr >> 8) & 0xF;
1744      opcode << "b";
1745      DumpCond(opcode, cond);
1746      DumpBranchTarget(args, instr_ptr + 4, (imm8 << 1));
1747    } else if ((instr & 0xF800) == 0xA800) {
1748      // Generate SP-relative address
1749      ThumbRegister rd(instr, 8);
1750      int imm8 = instr & 0xFF;
1751      opcode << "add";
1752      args << rd << ", sp, #" << (imm8 << 2);
1753    } else if ((instr & 0xF000) == 0xB000) {
1754      // Miscellaneous 16-bit instructions
1755      uint16_t opcode2 = (instr >> 5) & 0x7F;
1756      switch (opcode2) {
1757        case 0x00: case 0x01: case 0x02: case 0x03: case 0x04: case 0x05: case 0x06: case 0x07: {
1758          // Add immediate to SP        - 1011 00000 ii iiiii
1759          // Subtract immediate from SP - 1011 00001 ii iiiii
1760          int imm7 = instr & 0x7F;
1761          opcode << ((opcode2 & 4) == 0 ? "add" : "sub");
1762          args << "sp, sp, #" << (imm7 << 2);
1763          break;
1764        }
1765        case 0x08: case 0x09: case 0x0A: case 0x0B:  // 0001xxx
1766        case 0x0C: case 0x0D: case 0x0E: case 0x0F:
1767        case 0x18: case 0x19: case 0x1A: case 0x1B:  // 0011xxx
1768        case 0x1C: case 0x1D: case 0x1E: case 0x1F:
1769        case 0x48: case 0x49: case 0x4A: case 0x4B:  // 1001xxx
1770        case 0x4C: case 0x4D: case 0x4E: case 0x4F:
1771        case 0x58: case 0x59: case 0x5A: case 0x5B:  // 1011xxx
1772        case 0x5C: case 0x5D: case 0x5E: case 0x5F: {
1773          // CBNZ, CBZ
1774          uint16_t op = (instr >> 11) & 1;
1775          uint16_t i = (instr >> 9) & 1;
1776          uint16_t imm5 = (instr >> 3) & 0x1F;
1777          ThumbRegister Rn(instr, 0);
1778          opcode << (op != 0 ? "cbnz" : "cbz");
1779          uint32_t imm32 = (i << 6) | (imm5 << 1);
1780          args << Rn << ", ";
1781          DumpBranchTarget(args, instr_ptr + 4, imm32);
1782          break;
1783        }
1784        case 0x20: case 0x21: case 0x22: case 0x23: case 0x24: case 0x25: case 0x26: case 0x27:
1785        case 0x28: case 0x29: case 0x2A: case 0x2B: case 0x2C: case 0x2D: case 0x2E: case 0x2F: {
1786          opcode << "push";
1787          args << RegisterList((instr & 0xFF) | ((instr & 0x100) << 6));
1788          break;
1789        }
1790        case 0x60: case 0x61: case 0x62: case 0x63: case 0x64: case 0x65: case 0x66: case 0x67:
1791        case 0x68: case 0x69: case 0x6A: case 0x6B: case 0x6C: case 0x6D: case 0x6E: case 0x6F: {
1792          opcode << "pop";
1793          args << RegisterList((instr & 0xFF) | ((instr & 0x100) << 7));
1794          break;
1795        }
1796        case 0x70: case 0x71: case 0x72: case 0x73: case 0x74: case 0x75: case 0x76: case 0x77: {
1797          opcode << "bkpt";
1798          args << "#" << (instr & 0xFF);
1799          break;
1800        }
1801        case 0x50: case 0x51:    // 101000x
1802        case 0x52: case 0x53:    // 101001x
1803        case 0x56: case 0x57: {  // 101011x
1804          uint16_t op = (instr >> 6) & 3;
1805          opcode << kThumbReverseOperations[op];
1806          ThumbRegister Rm(instr, 3);
1807          ThumbRegister Rd(instr, 0);
1808          args << Rd << ", " << Rm;
1809          break;
1810        }
1811        case 0x78: case 0x79: case 0x7A: case 0x7B:  // 1111xxx
1812        case 0x7C: case 0x7D: case 0x7E: case 0x7F: {
1813          // If-Then, and hints
1814          uint16_t opA = (instr >> 4) & 0xF;
1815          uint16_t opB = instr & 0xF;
1816          if (opB == 0) {
1817            switch (opA) {
1818              case 0: opcode << "nop"; break;
1819              case 1: opcode << "yield"; break;
1820              case 2: opcode << "wfe";  break;
1821              case 3: opcode << "sev"; break;
1822              default: break;
1823            }
1824          } else {
1825            uint32_t first_cond = opA;
1826            uint32_t mask = opB;
1827            opcode << "it";
1828
1829            // Flesh out the base "it" opcode with the specific collection of 't's and 'e's,
1830            // and store up the actual condition codes we'll want to add to the next few opcodes.
1831            size_t count = 3 - CTZ(mask);
1832            it_conditions_.resize(count + 2);  // Plus the implicit 't', plus the "" for the IT itself.
1833            for (size_t i = 0; i < count; ++i) {
1834              bool positive_cond = ((first_cond & 1) != 0);
1835              bool positive_mask = ((mask & (1 << (3 - i))) != 0);
1836              if (positive_mask == positive_cond) {
1837                opcode << 't';
1838                it_conditions_[i] = kConditionCodeNames[first_cond];
1839              } else {
1840                opcode << 'e';
1841                it_conditions_[i] = kConditionCodeNames[first_cond ^ 1];
1842              }
1843            }
1844            it_conditions_[count] = kConditionCodeNames[first_cond];  // The implicit 't'.
1845
1846            it_conditions_[count + 1] = "";  // No condition code for the IT itself...
1847            DumpCond(args, first_cond);  // ...because it's considered an argument.
1848          }
1849          break;
1850        }
1851        default:
1852          break;
1853      }
1854    } else if (((instr & 0xF000) == 0x5000) || ((instr & 0xE000) == 0x6000) ||
1855        ((instr & 0xE000) == 0x8000)) {
1856      // Load/store single data item
1857      uint16_t opA = instr >> 12;
1858      // uint16_t opB = (instr >> 9) & 7;
1859      switch (opA) {
1860        case 0x6: {
1861          // STR Rt, [Rn, #imm] - 01100 iiiii nnn ttt
1862          // LDR Rt, [Rn, #imm] - 01101 iiiii nnn ttt
1863          uint16_t imm5 = (instr >> 6) & 0x1F;
1864          ThumbRegister Rn(instr, 3);
1865          ThumbRegister Rt(instr, 0);
1866          opcode << ((instr & 0x800) == 0 ? "str" : "ldr");
1867          args << Rt << ", [" << Rn << ", #" << (imm5 << 2) << "]";
1868          break;
1869        }
1870        case 0x9: {
1871          // STR Rt, [SP, #imm] - 01100 ttt iiiiiiii
1872          // LDR Rt, [SP, #imm] - 01101 ttt iiiiiiii
1873          uint16_t imm8 = instr & 0xFF;
1874          ThumbRegister Rt(instr, 8);
1875          opcode << ((instr & 0x800) == 0 ? "str" : "ldr");
1876          args << Rt << ", [sp, #" << (imm8 << 2) << "]";
1877          break;
1878        }
1879        default:
1880          break;
1881      }
1882    } else if (opcode1 == 0x38 || opcode1 == 0x39) {
1883      uint16_t imm11 = instr & 0x7FFF;
1884      int32_t imm32 = imm11 << 1;
1885      imm32 = (imm32 << 20) >> 20;  // sign extend 12 bit immediate
1886      opcode << "b";
1887      DumpBranchTarget(args, instr_ptr + 4, imm32);
1888    }
1889
1890    // Apply any IT-block conditions to the opcode if necessary.
1891    if (!it_conditions_.empty()) {
1892      opcode << it_conditions_.back();
1893      it_conditions_.pop_back();
1894    }
1895
1896    os << FormatInstructionPointer(instr_ptr)
1897       << StringPrintf(": %04x    \t%-7s ", instr, opcode.str().c_str())
1898       << args.str() << '\n';
1899  }
1900  return 2;
1901}
1902
1903}  // namespace arm
1904}  // namespace art
1905