1 //===-- DisassemblerLLVMC.cpp -----------------------------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "DisassemblerLLVMC.h"
10 
11 #include "llvm-c/Disassembler.h"
12 #include "llvm/ADT/SmallString.h"
13 #include "llvm/MC/MCAsmInfo.h"
14 #include "llvm/MC/MCContext.h"
15 #include "llvm/MC/MCDisassembler/MCDisassembler.h"
16 #include "llvm/MC/MCDisassembler/MCExternalSymbolizer.h"
17 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h"
18 #include "llvm/MC/MCInst.h"
19 #include "llvm/MC/MCInstPrinter.h"
20 #include "llvm/MC/MCInstrInfo.h"
21 #include "llvm/MC/MCRegisterInfo.h"
22 #include "llvm/MC/MCSubtargetInfo.h"
23 #include "llvm/MC/MCTargetOptions.h"
24 #include "llvm/Support/ErrorHandling.h"
25 #include "llvm/Support/ScopedPrinter.h"
26 #include "llvm/Support/TargetRegistry.h"
27 #include "llvm/Support/TargetSelect.h"
28 
29 #include "lldb/Core/Address.h"
30 #include "lldb/Core/Module.h"
31 #include "lldb/Symbol/SymbolContext.h"
32 #include "lldb/Target/ExecutionContext.h"
33 #include "lldb/Target/Process.h"
34 #include "lldb/Target/RegisterContext.h"
35 #include "lldb/Target/SectionLoadList.h"
36 #include "lldb/Target/StackFrame.h"
37 #include "lldb/Target/Target.h"
38 #include "lldb/Utility/DataExtractor.h"
39 #include "lldb/Utility/Log.h"
40 #include "lldb/Utility/RegularExpression.h"
41 #include "lldb/Utility/Stream.h"
42 
43 using namespace lldb;
44 using namespace lldb_private;
45 
46 class DisassemblerLLVMC::MCDisasmInstance {
47 public:
48   static std::unique_ptr<MCDisasmInstance>
49   Create(const char *triple, const char *cpu, const char *features_str,
50          unsigned flavor, DisassemblerLLVMC &owner);
51 
52   ~MCDisasmInstance() = default;
53 
54   uint64_t GetMCInst(const uint8_t *opcode_data, size_t opcode_data_len,
55                      lldb::addr_t pc, llvm::MCInst &mc_inst) const;
56   void PrintMCInst(llvm::MCInst &mc_inst, std::string &inst_string,
57                    std::string &comments_string);
58   void SetStyle(bool use_hex_immed, HexImmediateStyle hex_style);
59   bool CanBranch(llvm::MCInst &mc_inst) const;
60   bool HasDelaySlot(llvm::MCInst &mc_inst) const;
61   bool IsCall(llvm::MCInst &mc_inst) const;
62 
63 private:
64   MCDisasmInstance(std::unique_ptr<llvm::MCInstrInfo> &&instr_info_up,
65                    std::unique_ptr<llvm::MCRegisterInfo> &&reg_info_up,
66                    std::unique_ptr<llvm::MCSubtargetInfo> &&subtarget_info_up,
67                    std::unique_ptr<llvm::MCAsmInfo> &&asm_info_up,
68                    std::unique_ptr<llvm::MCContext> &&context_up,
69                    std::unique_ptr<llvm::MCDisassembler> &&disasm_up,
70                    std::unique_ptr<llvm::MCInstPrinter> &&instr_printer_up);
71 
72   std::unique_ptr<llvm::MCInstrInfo> m_instr_info_up;
73   std::unique_ptr<llvm::MCRegisterInfo> m_reg_info_up;
74   std::unique_ptr<llvm::MCSubtargetInfo> m_subtarget_info_up;
75   std::unique_ptr<llvm::MCAsmInfo> m_asm_info_up;
76   std::unique_ptr<llvm::MCContext> m_context_up;
77   std::unique_ptr<llvm::MCDisassembler> m_disasm_up;
78   std::unique_ptr<llvm::MCInstPrinter> m_instr_printer_up;
79 };
80 
81 class InstructionLLVMC : public lldb_private::Instruction {
82 public:
InstructionLLVMC(DisassemblerLLVMC & disasm,const lldb_private::Address & address,AddressClass addr_class)83   InstructionLLVMC(DisassemblerLLVMC &disasm,
84                    const lldb_private::Address &address,
85                    AddressClass addr_class)
86       : Instruction(address, addr_class),
87         m_disasm_wp(std::static_pointer_cast<DisassemblerLLVMC>(
88             disasm.shared_from_this())),
89         m_does_branch(eLazyBoolCalculate), m_has_delay_slot(eLazyBoolCalculate),
90         m_is_call(eLazyBoolCalculate), m_is_valid(false),
91         m_using_file_addr(false) {}
92 
93   ~InstructionLLVMC() override = default;
94 
DoesBranch()95   bool DoesBranch() override {
96     if (m_does_branch == eLazyBoolCalculate) {
97       DisassemblerScope disasm(*this);
98       if (disasm) {
99         DataExtractor data;
100         if (m_opcode.GetData(data)) {
101           bool is_alternate_isa;
102           lldb::addr_t pc = m_address.GetFileAddress();
103 
104           DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
105               GetDisasmToUse(is_alternate_isa, disasm);
106           const uint8_t *opcode_data = data.GetDataStart();
107           const size_t opcode_data_len = data.GetByteSize();
108           llvm::MCInst inst;
109           const size_t inst_size =
110               mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
111           // Be conservative, if we didn't understand the instruction, say it
112           // might branch...
113           if (inst_size == 0)
114             m_does_branch = eLazyBoolYes;
115           else {
116             const bool can_branch = mc_disasm_ptr->CanBranch(inst);
117             if (can_branch)
118               m_does_branch = eLazyBoolYes;
119             else
120               m_does_branch = eLazyBoolNo;
121           }
122         }
123       }
124     }
125     return m_does_branch == eLazyBoolYes;
126   }
127 
HasDelaySlot()128   bool HasDelaySlot() override {
129     if (m_has_delay_slot == eLazyBoolCalculate) {
130       DisassemblerScope disasm(*this);
131       if (disasm) {
132         DataExtractor data;
133         if (m_opcode.GetData(data)) {
134           bool is_alternate_isa;
135           lldb::addr_t pc = m_address.GetFileAddress();
136 
137           DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
138               GetDisasmToUse(is_alternate_isa, disasm);
139           const uint8_t *opcode_data = data.GetDataStart();
140           const size_t opcode_data_len = data.GetByteSize();
141           llvm::MCInst inst;
142           const size_t inst_size =
143               mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
144           // if we didn't understand the instruction, say it doesn't have a
145           // delay slot...
146           if (inst_size == 0)
147             m_has_delay_slot = eLazyBoolNo;
148           else {
149             const bool has_delay_slot = mc_disasm_ptr->HasDelaySlot(inst);
150             if (has_delay_slot)
151               m_has_delay_slot = eLazyBoolYes;
152             else
153               m_has_delay_slot = eLazyBoolNo;
154           }
155         }
156       }
157     }
158     return m_has_delay_slot == eLazyBoolYes;
159   }
160 
GetDisasmToUse(bool & is_alternate_isa)161   DisassemblerLLVMC::MCDisasmInstance *GetDisasmToUse(bool &is_alternate_isa) {
162     DisassemblerScope disasm(*this);
163     return GetDisasmToUse(is_alternate_isa, disasm);
164   }
165 
Decode(const lldb_private::Disassembler & disassembler,const lldb_private::DataExtractor & data,lldb::offset_t data_offset)166   size_t Decode(const lldb_private::Disassembler &disassembler,
167                 const lldb_private::DataExtractor &data,
168                 lldb::offset_t data_offset) override {
169     // All we have to do is read the opcode which can be easy for some
170     // architectures
171     bool got_op = false;
172     DisassemblerScope disasm(*this);
173     if (disasm) {
174       const ArchSpec &arch = disasm->GetArchitecture();
175       const lldb::ByteOrder byte_order = data.GetByteOrder();
176 
177       const uint32_t min_op_byte_size = arch.GetMinimumOpcodeByteSize();
178       const uint32_t max_op_byte_size = arch.GetMaximumOpcodeByteSize();
179       if (min_op_byte_size == max_op_byte_size) {
180         // Fixed size instructions, just read that amount of data.
181         if (!data.ValidOffsetForDataOfSize(data_offset, min_op_byte_size))
182           return false;
183 
184         switch (min_op_byte_size) {
185         case 1:
186           m_opcode.SetOpcode8(data.GetU8(&data_offset), byte_order);
187           got_op = true;
188           break;
189 
190         case 2:
191           m_opcode.SetOpcode16(data.GetU16(&data_offset), byte_order);
192           got_op = true;
193           break;
194 
195         case 4:
196           m_opcode.SetOpcode32(data.GetU32(&data_offset), byte_order);
197           got_op = true;
198           break;
199 
200         case 8:
201           m_opcode.SetOpcode64(data.GetU64(&data_offset), byte_order);
202           got_op = true;
203           break;
204 
205         default:
206           m_opcode.SetOpcodeBytes(data.PeekData(data_offset, min_op_byte_size),
207                                   min_op_byte_size);
208           got_op = true;
209           break;
210         }
211       }
212       if (!got_op) {
213         bool is_alternate_isa = false;
214         DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
215             GetDisasmToUse(is_alternate_isa, disasm);
216 
217         const llvm::Triple::ArchType machine = arch.GetMachine();
218         if (machine == llvm::Triple::arm || machine == llvm::Triple::thumb) {
219           if (machine == llvm::Triple::thumb || is_alternate_isa) {
220             uint32_t thumb_opcode = data.GetU16(&data_offset);
221             if ((thumb_opcode & 0xe000) != 0xe000 ||
222                 ((thumb_opcode & 0x1800u) == 0)) {
223               m_opcode.SetOpcode16(thumb_opcode, byte_order);
224               m_is_valid = true;
225             } else {
226               thumb_opcode <<= 16;
227               thumb_opcode |= data.GetU16(&data_offset);
228               m_opcode.SetOpcode16_2(thumb_opcode, byte_order);
229               m_is_valid = true;
230             }
231           } else {
232             m_opcode.SetOpcode32(data.GetU32(&data_offset), byte_order);
233             m_is_valid = true;
234           }
235         } else {
236           // The opcode isn't evenly sized, so we need to actually use the llvm
237           // disassembler to parse it and get the size.
238           uint8_t *opcode_data =
239               const_cast<uint8_t *>(data.PeekData(data_offset, 1));
240           const size_t opcode_data_len = data.BytesLeft(data_offset);
241           const addr_t pc = m_address.GetFileAddress();
242           llvm::MCInst inst;
243 
244           const size_t inst_size =
245               mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
246           if (inst_size == 0)
247             m_opcode.Clear();
248           else {
249             m_opcode.SetOpcodeBytes(opcode_data, inst_size);
250             m_is_valid = true;
251           }
252         }
253       }
254       return m_opcode.GetByteSize();
255     }
256     return 0;
257   }
258 
AppendComment(std::string & description)259   void AppendComment(std::string &description) {
260     if (m_comment.empty())
261       m_comment.swap(description);
262     else {
263       m_comment.append(", ");
264       m_comment.append(description);
265     }
266   }
267 
CalculateMnemonicOperandsAndComment(const lldb_private::ExecutionContext * exe_ctx)268   void CalculateMnemonicOperandsAndComment(
269       const lldb_private::ExecutionContext *exe_ctx) override {
270     DataExtractor data;
271     const AddressClass address_class = GetAddressClass();
272 
273     if (m_opcode.GetData(data)) {
274       std::string out_string;
275       std::string comment_string;
276 
277       DisassemblerScope disasm(*this, exe_ctx);
278       if (disasm) {
279         DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr;
280 
281         if (address_class == AddressClass::eCodeAlternateISA)
282           mc_disasm_ptr = disasm->m_alternate_disasm_up.get();
283         else
284           mc_disasm_ptr = disasm->m_disasm_up.get();
285 
286         lldb::addr_t pc = m_address.GetFileAddress();
287         m_using_file_addr = true;
288 
289         const bool data_from_file = disasm->m_data_from_file;
290         bool use_hex_immediates = true;
291         Disassembler::HexImmediateStyle hex_style = Disassembler::eHexStyleC;
292 
293         if (exe_ctx) {
294           Target *target = exe_ctx->GetTargetPtr();
295           if (target) {
296             use_hex_immediates = target->GetUseHexImmediates();
297             hex_style = target->GetHexImmediateStyle();
298 
299             if (!data_from_file) {
300               const lldb::addr_t load_addr = m_address.GetLoadAddress(target);
301               if (load_addr != LLDB_INVALID_ADDRESS) {
302                 pc = load_addr;
303                 m_using_file_addr = false;
304               }
305             }
306           }
307         }
308 
309         const uint8_t *opcode_data = data.GetDataStart();
310         const size_t opcode_data_len = data.GetByteSize();
311         llvm::MCInst inst;
312         size_t inst_size =
313             mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
314 
315         if (inst_size > 0) {
316           mc_disasm_ptr->SetStyle(use_hex_immediates, hex_style);
317           mc_disasm_ptr->PrintMCInst(inst, out_string, comment_string);
318 
319           if (!comment_string.empty()) {
320             AppendComment(comment_string);
321           }
322         }
323 
324         if (inst_size == 0) {
325           m_comment.assign("unknown opcode");
326           inst_size = m_opcode.GetByteSize();
327           StreamString mnemonic_strm;
328           lldb::offset_t offset = 0;
329           lldb::ByteOrder byte_order = data.GetByteOrder();
330           switch (inst_size) {
331           case 1: {
332             const uint8_t uval8 = data.GetU8(&offset);
333             m_opcode.SetOpcode8(uval8, byte_order);
334             m_opcode_name.assign(".byte");
335             mnemonic_strm.Printf("0x%2.2x", uval8);
336           } break;
337           case 2: {
338             const uint16_t uval16 = data.GetU16(&offset);
339             m_opcode.SetOpcode16(uval16, byte_order);
340             m_opcode_name.assign(".short");
341             mnemonic_strm.Printf("0x%4.4x", uval16);
342           } break;
343           case 4: {
344             const uint32_t uval32 = data.GetU32(&offset);
345             m_opcode.SetOpcode32(uval32, byte_order);
346             m_opcode_name.assign(".long");
347             mnemonic_strm.Printf("0x%8.8x", uval32);
348           } break;
349           case 8: {
350             const uint64_t uval64 = data.GetU64(&offset);
351             m_opcode.SetOpcode64(uval64, byte_order);
352             m_opcode_name.assign(".quad");
353             mnemonic_strm.Printf("0x%16.16" PRIx64, uval64);
354           } break;
355           default:
356             if (inst_size == 0)
357               return;
358             else {
359               const uint8_t *bytes = data.PeekData(offset, inst_size);
360               if (bytes == nullptr)
361                 return;
362               m_opcode_name.assign(".byte");
363               m_opcode.SetOpcodeBytes(bytes, inst_size);
364               mnemonic_strm.Printf("0x%2.2x", bytes[0]);
365               for (uint32_t i = 1; i < inst_size; ++i)
366                 mnemonic_strm.Printf(" 0x%2.2x", bytes[i]);
367             }
368             break;
369           }
370           m_mnemonics = mnemonic_strm.GetString();
371           return;
372         } else {
373           if (m_does_branch == eLazyBoolCalculate) {
374             const bool can_branch = mc_disasm_ptr->CanBranch(inst);
375             if (can_branch)
376               m_does_branch = eLazyBoolYes;
377             else
378               m_does_branch = eLazyBoolNo;
379           }
380         }
381 
382         static RegularExpression s_regex(
383             llvm::StringRef("[ \t]*([^ ^\t]+)[ \t]*([^ ^\t].*)?"));
384 
385         llvm::SmallVector<llvm::StringRef, 4> matches;
386         if (s_regex.Execute(out_string, &matches)) {
387           m_opcode_name = matches[1].str();
388           m_mnemonics = matches[2].str();
389         }
390       }
391     }
392   }
393 
IsValid() const394   bool IsValid() const { return m_is_valid; }
395 
UsingFileAddress() const396   bool UsingFileAddress() const { return m_using_file_addr; }
GetByteSize() const397   size_t GetByteSize() const { return m_opcode.GetByteSize(); }
398 
399   /// Grants exclusive access to the disassembler and initializes it with the
400   /// given InstructionLLVMC and an optional ExecutionContext.
401   class DisassemblerScope {
402     std::shared_ptr<DisassemblerLLVMC> m_disasm;
403 
404   public:
DisassemblerScope(InstructionLLVMC & i,const lldb_private::ExecutionContext * exe_ctx=nullptr)405     explicit DisassemblerScope(
406         InstructionLLVMC &i,
407         const lldb_private::ExecutionContext *exe_ctx = nullptr)
408         : m_disasm(i.m_disasm_wp.lock()) {
409       m_disasm->m_mutex.lock();
410       m_disasm->m_inst = &i;
411       m_disasm->m_exe_ctx = exe_ctx;
412     }
~DisassemblerScope()413     ~DisassemblerScope() { m_disasm->m_mutex.unlock(); }
414 
415     /// Evaluates to true if this scope contains a valid disassembler.
operator bool() const416     operator bool() const { return static_cast<bool>(m_disasm); }
417 
operator ->()418     std::shared_ptr<DisassemblerLLVMC> operator->() { return m_disasm; }
419   };
420 
421   static llvm::StringRef::const_iterator
ConsumeWhitespace(llvm::StringRef::const_iterator osi,llvm::StringRef::const_iterator ose)422   ConsumeWhitespace(llvm::StringRef::const_iterator osi,
423                     llvm::StringRef::const_iterator ose) {
424     while (osi != ose) {
425       switch (*osi) {
426       default:
427         return osi;
428       case ' ':
429       case '\t':
430         break;
431       }
432       ++osi;
433     }
434 
435     return osi;
436   }
437 
438   static std::pair<bool, llvm::StringRef::const_iterator>
ConsumeChar(llvm::StringRef::const_iterator osi,const char c,llvm::StringRef::const_iterator ose)439   ConsumeChar(llvm::StringRef::const_iterator osi, const char c,
440               llvm::StringRef::const_iterator ose) {
441     bool found = false;
442 
443     osi = ConsumeWhitespace(osi, ose);
444     if (osi != ose && *osi == c) {
445       found = true;
446       ++osi;
447     }
448 
449     return std::make_pair(found, osi);
450   }
451 
452   static std::pair<Operand, llvm::StringRef::const_iterator>
ParseRegisterName(llvm::StringRef::const_iterator osi,llvm::StringRef::const_iterator ose)453   ParseRegisterName(llvm::StringRef::const_iterator osi,
454                     llvm::StringRef::const_iterator ose) {
455     Operand ret;
456     ret.m_type = Operand::Type::Register;
457     std::string str;
458 
459     osi = ConsumeWhitespace(osi, ose);
460 
461     while (osi != ose) {
462       if (*osi >= '0' && *osi <= '9') {
463         if (str.empty()) {
464           return std::make_pair(Operand(), osi);
465         } else {
466           str.push_back(*osi);
467         }
468       } else if (*osi >= 'a' && *osi <= 'z') {
469         str.push_back(*osi);
470       } else {
471         switch (*osi) {
472         default:
473           if (str.empty()) {
474             return std::make_pair(Operand(), osi);
475           } else {
476             ret.m_register = ConstString(str);
477             return std::make_pair(ret, osi);
478           }
479         case '%':
480           if (!str.empty()) {
481             return std::make_pair(Operand(), osi);
482           }
483           break;
484         }
485       }
486       ++osi;
487     }
488 
489     ret.m_register = ConstString(str);
490     return std::make_pair(ret, osi);
491   }
492 
493   static std::pair<Operand, llvm::StringRef::const_iterator>
ParseImmediate(llvm::StringRef::const_iterator osi,llvm::StringRef::const_iterator ose)494   ParseImmediate(llvm::StringRef::const_iterator osi,
495                  llvm::StringRef::const_iterator ose) {
496     Operand ret;
497     ret.m_type = Operand::Type::Immediate;
498     std::string str;
499     bool is_hex = false;
500 
501     osi = ConsumeWhitespace(osi, ose);
502 
503     while (osi != ose) {
504       if (*osi >= '0' && *osi <= '9') {
505         str.push_back(*osi);
506       } else if (*osi >= 'a' && *osi <= 'f') {
507         if (is_hex) {
508           str.push_back(*osi);
509         } else {
510           return std::make_pair(Operand(), osi);
511         }
512       } else {
513         switch (*osi) {
514         default:
515           if (str.empty()) {
516             return std::make_pair(Operand(), osi);
517           } else {
518             ret.m_immediate = strtoull(str.c_str(), nullptr, 0);
519             return std::make_pair(ret, osi);
520           }
521         case 'x':
522           if (!str.compare("0")) {
523             is_hex = true;
524             str.push_back(*osi);
525           } else {
526             return std::make_pair(Operand(), osi);
527           }
528           break;
529         case '#':
530         case '$':
531           if (!str.empty()) {
532             return std::make_pair(Operand(), osi);
533           }
534           break;
535         case '-':
536           if (str.empty()) {
537             ret.m_negative = true;
538           } else {
539             return std::make_pair(Operand(), osi);
540           }
541         }
542       }
543       ++osi;
544     }
545 
546     ret.m_immediate = strtoull(str.c_str(), nullptr, 0);
547     return std::make_pair(ret, osi);
548   }
549 
550   // -0x5(%rax,%rax,2)
551   static std::pair<Operand, llvm::StringRef::const_iterator>
ParseIntelIndexedAccess(llvm::StringRef::const_iterator osi,llvm::StringRef::const_iterator ose)552   ParseIntelIndexedAccess(llvm::StringRef::const_iterator osi,
553                           llvm::StringRef::const_iterator ose) {
554     std::pair<Operand, llvm::StringRef::const_iterator> offset_and_iterator =
555         ParseImmediate(osi, ose);
556     if (offset_and_iterator.first.IsValid()) {
557       osi = offset_and_iterator.second;
558     }
559 
560     bool found = false;
561     std::tie(found, osi) = ConsumeChar(osi, '(', ose);
562     if (!found) {
563       return std::make_pair(Operand(), osi);
564     }
565 
566     std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
567         ParseRegisterName(osi, ose);
568     if (base_and_iterator.first.IsValid()) {
569       osi = base_and_iterator.second;
570     } else {
571       return std::make_pair(Operand(), osi);
572     }
573 
574     std::tie(found, osi) = ConsumeChar(osi, ',', ose);
575     if (!found) {
576       return std::make_pair(Operand(), osi);
577     }
578 
579     std::pair<Operand, llvm::StringRef::const_iterator> index_and_iterator =
580         ParseRegisterName(osi, ose);
581     if (index_and_iterator.first.IsValid()) {
582       osi = index_and_iterator.second;
583     } else {
584       return std::make_pair(Operand(), osi);
585     }
586 
587     std::tie(found, osi) = ConsumeChar(osi, ',', ose);
588     if (!found) {
589       return std::make_pair(Operand(), osi);
590     }
591 
592     std::pair<Operand, llvm::StringRef::const_iterator>
593         multiplier_and_iterator = ParseImmediate(osi, ose);
594     if (index_and_iterator.first.IsValid()) {
595       osi = index_and_iterator.second;
596     } else {
597       return std::make_pair(Operand(), osi);
598     }
599 
600     std::tie(found, osi) = ConsumeChar(osi, ')', ose);
601     if (!found) {
602       return std::make_pair(Operand(), osi);
603     }
604 
605     Operand product;
606     product.m_type = Operand::Type::Product;
607     product.m_children.push_back(index_and_iterator.first);
608     product.m_children.push_back(multiplier_and_iterator.first);
609 
610     Operand index;
611     index.m_type = Operand::Type::Sum;
612     index.m_children.push_back(base_and_iterator.first);
613     index.m_children.push_back(product);
614 
615     if (offset_and_iterator.first.IsValid()) {
616       Operand offset;
617       offset.m_type = Operand::Type::Sum;
618       offset.m_children.push_back(offset_and_iterator.first);
619       offset.m_children.push_back(index);
620 
621       Operand deref;
622       deref.m_type = Operand::Type::Dereference;
623       deref.m_children.push_back(offset);
624       return std::make_pair(deref, osi);
625     } else {
626       Operand deref;
627       deref.m_type = Operand::Type::Dereference;
628       deref.m_children.push_back(index);
629       return std::make_pair(deref, osi);
630     }
631   }
632 
633   // -0x10(%rbp)
634   static std::pair<Operand, llvm::StringRef::const_iterator>
ParseIntelDerefAccess(llvm::StringRef::const_iterator osi,llvm::StringRef::const_iterator ose)635   ParseIntelDerefAccess(llvm::StringRef::const_iterator osi,
636                         llvm::StringRef::const_iterator ose) {
637     std::pair<Operand, llvm::StringRef::const_iterator> offset_and_iterator =
638         ParseImmediate(osi, ose);
639     if (offset_and_iterator.first.IsValid()) {
640       osi = offset_and_iterator.second;
641     }
642 
643     bool found = false;
644     std::tie(found, osi) = ConsumeChar(osi, '(', ose);
645     if (!found) {
646       return std::make_pair(Operand(), osi);
647     }
648 
649     std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
650         ParseRegisterName(osi, ose);
651     if (base_and_iterator.first.IsValid()) {
652       osi = base_and_iterator.second;
653     } else {
654       return std::make_pair(Operand(), osi);
655     }
656 
657     std::tie(found, osi) = ConsumeChar(osi, ')', ose);
658     if (!found) {
659       return std::make_pair(Operand(), osi);
660     }
661 
662     if (offset_and_iterator.first.IsValid()) {
663       Operand offset;
664       offset.m_type = Operand::Type::Sum;
665       offset.m_children.push_back(offset_and_iterator.first);
666       offset.m_children.push_back(base_and_iterator.first);
667 
668       Operand deref;
669       deref.m_type = Operand::Type::Dereference;
670       deref.m_children.push_back(offset);
671       return std::make_pair(deref, osi);
672     } else {
673       Operand deref;
674       deref.m_type = Operand::Type::Dereference;
675       deref.m_children.push_back(base_and_iterator.first);
676       return std::make_pair(deref, osi);
677     }
678   }
679 
680   // [sp, #8]!
681   static std::pair<Operand, llvm::StringRef::const_iterator>
ParseARMOffsetAccess(llvm::StringRef::const_iterator osi,llvm::StringRef::const_iterator ose)682   ParseARMOffsetAccess(llvm::StringRef::const_iterator osi,
683                        llvm::StringRef::const_iterator ose) {
684     bool found = false;
685     std::tie(found, osi) = ConsumeChar(osi, '[', ose);
686     if (!found) {
687       return std::make_pair(Operand(), osi);
688     }
689 
690     std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
691         ParseRegisterName(osi, ose);
692     if (base_and_iterator.first.IsValid()) {
693       osi = base_and_iterator.second;
694     } else {
695       return std::make_pair(Operand(), osi);
696     }
697 
698     std::tie(found, osi) = ConsumeChar(osi, ',', ose);
699     if (!found) {
700       return std::make_pair(Operand(), osi);
701     }
702 
703     std::pair<Operand, llvm::StringRef::const_iterator> offset_and_iterator =
704         ParseImmediate(osi, ose);
705     if (offset_and_iterator.first.IsValid()) {
706       osi = offset_and_iterator.second;
707     }
708 
709     std::tie(found, osi) = ConsumeChar(osi, ']', ose);
710     if (!found) {
711       return std::make_pair(Operand(), osi);
712     }
713 
714     Operand offset;
715     offset.m_type = Operand::Type::Sum;
716     offset.m_children.push_back(offset_and_iterator.first);
717     offset.m_children.push_back(base_and_iterator.first);
718 
719     Operand deref;
720     deref.m_type = Operand::Type::Dereference;
721     deref.m_children.push_back(offset);
722     return std::make_pair(deref, osi);
723   }
724 
725   // [sp]
726   static std::pair<Operand, llvm::StringRef::const_iterator>
ParseARMDerefAccess(llvm::StringRef::const_iterator osi,llvm::StringRef::const_iterator ose)727   ParseARMDerefAccess(llvm::StringRef::const_iterator osi,
728                       llvm::StringRef::const_iterator ose) {
729     bool found = false;
730     std::tie(found, osi) = ConsumeChar(osi, '[', ose);
731     if (!found) {
732       return std::make_pair(Operand(), osi);
733     }
734 
735     std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
736         ParseRegisterName(osi, ose);
737     if (base_and_iterator.first.IsValid()) {
738       osi = base_and_iterator.second;
739     } else {
740       return std::make_pair(Operand(), osi);
741     }
742 
743     std::tie(found, osi) = ConsumeChar(osi, ']', ose);
744     if (!found) {
745       return std::make_pair(Operand(), osi);
746     }
747 
748     Operand deref;
749     deref.m_type = Operand::Type::Dereference;
750     deref.m_children.push_back(base_and_iterator.first);
751     return std::make_pair(deref, osi);
752   }
753 
DumpOperand(const Operand & op,Stream & s)754   static void DumpOperand(const Operand &op, Stream &s) {
755     switch (op.m_type) {
756     case Operand::Type::Dereference:
757       s.PutCString("*");
758       DumpOperand(op.m_children[0], s);
759       break;
760     case Operand::Type::Immediate:
761       if (op.m_negative) {
762         s.PutCString("-");
763       }
764       s.PutCString(llvm::to_string(op.m_immediate));
765       break;
766     case Operand::Type::Invalid:
767       s.PutCString("Invalid");
768       break;
769     case Operand::Type::Product:
770       s.PutCString("(");
771       DumpOperand(op.m_children[0], s);
772       s.PutCString("*");
773       DumpOperand(op.m_children[1], s);
774       s.PutCString(")");
775       break;
776     case Operand::Type::Register:
777       s.PutCString(op.m_register.AsCString());
778       break;
779     case Operand::Type::Sum:
780       s.PutCString("(");
781       DumpOperand(op.m_children[0], s);
782       s.PutCString("+");
783       DumpOperand(op.m_children[1], s);
784       s.PutCString(")");
785       break;
786     }
787   }
788 
ParseOperands(llvm::SmallVectorImpl<Instruction::Operand> & operands)789   bool ParseOperands(
790       llvm::SmallVectorImpl<Instruction::Operand> &operands) override {
791     const char *operands_string = GetOperands(nullptr);
792 
793     if (!operands_string) {
794       return false;
795     }
796 
797     llvm::StringRef operands_ref(operands_string);
798 
799     llvm::StringRef::const_iterator osi = operands_ref.begin();
800     llvm::StringRef::const_iterator ose = operands_ref.end();
801 
802     while (osi != ose) {
803       Operand operand;
804       llvm::StringRef::const_iterator iter;
805 
806       if ((std::tie(operand, iter) = ParseIntelIndexedAccess(osi, ose),
807            operand.IsValid()) ||
808           (std::tie(operand, iter) = ParseIntelDerefAccess(osi, ose),
809            operand.IsValid()) ||
810           (std::tie(operand, iter) = ParseARMOffsetAccess(osi, ose),
811            operand.IsValid()) ||
812           (std::tie(operand, iter) = ParseARMDerefAccess(osi, ose),
813            operand.IsValid()) ||
814           (std::tie(operand, iter) = ParseRegisterName(osi, ose),
815            operand.IsValid()) ||
816           (std::tie(operand, iter) = ParseImmediate(osi, ose),
817            operand.IsValid())) {
818         osi = iter;
819         operands.push_back(operand);
820       } else {
821         return false;
822       }
823 
824       std::pair<bool, llvm::StringRef::const_iterator> found_and_iter =
825           ConsumeChar(osi, ',', ose);
826       if (found_and_iter.first) {
827         osi = found_and_iter.second;
828       }
829 
830       osi = ConsumeWhitespace(osi, ose);
831     }
832 
833     DisassemblerSP disasm_sp = m_disasm_wp.lock();
834 
835     if (disasm_sp && operands.size() > 1) {
836       // TODO tie this into the MC Disassembler's notion of clobbers.
837       switch (disasm_sp->GetArchitecture().GetMachine()) {
838       default:
839         break;
840       case llvm::Triple::x86:
841       case llvm::Triple::x86_64:
842         operands[operands.size() - 1].m_clobbered = true;
843         break;
844       case llvm::Triple::arm:
845         operands[0].m_clobbered = true;
846         break;
847       }
848     }
849 
850     if (Log *log =
851             lldb_private::GetLogIfAllCategoriesSet(LIBLLDB_LOG_PROCESS)) {
852       StreamString ss;
853 
854       ss.Printf("[%s] expands to %zu operands:\n", operands_string,
855                 operands.size());
856       for (const Operand &operand : operands) {
857         ss.PutCString("  ");
858         DumpOperand(operand, ss);
859         ss.PutCString("\n");
860       }
861 
862       log->PutString(ss.GetString());
863     }
864 
865     return true;
866   }
867 
IsCall()868   bool IsCall() override {
869     if (m_is_call == eLazyBoolCalculate) {
870       DisassemblerScope disasm(*this);
871       if (disasm) {
872         DataExtractor data;
873         if (m_opcode.GetData(data)) {
874           bool is_alternate_isa;
875           lldb::addr_t pc = m_address.GetFileAddress();
876 
877           DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
878               GetDisasmToUse(is_alternate_isa, disasm);
879           const uint8_t *opcode_data = data.GetDataStart();
880           const size_t opcode_data_len = data.GetByteSize();
881           llvm::MCInst inst;
882           const size_t inst_size =
883               mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
884           if (inst_size == 0) {
885             m_is_call = eLazyBoolNo;
886           } else {
887             if (mc_disasm_ptr->IsCall(inst))
888               m_is_call = eLazyBoolYes;
889             else
890               m_is_call = eLazyBoolNo;
891           }
892         }
893       }
894     }
895     return m_is_call == eLazyBoolYes;
896   }
897 
898 protected:
899   std::weak_ptr<DisassemblerLLVMC> m_disasm_wp;
900   LazyBool m_does_branch;
901   LazyBool m_has_delay_slot;
902   LazyBool m_is_call;
903   bool m_is_valid;
904   bool m_using_file_addr;
905 
906 private:
907   DisassemblerLLVMC::MCDisasmInstance *
GetDisasmToUse(bool & is_alternate_isa,DisassemblerScope & disasm)908   GetDisasmToUse(bool &is_alternate_isa, DisassemblerScope &disasm) {
909     is_alternate_isa = false;
910     if (disasm) {
911       if (disasm->m_alternate_disasm_up) {
912         const AddressClass address_class = GetAddressClass();
913 
914         if (address_class == AddressClass::eCodeAlternateISA) {
915           is_alternate_isa = true;
916           return disasm->m_alternate_disasm_up.get();
917         }
918       }
919       return disasm->m_disasm_up.get();
920     }
921     return nullptr;
922   }
923 };
924 
925 std::unique_ptr<DisassemblerLLVMC::MCDisasmInstance>
Create(const char * triple,const char * cpu,const char * features_str,unsigned flavor,DisassemblerLLVMC & owner)926 DisassemblerLLVMC::MCDisasmInstance::Create(const char *triple, const char *cpu,
927                                             const char *features_str,
928                                             unsigned flavor,
929                                             DisassemblerLLVMC &owner) {
930   using Instance = std::unique_ptr<DisassemblerLLVMC::MCDisasmInstance>;
931 
932   std::string Status;
933   const llvm::Target *curr_target =
934       llvm::TargetRegistry::lookupTarget(triple, Status);
935   if (!curr_target)
936     return Instance();
937 
938   std::unique_ptr<llvm::MCInstrInfo> instr_info_up(
939       curr_target->createMCInstrInfo());
940   if (!instr_info_up)
941     return Instance();
942 
943   std::unique_ptr<llvm::MCRegisterInfo> reg_info_up(
944       curr_target->createMCRegInfo(triple));
945   if (!reg_info_up)
946     return Instance();
947 
948   std::unique_ptr<llvm::MCSubtargetInfo> subtarget_info_up(
949       curr_target->createMCSubtargetInfo(triple, cpu, features_str));
950   if (!subtarget_info_up)
951     return Instance();
952 
953   llvm::MCTargetOptions MCOptions;
954   std::unique_ptr<llvm::MCAsmInfo> asm_info_up(
955       curr_target->createMCAsmInfo(*reg_info_up, triple, MCOptions));
956   if (!asm_info_up)
957     return Instance();
958 
959   std::unique_ptr<llvm::MCContext> context_up(
960       new llvm::MCContext(asm_info_up.get(), reg_info_up.get(), nullptr));
961   if (!context_up)
962     return Instance();
963 
964   std::unique_ptr<llvm::MCDisassembler> disasm_up(
965       curr_target->createMCDisassembler(*subtarget_info_up, *context_up));
966   if (!disasm_up)
967     return Instance();
968 
969   std::unique_ptr<llvm::MCRelocationInfo> rel_info_up(
970       curr_target->createMCRelocationInfo(triple, *context_up));
971   if (!rel_info_up)
972     return Instance();
973 
974   std::unique_ptr<llvm::MCSymbolizer> symbolizer_up(
975       curr_target->createMCSymbolizer(
976           triple, nullptr, DisassemblerLLVMC::SymbolLookupCallback, &owner,
977           context_up.get(), std::move(rel_info_up)));
978   disasm_up->setSymbolizer(std::move(symbolizer_up));
979 
980   unsigned asm_printer_variant =
981       flavor == ~0U ? asm_info_up->getAssemblerDialect() : flavor;
982 
983   std::unique_ptr<llvm::MCInstPrinter> instr_printer_up(
984       curr_target->createMCInstPrinter(llvm::Triple{triple},
985                                        asm_printer_variant, *asm_info_up,
986                                        *instr_info_up, *reg_info_up));
987   if (!instr_printer_up)
988     return Instance();
989 
990   return Instance(
991       new MCDisasmInstance(std::move(instr_info_up), std::move(reg_info_up),
992                            std::move(subtarget_info_up), std::move(asm_info_up),
993                            std::move(context_up), std::move(disasm_up),
994                            std::move(instr_printer_up)));
995 }
996 
MCDisasmInstance(std::unique_ptr<llvm::MCInstrInfo> && instr_info_up,std::unique_ptr<llvm::MCRegisterInfo> && reg_info_up,std::unique_ptr<llvm::MCSubtargetInfo> && subtarget_info_up,std::unique_ptr<llvm::MCAsmInfo> && asm_info_up,std::unique_ptr<llvm::MCContext> && context_up,std::unique_ptr<llvm::MCDisassembler> && disasm_up,std::unique_ptr<llvm::MCInstPrinter> && instr_printer_up)997 DisassemblerLLVMC::MCDisasmInstance::MCDisasmInstance(
998     std::unique_ptr<llvm::MCInstrInfo> &&instr_info_up,
999     std::unique_ptr<llvm::MCRegisterInfo> &&reg_info_up,
1000     std::unique_ptr<llvm::MCSubtargetInfo> &&subtarget_info_up,
1001     std::unique_ptr<llvm::MCAsmInfo> &&asm_info_up,
1002     std::unique_ptr<llvm::MCContext> &&context_up,
1003     std::unique_ptr<llvm::MCDisassembler> &&disasm_up,
1004     std::unique_ptr<llvm::MCInstPrinter> &&instr_printer_up)
1005     : m_instr_info_up(std::move(instr_info_up)),
1006       m_reg_info_up(std::move(reg_info_up)),
1007       m_subtarget_info_up(std::move(subtarget_info_up)),
1008       m_asm_info_up(std::move(asm_info_up)),
1009       m_context_up(std::move(context_up)), m_disasm_up(std::move(disasm_up)),
1010       m_instr_printer_up(std::move(instr_printer_up)) {
1011   assert(m_instr_info_up && m_reg_info_up && m_subtarget_info_up &&
1012          m_asm_info_up && m_context_up && m_disasm_up && m_instr_printer_up);
1013 }
1014 
GetMCInst(const uint8_t * opcode_data,size_t opcode_data_len,lldb::addr_t pc,llvm::MCInst & mc_inst) const1015 uint64_t DisassemblerLLVMC::MCDisasmInstance::GetMCInst(
1016     const uint8_t *opcode_data, size_t opcode_data_len, lldb::addr_t pc,
1017     llvm::MCInst &mc_inst) const {
1018   llvm::ArrayRef<uint8_t> data(opcode_data, opcode_data_len);
1019   llvm::MCDisassembler::DecodeStatus status;
1020 
1021   uint64_t new_inst_size;
1022   status = m_disasm_up->getInstruction(mc_inst, new_inst_size, data, pc,
1023                                        llvm::nulls());
1024   if (status == llvm::MCDisassembler::Success)
1025     return new_inst_size;
1026   else
1027     return 0;
1028 }
1029 
PrintMCInst(llvm::MCInst & mc_inst,std::string & inst_string,std::string & comments_string)1030 void DisassemblerLLVMC::MCDisasmInstance::PrintMCInst(
1031     llvm::MCInst &mc_inst, std::string &inst_string,
1032     std::string &comments_string) {
1033   llvm::raw_string_ostream inst_stream(inst_string);
1034   llvm::raw_string_ostream comments_stream(comments_string);
1035 
1036   m_instr_printer_up->setCommentStream(comments_stream);
1037   m_instr_printer_up->printInst(&mc_inst, 0, llvm::StringRef(),
1038                                 *m_subtarget_info_up, inst_stream);
1039   m_instr_printer_up->setCommentStream(llvm::nulls());
1040   comments_stream.flush();
1041 
1042   static std::string g_newlines("\r\n");
1043 
1044   for (size_t newline_pos = 0;
1045        (newline_pos = comments_string.find_first_of(g_newlines, newline_pos)) !=
1046        comments_string.npos;
1047        /**/) {
1048     comments_string.replace(comments_string.begin() + newline_pos,
1049                             comments_string.begin() + newline_pos + 1, 1, ' ');
1050   }
1051 }
1052 
SetStyle(bool use_hex_immed,HexImmediateStyle hex_style)1053 void DisassemblerLLVMC::MCDisasmInstance::SetStyle(
1054     bool use_hex_immed, HexImmediateStyle hex_style) {
1055   m_instr_printer_up->setPrintImmHex(use_hex_immed);
1056   switch (hex_style) {
1057   case eHexStyleC:
1058     m_instr_printer_up->setPrintHexStyle(llvm::HexStyle::C);
1059     break;
1060   case eHexStyleAsm:
1061     m_instr_printer_up->setPrintHexStyle(llvm::HexStyle::Asm);
1062     break;
1063   }
1064 }
1065 
CanBranch(llvm::MCInst & mc_inst) const1066 bool DisassemblerLLVMC::MCDisasmInstance::CanBranch(
1067     llvm::MCInst &mc_inst) const {
1068   return m_instr_info_up->get(mc_inst.getOpcode())
1069       .mayAffectControlFlow(mc_inst, *m_reg_info_up);
1070 }
1071 
HasDelaySlot(llvm::MCInst & mc_inst) const1072 bool DisassemblerLLVMC::MCDisasmInstance::HasDelaySlot(
1073     llvm::MCInst &mc_inst) const {
1074   return m_instr_info_up->get(mc_inst.getOpcode()).hasDelaySlot();
1075 }
1076 
IsCall(llvm::MCInst & mc_inst) const1077 bool DisassemblerLLVMC::MCDisasmInstance::IsCall(llvm::MCInst &mc_inst) const {
1078   return m_instr_info_up->get(mc_inst.getOpcode()).isCall();
1079 }
1080 
DisassemblerLLVMC(const ArchSpec & arch,const char * flavor_string)1081 DisassemblerLLVMC::DisassemblerLLVMC(const ArchSpec &arch,
1082                                      const char *flavor_string)
1083     : Disassembler(arch, flavor_string), m_exe_ctx(nullptr), m_inst(nullptr),
1084       m_data_from_file(false) {
1085   if (!FlavorValidForArchSpec(arch, m_flavor.c_str())) {
1086     m_flavor.assign("default");
1087   }
1088 
1089   unsigned flavor = ~0U;
1090   llvm::Triple triple = arch.GetTriple();
1091 
1092   // So far the only supported flavor is "intel" on x86.  The base class will
1093   // set this correctly coming in.
1094   if (triple.getArch() == llvm::Triple::x86 ||
1095       triple.getArch() == llvm::Triple::x86_64) {
1096     if (m_flavor == "intel") {
1097       flavor = 1;
1098     } else if (m_flavor == "att") {
1099       flavor = 0;
1100     }
1101   }
1102 
1103   ArchSpec thumb_arch(arch);
1104   if (triple.getArch() == llvm::Triple::arm) {
1105     std::string thumb_arch_name(thumb_arch.GetTriple().getArchName().str());
1106     // Replace "arm" with "thumb" so we get all thumb variants correct
1107     if (thumb_arch_name.size() > 3) {
1108       thumb_arch_name.erase(0, 3);
1109       thumb_arch_name.insert(0, "thumb");
1110     } else {
1111       thumb_arch_name = "thumbv8.2a";
1112     }
1113     thumb_arch.GetTriple().setArchName(llvm::StringRef(thumb_arch_name));
1114   }
1115 
1116   // If no sub architecture specified then use the most recent arm architecture
1117   // so the disassembler will return all instruction. Without it we will see a
1118   // lot of unknow opcode in case the code uses instructions which are not
1119   // available in the oldest arm version (used when no sub architecture is
1120   // specified)
1121   if (triple.getArch() == llvm::Triple::arm &&
1122       triple.getSubArch() == llvm::Triple::NoSubArch)
1123     triple.setArchName("armv8.2a");
1124 
1125   std::string features_str = "";
1126   const char *triple_str = triple.getTriple().c_str();
1127 
1128   // ARM Cortex M0-M7 devices only execute thumb instructions
1129   if (arch.IsAlwaysThumbInstructions()) {
1130     triple_str = thumb_arch.GetTriple().getTriple().c_str();
1131     features_str += "+fp-armv8,";
1132   }
1133 
1134   const char *cpu = "";
1135 
1136   switch (arch.GetCore()) {
1137   case ArchSpec::eCore_mips32:
1138   case ArchSpec::eCore_mips32el:
1139     cpu = "mips32";
1140     break;
1141   case ArchSpec::eCore_mips32r2:
1142   case ArchSpec::eCore_mips32r2el:
1143     cpu = "mips32r2";
1144     break;
1145   case ArchSpec::eCore_mips32r3:
1146   case ArchSpec::eCore_mips32r3el:
1147     cpu = "mips32r3";
1148     break;
1149   case ArchSpec::eCore_mips32r5:
1150   case ArchSpec::eCore_mips32r5el:
1151     cpu = "mips32r5";
1152     break;
1153   case ArchSpec::eCore_mips32r6:
1154   case ArchSpec::eCore_mips32r6el:
1155     cpu = "mips32r6";
1156     break;
1157   case ArchSpec::eCore_mips64:
1158   case ArchSpec::eCore_mips64el:
1159     cpu = "mips64";
1160     break;
1161   case ArchSpec::eCore_mips64r2:
1162   case ArchSpec::eCore_mips64r2el:
1163     cpu = "mips64r2";
1164     break;
1165   case ArchSpec::eCore_mips64r3:
1166   case ArchSpec::eCore_mips64r3el:
1167     cpu = "mips64r3";
1168     break;
1169   case ArchSpec::eCore_mips64r5:
1170   case ArchSpec::eCore_mips64r5el:
1171     cpu = "mips64r5";
1172     break;
1173   case ArchSpec::eCore_mips64r6:
1174   case ArchSpec::eCore_mips64r6el:
1175     cpu = "mips64r6";
1176     break;
1177   default:
1178     cpu = "";
1179     break;
1180   }
1181 
1182   if (arch.IsMIPS()) {
1183     uint32_t arch_flags = arch.GetFlags();
1184     if (arch_flags & ArchSpec::eMIPSAse_msa)
1185       features_str += "+msa,";
1186     if (arch_flags & ArchSpec::eMIPSAse_dsp)
1187       features_str += "+dsp,";
1188     if (arch_flags & ArchSpec::eMIPSAse_dspr2)
1189       features_str += "+dspr2,";
1190   }
1191 
1192   // If any AArch64 variant, enable the ARMv8.5 ISA with SVE extensions so we
1193   // can disassemble newer instructions.
1194   if (triple.getArch() == llvm::Triple::aarch64 ||
1195       triple.getArch() == llvm::Triple::aarch64_32)
1196     features_str += "+v8.5a,+sve2";
1197 
1198   if ((triple.getArch() == llvm::Triple::aarch64 ||
1199        triple.getArch() == llvm::Triple::aarch64_32)
1200       && triple.getVendor() == llvm::Triple::Apple) {
1201     cpu = "apple-latest";
1202   }
1203 
1204   // We use m_disasm_up.get() to tell whether we are valid or not, so if this
1205   // isn't good for some reason, we won't be valid and FindPlugin will fail and
1206   // we won't get used.
1207   m_disasm_up = MCDisasmInstance::Create(triple_str, cpu, features_str.c_str(),
1208                                          flavor, *this);
1209 
1210   llvm::Triple::ArchType llvm_arch = triple.getArch();
1211 
1212   // For arm CPUs that can execute arm or thumb instructions, also create a
1213   // thumb instruction disassembler.
1214   if (llvm_arch == llvm::Triple::arm) {
1215     std::string thumb_triple(thumb_arch.GetTriple().getTriple());
1216     m_alternate_disasm_up =
1217         MCDisasmInstance::Create(thumb_triple.c_str(), "", features_str.c_str(),
1218                                  flavor, *this);
1219     if (!m_alternate_disasm_up)
1220       m_disasm_up.reset();
1221 
1222   } else if (arch.IsMIPS()) {
1223     /* Create alternate disassembler for MIPS16 and microMIPS */
1224     uint32_t arch_flags = arch.GetFlags();
1225     if (arch_flags & ArchSpec::eMIPSAse_mips16)
1226       features_str += "+mips16,";
1227     else if (arch_flags & ArchSpec::eMIPSAse_micromips)
1228       features_str += "+micromips,";
1229 
1230     m_alternate_disasm_up = MCDisasmInstance::Create(
1231         triple_str, cpu, features_str.c_str(), flavor, *this);
1232     if (!m_alternate_disasm_up)
1233       m_disasm_up.reset();
1234   }
1235 }
1236 
1237 DisassemblerLLVMC::~DisassemblerLLVMC() = default;
1238 
CreateInstance(const ArchSpec & arch,const char * flavor)1239 Disassembler *DisassemblerLLVMC::CreateInstance(const ArchSpec &arch,
1240                                                 const char *flavor) {
1241   if (arch.GetTriple().getArch() != llvm::Triple::UnknownArch) {
1242     std::unique_ptr<DisassemblerLLVMC> disasm_up(
1243         new DisassemblerLLVMC(arch, flavor));
1244 
1245     if (disasm_up.get() && disasm_up->IsValid())
1246       return disasm_up.release();
1247   }
1248   return nullptr;
1249 }
1250 
DecodeInstructions(const Address & base_addr,const DataExtractor & data,lldb::offset_t data_offset,size_t num_instructions,bool append,bool data_from_file)1251 size_t DisassemblerLLVMC::DecodeInstructions(const Address &base_addr,
1252                                              const DataExtractor &data,
1253                                              lldb::offset_t data_offset,
1254                                              size_t num_instructions,
1255                                              bool append, bool data_from_file) {
1256   if (!append)
1257     m_instruction_list.Clear();
1258 
1259   if (!IsValid())
1260     return 0;
1261 
1262   m_data_from_file = data_from_file;
1263   uint32_t data_cursor = data_offset;
1264   const size_t data_byte_size = data.GetByteSize();
1265   uint32_t instructions_parsed = 0;
1266   Address inst_addr(base_addr);
1267 
1268   while (data_cursor < data_byte_size &&
1269          instructions_parsed < num_instructions) {
1270 
1271     AddressClass address_class = AddressClass::eCode;
1272 
1273     if (m_alternate_disasm_up)
1274       address_class = inst_addr.GetAddressClass();
1275 
1276     InstructionSP inst_sp(
1277         new InstructionLLVMC(*this, inst_addr, address_class));
1278 
1279     if (!inst_sp)
1280       break;
1281 
1282     uint32_t inst_size = inst_sp->Decode(*this, data, data_cursor);
1283 
1284     if (inst_size == 0)
1285       break;
1286 
1287     m_instruction_list.Append(inst_sp);
1288     data_cursor += inst_size;
1289     inst_addr.Slide(inst_size);
1290     instructions_parsed++;
1291   }
1292 
1293   return data_cursor - data_offset;
1294 }
1295 
Initialize()1296 void DisassemblerLLVMC::Initialize() {
1297   PluginManager::RegisterPlugin(GetPluginNameStatic(),
1298                                 "Disassembler that uses LLVM MC to disassemble "
1299                                 "i386, x86_64, ARM, and ARM64.",
1300                                 CreateInstance);
1301 
1302   llvm::InitializeAllTargetInfos();
1303   llvm::InitializeAllTargetMCs();
1304   llvm::InitializeAllAsmParsers();
1305   llvm::InitializeAllDisassemblers();
1306 }
1307 
Terminate()1308 void DisassemblerLLVMC::Terminate() {
1309   PluginManager::UnregisterPlugin(CreateInstance);
1310 }
1311 
GetPluginNameStatic()1312 ConstString DisassemblerLLVMC::GetPluginNameStatic() {
1313   static ConstString g_name("llvm-mc");
1314   return g_name;
1315 }
1316 
OpInfoCallback(void * disassembler,uint64_t pc,uint64_t offset,uint64_t size,int tag_type,void * tag_bug)1317 int DisassemblerLLVMC::OpInfoCallback(void *disassembler, uint64_t pc,
1318                                       uint64_t offset, uint64_t size,
1319                                       int tag_type, void *tag_bug) {
1320   return static_cast<DisassemblerLLVMC *>(disassembler)
1321       ->OpInfo(pc, offset, size, tag_type, tag_bug);
1322 }
1323 
SymbolLookupCallback(void * disassembler,uint64_t value,uint64_t * type,uint64_t pc,const char ** name)1324 const char *DisassemblerLLVMC::SymbolLookupCallback(void *disassembler,
1325                                                     uint64_t value,
1326                                                     uint64_t *type, uint64_t pc,
1327                                                     const char **name) {
1328   return static_cast<DisassemblerLLVMC *>(disassembler)
1329       ->SymbolLookup(value, type, pc, name);
1330 }
1331 
FlavorValidForArchSpec(const lldb_private::ArchSpec & arch,const char * flavor)1332 bool DisassemblerLLVMC::FlavorValidForArchSpec(
1333     const lldb_private::ArchSpec &arch, const char *flavor) {
1334   llvm::Triple triple = arch.GetTriple();
1335   if (flavor == nullptr || strcmp(flavor, "default") == 0)
1336     return true;
1337 
1338   if (triple.getArch() == llvm::Triple::x86 ||
1339       triple.getArch() == llvm::Triple::x86_64) {
1340     return strcmp(flavor, "intel") == 0 || strcmp(flavor, "att") == 0;
1341   } else
1342     return false;
1343 }
1344 
IsValid() const1345 bool DisassemblerLLVMC::IsValid() const { return m_disasm_up.operator bool(); }
1346 
OpInfo(uint64_t PC,uint64_t Offset,uint64_t Size,int tag_type,void * tag_bug)1347 int DisassemblerLLVMC::OpInfo(uint64_t PC, uint64_t Offset, uint64_t Size,
1348                               int tag_type, void *tag_bug) {
1349   switch (tag_type) {
1350   default:
1351     break;
1352   case 1:
1353     memset(tag_bug, 0, sizeof(::LLVMOpInfo1));
1354     break;
1355   }
1356   return 0;
1357 }
1358 
SymbolLookup(uint64_t value,uint64_t * type_ptr,uint64_t pc,const char ** name)1359 const char *DisassemblerLLVMC::SymbolLookup(uint64_t value, uint64_t *type_ptr,
1360                                             uint64_t pc, const char **name) {
1361   if (*type_ptr) {
1362     if (m_exe_ctx && m_inst) {
1363       // std::string remove_this_prior_to_checkin;
1364       Target *target = m_exe_ctx ? m_exe_ctx->GetTargetPtr() : nullptr;
1365       Address value_so_addr;
1366       Address pc_so_addr;
1367       if (m_inst->UsingFileAddress()) {
1368         ModuleSP module_sp(m_inst->GetAddress().GetModule());
1369         if (module_sp) {
1370           module_sp->ResolveFileAddress(value, value_so_addr);
1371           module_sp->ResolveFileAddress(pc, pc_so_addr);
1372         }
1373       } else if (target && !target->GetSectionLoadList().IsEmpty()) {
1374         target->GetSectionLoadList().ResolveLoadAddress(value, value_so_addr);
1375         target->GetSectionLoadList().ResolveLoadAddress(pc, pc_so_addr);
1376       }
1377 
1378       SymbolContext sym_ctx;
1379       const SymbolContextItem resolve_scope =
1380           eSymbolContextFunction | eSymbolContextSymbol;
1381       if (pc_so_addr.IsValid() && pc_so_addr.GetModule()) {
1382         pc_so_addr.GetModule()->ResolveSymbolContextForAddress(
1383             pc_so_addr, resolve_scope, sym_ctx);
1384       }
1385 
1386       if (value_so_addr.IsValid() && value_so_addr.GetSection()) {
1387         StreamString ss;
1388 
1389         bool format_omitting_current_func_name = false;
1390         if (sym_ctx.symbol || sym_ctx.function) {
1391           AddressRange range;
1392           if (sym_ctx.GetAddressRange(resolve_scope, 0, false, range) &&
1393               range.GetBaseAddress().IsValid() &&
1394               range.ContainsLoadAddress(value_so_addr, target)) {
1395             format_omitting_current_func_name = true;
1396           }
1397         }
1398 
1399         // If the "value" address (the target address we're symbolicating) is
1400         // inside the same SymbolContext as the current instruction pc
1401         // (pc_so_addr), don't print the full function name - just print it
1402         // with DumpStyleNoFunctionName style, e.g. "<+36>".
1403         if (format_omitting_current_func_name) {
1404           value_so_addr.Dump(&ss, target, Address::DumpStyleNoFunctionName,
1405                              Address::DumpStyleSectionNameOffset);
1406         } else {
1407           value_so_addr.Dump(
1408               &ss, target,
1409               Address::DumpStyleResolvedDescriptionNoFunctionArguments,
1410               Address::DumpStyleSectionNameOffset);
1411         }
1412 
1413         if (!ss.GetString().empty()) {
1414           // If Address::Dump returned a multi-line description, most commonly
1415           // seen when we have multiple levels of inlined functions at an
1416           // address, only show the first line.
1417           std::string str = ss.GetString();
1418           size_t first_eol_char = str.find_first_of("\r\n");
1419           if (first_eol_char != std::string::npos) {
1420             str.erase(first_eol_char);
1421           }
1422           m_inst->AppendComment(str);
1423         }
1424       }
1425     }
1426   }
1427 
1428   *type_ptr = LLVMDisassembler_ReferenceType_InOut_None;
1429   *name = nullptr;
1430   return nullptr;
1431 }
1432 
1433 // PluginInterface protocol
GetPluginName()1434 ConstString DisassemblerLLVMC::GetPluginName() { return GetPluginNameStatic(); }
1435 
GetPluginVersion()1436 uint32_t DisassemblerLLVMC::GetPluginVersion() { return 1; }
1437