1 //===-- DisassemblerLLVMC.cpp -----------------------------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "DisassemblerLLVMC.h"
12 #include "llvm-c/Disassembler.h"
13 #include "llvm/ADT/SmallString.h"
14 #include "llvm/MC/MCAsmInfo.h"
15 #include "llvm/MC/MCContext.h"
16 #include "llvm/MC/MCDisassembler/MCDisassembler.h"
17 #include "llvm/MC/MCDisassembler/MCExternalSymbolizer.h"
18 #include "llvm/MC/MCDisassembler/MCRelocationInfo.h"
19 #include "llvm/MC/MCInst.h"
20 #include "llvm/MC/MCInstPrinter.h"
21 #include "llvm/MC/MCInstrInfo.h"
22 #include "llvm/MC/MCRegisterInfo.h"
23 #include "llvm/MC/MCSubtargetInfo.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"
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"
44 using namespace lldb_private;
46 class DisassemblerLLVMC::MCDisasmInstance {
48 static std::unique_ptr<MCDisasmInstance>
49 Create(const char *triple, const char *cpu, const char *features_str,
50 unsigned flavor, DisassemblerLLVMC &owner);
52 ~MCDisasmInstance() = default;
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;
64 MCDisasmInstance(std::unique_ptr<llvm::MCInstrInfo> &&instr_info_up,
65 std::unique_ptr<llvm::MCRegisterInfo> &®_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);
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;
81 class InstructionLLVMC : public lldb_private::Instruction {
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) {}
93 ~InstructionLLVMC() override = default;
95 bool DoesBranch() override {
96 if (m_does_branch == eLazyBoolCalculate) {
97 DisassemblerScope disasm(*this);
100 if (m_opcode.GetData(data)) {
101 bool is_alternate_isa;
102 lldb::addr_t pc = m_address.GetFileAddress();
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();
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
114 m_does_branch = eLazyBoolYes;
116 const bool can_branch = mc_disasm_ptr->CanBranch(inst);
118 m_does_branch = eLazyBoolYes;
120 m_does_branch = eLazyBoolNo;
125 return m_does_branch == eLazyBoolYes;
128 bool HasDelaySlot() override {
129 if (m_has_delay_slot == eLazyBoolCalculate) {
130 DisassemblerScope disasm(*this);
133 if (m_opcode.GetData(data)) {
134 bool is_alternate_isa;
135 lldb::addr_t pc = m_address.GetFileAddress();
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();
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
147 m_has_delay_slot = eLazyBoolNo;
149 const bool has_delay_slot = mc_disasm_ptr->HasDelaySlot(inst);
151 m_has_delay_slot = eLazyBoolYes;
153 m_has_delay_slot = eLazyBoolNo;
158 return m_has_delay_slot == eLazyBoolYes;
161 DisassemblerLLVMC::MCDisasmInstance *GetDisasmToUse(bool &is_alternate_isa) {
162 DisassemblerScope disasm(*this);
163 return GetDisasmToUse(is_alternate_isa, disasm);
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
172 DisassemblerScope disasm(*this);
174 const ArchSpec &arch = disasm->GetArchitecture();
175 const lldb::ByteOrder byte_order = data.GetByteOrder();
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))
184 switch (min_op_byte_size) {
186 m_opcode.SetOpcode8(data.GetU8(&data_offset), byte_order);
191 m_opcode.SetOpcode16(data.GetU16(&data_offset), byte_order);
196 m_opcode.SetOpcode32(data.GetU32(&data_offset), byte_order);
201 m_opcode.SetOpcode64(data.GetU64(&data_offset), byte_order);
206 m_opcode.SetOpcodeBytes(data.PeekData(data_offset, min_op_byte_size),
213 bool is_alternate_isa = false;
214 DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
215 GetDisasmToUse(is_alternate_isa, disasm);
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);
227 thumb_opcode |= data.GetU16(&data_offset);
228 m_opcode.SetOpcode16_2(thumb_opcode, byte_order);
232 m_opcode.SetOpcode32(data.GetU32(&data_offset), byte_order);
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();
244 const size_t inst_size =
245 mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
249 m_opcode.SetOpcodeBytes(opcode_data, inst_size);
254 return m_opcode.GetByteSize();
259 void AppendComment(std::string &description) {
260 if (m_comment.empty())
261 m_comment.swap(description);
263 m_comment.append(", ");
264 m_comment.append(description);
268 void CalculateMnemonicOperandsAndComment(
269 const lldb_private::ExecutionContext *exe_ctx) override {
271 const AddressClass address_class = GetAddressClass();
273 if (m_opcode.GetData(data)) {
274 std::string out_string;
275 std::string comment_string;
277 DisassemblerScope disasm(*this, exe_ctx);
279 DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr;
281 if (address_class == AddressClass::eCodeAlternateISA)
282 mc_disasm_ptr = disasm->m_alternate_disasm_up.get();
284 mc_disasm_ptr = disasm->m_disasm_up.get();
286 lldb::addr_t pc = m_address.GetFileAddress();
287 m_using_file_addr = true;
289 const bool data_from_file = disasm->m_data_from_file;
290 bool use_hex_immediates = true;
291 Disassembler::HexImmediateStyle hex_style = Disassembler::eHexStyleC;
294 Target *target = exe_ctx->GetTargetPtr();
296 use_hex_immediates = target->GetUseHexImmediates();
297 hex_style = target->GetHexImmediateStyle();
299 if (!data_from_file) {
300 const lldb::addr_t load_addr = m_address.GetLoadAddress(target);
301 if (load_addr != LLDB_INVALID_ADDRESS) {
303 m_using_file_addr = false;
309 const uint8_t *opcode_data = data.GetDataStart();
310 const size_t opcode_data_len = data.GetByteSize();
313 mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
316 mc_disasm_ptr->SetStyle(use_hex_immediates, hex_style);
317 mc_disasm_ptr->PrintMCInst(inst, out_string, comment_string);
319 if (!comment_string.empty()) {
320 AppendComment(comment_string);
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();
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);
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);
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);
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);
359 const uint8_t *bytes = data.PeekData(offset, inst_size);
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]);
370 m_mnemonics = mnemonic_strm.GetString();
373 if (m_does_branch == eLazyBoolCalculate) {
374 const bool can_branch = mc_disasm_ptr->CanBranch(inst);
376 m_does_branch = eLazyBoolYes;
378 m_does_branch = eLazyBoolNo;
382 static RegularExpression s_regex(
383 llvm::StringRef("[ \t]*([^ ^\t]+)[ \t]*([^ ^\t].*)?"));
385 RegularExpression::Match matches(3);
387 if (s_regex.Execute(out_string, &matches)) {
388 matches.GetMatchAtIndex(out_string.c_str(), 1, m_opcode_name);
389 matches.GetMatchAtIndex(out_string.c_str(), 2, m_mnemonics);
395 bool IsValid() const { return m_is_valid; }
397 bool UsingFileAddress() const { return m_using_file_addr; }
398 size_t GetByteSize() const { return m_opcode.GetByteSize(); }
400 /// Grants exclusive access to the disassembler and initializes it with the
401 /// given InstructionLLVMC and an optional ExecutionContext.
402 class DisassemblerScope {
403 std::shared_ptr<DisassemblerLLVMC> m_disasm;
406 explicit DisassemblerScope(
408 const lldb_private::ExecutionContext *exe_ctx = nullptr)
409 : m_disasm(i.m_disasm_wp.lock()) {
410 m_disasm->m_mutex.lock();
411 m_disasm->m_inst = &i;
412 m_disasm->m_exe_ctx = exe_ctx;
414 ~DisassemblerScope() { m_disasm->m_mutex.unlock(); }
416 /// Evaluates to true if this scope contains a valid disassembler.
417 operator bool() const { return static_cast<bool>(m_disasm); }
419 std::shared_ptr<DisassemblerLLVMC> operator->() { return m_disasm; }
422 static llvm::StringRef::const_iterator
423 ConsumeWhitespace(llvm::StringRef::const_iterator osi,
424 llvm::StringRef::const_iterator ose) {
439 static std::pair<bool, llvm::StringRef::const_iterator>
440 ConsumeChar(llvm::StringRef::const_iterator osi, const char c,
441 llvm::StringRef::const_iterator ose) {
444 osi = ConsumeWhitespace(osi, ose);
445 if (osi != ose && *osi == c) {
450 return std::make_pair(found, osi);
453 static std::pair<Operand, llvm::StringRef::const_iterator>
454 ParseRegisterName(llvm::StringRef::const_iterator osi,
455 llvm::StringRef::const_iterator ose) {
457 ret.m_type = Operand::Type::Register;
460 osi = ConsumeWhitespace(osi, ose);
463 if (*osi >= '0' && *osi <= '9') {
465 return std::make_pair(Operand(), osi);
469 } else if (*osi >= 'a' && *osi <= 'z') {
475 return std::make_pair(Operand(), osi);
477 ret.m_register = ConstString(str);
478 return std::make_pair(ret, osi);
482 return std::make_pair(Operand(), osi);
490 ret.m_register = ConstString(str);
491 return std::make_pair(ret, osi);
494 static std::pair<Operand, llvm::StringRef::const_iterator>
495 ParseImmediate(llvm::StringRef::const_iterator osi,
496 llvm::StringRef::const_iterator ose) {
498 ret.m_type = Operand::Type::Immediate;
502 osi = ConsumeWhitespace(osi, ose);
505 if (*osi >= '0' && *osi <= '9') {
507 } else if (*osi >= 'a' && *osi <= 'f') {
511 return std::make_pair(Operand(), osi);
517 return std::make_pair(Operand(), osi);
519 ret.m_immediate = strtoull(str.c_str(), nullptr, 0);
520 return std::make_pair(ret, osi);
523 if (!str.compare("0")) {
527 return std::make_pair(Operand(), osi);
533 return std::make_pair(Operand(), osi);
538 ret.m_negative = true;
540 return std::make_pair(Operand(), osi);
547 ret.m_immediate = strtoull(str.c_str(), nullptr, 0);
548 return std::make_pair(ret, osi);
552 static std::pair<Operand, llvm::StringRef::const_iterator>
553 ParseIntelIndexedAccess(llvm::StringRef::const_iterator osi,
554 llvm::StringRef::const_iterator ose) {
555 std::pair<Operand, llvm::StringRef::const_iterator> offset_and_iterator =
556 ParseImmediate(osi, ose);
557 if (offset_and_iterator.first.IsValid()) {
558 osi = offset_and_iterator.second;
562 std::tie(found, osi) = ConsumeChar(osi, '(', ose);
564 return std::make_pair(Operand(), osi);
567 std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
568 ParseRegisterName(osi, ose);
569 if (base_and_iterator.first.IsValid()) {
570 osi = base_and_iterator.second;
572 return std::make_pair(Operand(), osi);
575 std::tie(found, osi) = ConsumeChar(osi, ',', ose);
577 return std::make_pair(Operand(), osi);
580 std::pair<Operand, llvm::StringRef::const_iterator> index_and_iterator =
581 ParseRegisterName(osi, ose);
582 if (index_and_iterator.first.IsValid()) {
583 osi = index_and_iterator.second;
585 return std::make_pair(Operand(), osi);
588 std::tie(found, osi) = ConsumeChar(osi, ',', ose);
590 return std::make_pair(Operand(), osi);
593 std::pair<Operand, llvm::StringRef::const_iterator>
594 multiplier_and_iterator = ParseImmediate(osi, ose);
595 if (index_and_iterator.first.IsValid()) {
596 osi = index_and_iterator.second;
598 return std::make_pair(Operand(), osi);
601 std::tie(found, osi) = ConsumeChar(osi, ')', ose);
603 return std::make_pair(Operand(), osi);
607 product.m_type = Operand::Type::Product;
608 product.m_children.push_back(index_and_iterator.first);
609 product.m_children.push_back(multiplier_and_iterator.first);
612 index.m_type = Operand::Type::Sum;
613 index.m_children.push_back(base_and_iterator.first);
614 index.m_children.push_back(product);
616 if (offset_and_iterator.first.IsValid()) {
618 offset.m_type = Operand::Type::Sum;
619 offset.m_children.push_back(offset_and_iterator.first);
620 offset.m_children.push_back(index);
623 deref.m_type = Operand::Type::Dereference;
624 deref.m_children.push_back(offset);
625 return std::make_pair(deref, osi);
628 deref.m_type = Operand::Type::Dereference;
629 deref.m_children.push_back(index);
630 return std::make_pair(deref, osi);
635 static std::pair<Operand, llvm::StringRef::const_iterator>
636 ParseIntelDerefAccess(llvm::StringRef::const_iterator osi,
637 llvm::StringRef::const_iterator ose) {
638 std::pair<Operand, llvm::StringRef::const_iterator> offset_and_iterator =
639 ParseImmediate(osi, ose);
640 if (offset_and_iterator.first.IsValid()) {
641 osi = offset_and_iterator.second;
645 std::tie(found, osi) = ConsumeChar(osi, '(', ose);
647 return std::make_pair(Operand(), osi);
650 std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
651 ParseRegisterName(osi, ose);
652 if (base_and_iterator.first.IsValid()) {
653 osi = base_and_iterator.second;
655 return std::make_pair(Operand(), osi);
658 std::tie(found, osi) = ConsumeChar(osi, ')', ose);
660 return std::make_pair(Operand(), osi);
663 if (offset_and_iterator.first.IsValid()) {
665 offset.m_type = Operand::Type::Sum;
666 offset.m_children.push_back(offset_and_iterator.first);
667 offset.m_children.push_back(base_and_iterator.first);
670 deref.m_type = Operand::Type::Dereference;
671 deref.m_children.push_back(offset);
672 return std::make_pair(deref, osi);
675 deref.m_type = Operand::Type::Dereference;
676 deref.m_children.push_back(base_and_iterator.first);
677 return std::make_pair(deref, osi);
682 static std::pair<Operand, llvm::StringRef::const_iterator>
683 ParseARMOffsetAccess(llvm::StringRef::const_iterator osi,
684 llvm::StringRef::const_iterator ose) {
686 std::tie(found, osi) = ConsumeChar(osi, '[', ose);
688 return std::make_pair(Operand(), osi);
691 std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
692 ParseRegisterName(osi, ose);
693 if (base_and_iterator.first.IsValid()) {
694 osi = base_and_iterator.second;
696 return std::make_pair(Operand(), osi);
699 std::tie(found, osi) = ConsumeChar(osi, ',', ose);
701 return std::make_pair(Operand(), osi);
704 std::pair<Operand, llvm::StringRef::const_iterator> offset_and_iterator =
705 ParseImmediate(osi, ose);
706 if (offset_and_iterator.first.IsValid()) {
707 osi = offset_and_iterator.second;
710 std::tie(found, osi) = ConsumeChar(osi, ']', ose);
712 return std::make_pair(Operand(), osi);
716 offset.m_type = Operand::Type::Sum;
717 offset.m_children.push_back(offset_and_iterator.first);
718 offset.m_children.push_back(base_and_iterator.first);
721 deref.m_type = Operand::Type::Dereference;
722 deref.m_children.push_back(offset);
723 return std::make_pair(deref, osi);
727 static std::pair<Operand, llvm::StringRef::const_iterator>
728 ParseARMDerefAccess(llvm::StringRef::const_iterator osi,
729 llvm::StringRef::const_iterator ose) {
731 std::tie(found, osi) = ConsumeChar(osi, '[', ose);
733 return std::make_pair(Operand(), osi);
736 std::pair<Operand, llvm::StringRef::const_iterator> base_and_iterator =
737 ParseRegisterName(osi, ose);
738 if (base_and_iterator.first.IsValid()) {
739 osi = base_and_iterator.second;
741 return std::make_pair(Operand(), osi);
744 std::tie(found, osi) = ConsumeChar(osi, ']', ose);
746 return std::make_pair(Operand(), osi);
750 deref.m_type = Operand::Type::Dereference;
751 deref.m_children.push_back(base_and_iterator.first);
752 return std::make_pair(deref, osi);
755 static void DumpOperand(const Operand &op, Stream &s) {
757 case Operand::Type::Dereference:
759 DumpOperand(op.m_children[0], s);
761 case Operand::Type::Immediate:
765 s.PutCString(llvm::to_string(op.m_immediate));
767 case Operand::Type::Invalid:
768 s.PutCString("Invalid");
770 case Operand::Type::Product:
772 DumpOperand(op.m_children[0], s);
774 DumpOperand(op.m_children[1], s);
777 case Operand::Type::Register:
778 s.PutCString(op.m_register.AsCString());
780 case Operand::Type::Sum:
782 DumpOperand(op.m_children[0], s);
784 DumpOperand(op.m_children[1], s);
791 llvm::SmallVectorImpl<Instruction::Operand> &operands) override {
792 const char *operands_string = GetOperands(nullptr);
794 if (!operands_string) {
798 llvm::StringRef operands_ref(operands_string);
800 llvm::StringRef::const_iterator osi = operands_ref.begin();
801 llvm::StringRef::const_iterator ose = operands_ref.end();
805 llvm::StringRef::const_iterator iter;
807 if ((std::tie(operand, iter) = ParseIntelIndexedAccess(osi, ose),
808 operand.IsValid()) ||
809 (std::tie(operand, iter) = ParseIntelDerefAccess(osi, ose),
810 operand.IsValid()) ||
811 (std::tie(operand, iter) = ParseARMOffsetAccess(osi, ose),
812 operand.IsValid()) ||
813 (std::tie(operand, iter) = ParseARMDerefAccess(osi, ose),
814 operand.IsValid()) ||
815 (std::tie(operand, iter) = ParseRegisterName(osi, ose),
816 operand.IsValid()) ||
817 (std::tie(operand, iter) = ParseImmediate(osi, ose),
818 operand.IsValid())) {
820 operands.push_back(operand);
825 std::pair<bool, llvm::StringRef::const_iterator> found_and_iter =
826 ConsumeChar(osi, ',', ose);
827 if (found_and_iter.first) {
828 osi = found_and_iter.second;
831 osi = ConsumeWhitespace(osi, ose);
834 DisassemblerSP disasm_sp = m_disasm_wp.lock();
836 if (disasm_sp && operands.size() > 1) {
837 // TODO tie this into the MC Disassembler's notion of clobbers.
838 switch (disasm_sp->GetArchitecture().GetMachine()) {
841 case llvm::Triple::x86:
842 case llvm::Triple::x86_64:
843 operands[operands.size() - 1].m_clobbered = true;
845 case llvm::Triple::arm:
846 operands[0].m_clobbered = true;
852 lldb_private::GetLogIfAllCategoriesSet(LIBLLDB_LOG_PROCESS)) {
855 ss.Printf("[%s] expands to %zu operands:\n", operands_string,
857 for (const Operand &operand : operands) {
859 DumpOperand(operand, ss);
863 log->PutString(ss.GetString());
869 bool IsCall() override {
870 if (m_is_call == eLazyBoolCalculate) {
871 DisassemblerScope disasm(*this);
874 if (m_opcode.GetData(data)) {
875 bool is_alternate_isa;
876 lldb::addr_t pc = m_address.GetFileAddress();
878 DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
879 GetDisasmToUse(is_alternate_isa, disasm);
880 const uint8_t *opcode_data = data.GetDataStart();
881 const size_t opcode_data_len = data.GetByteSize();
883 const size_t inst_size =
884 mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
885 if (inst_size == 0) {
886 m_is_call = eLazyBoolNo;
888 if (mc_disasm_ptr->IsCall(inst))
889 m_is_call = eLazyBoolYes;
891 m_is_call = eLazyBoolNo;
896 return m_is_call == eLazyBoolYes;
900 std::weak_ptr<DisassemblerLLVMC> m_disasm_wp;
901 LazyBool m_does_branch;
902 LazyBool m_has_delay_slot;
905 bool m_using_file_addr;
908 DisassemblerLLVMC::MCDisasmInstance *
909 GetDisasmToUse(bool &is_alternate_isa, DisassemblerScope &disasm) {
910 is_alternate_isa = false;
912 if (disasm->m_alternate_disasm_up) {
913 const AddressClass address_class = GetAddressClass();
915 if (address_class == AddressClass::eCodeAlternateISA) {
916 is_alternate_isa = true;
917 return disasm->m_alternate_disasm_up.get();
920 return disasm->m_disasm_up.get();
926 std::unique_ptr<DisassemblerLLVMC::MCDisasmInstance>
927 DisassemblerLLVMC::MCDisasmInstance::Create(const char *triple, const char *cpu,
928 const char *features_str,
930 DisassemblerLLVMC &owner) {
931 using Instance = std::unique_ptr<DisassemblerLLVMC::MCDisasmInstance>;
934 const llvm::Target *curr_target =
935 llvm::TargetRegistry::lookupTarget(triple, Status);
939 std::unique_ptr<llvm::MCInstrInfo> instr_info_up(
940 curr_target->createMCInstrInfo());
944 std::unique_ptr<llvm::MCRegisterInfo> reg_info_up(
945 curr_target->createMCRegInfo(triple));
949 std::unique_ptr<llvm::MCSubtargetInfo> subtarget_info_up(
950 curr_target->createMCSubtargetInfo(triple, cpu, features_str));
951 if (!subtarget_info_up)
954 std::unique_ptr<llvm::MCAsmInfo> asm_info_up(
955 curr_target->createMCAsmInfo(*reg_info_up, triple));
959 std::unique_ptr<llvm::MCContext> context_up(
960 new llvm::MCContext(asm_info_up.get(), reg_info_up.get(), 0));
964 std::unique_ptr<llvm::MCDisassembler> disasm_up(
965 curr_target->createMCDisassembler(*subtarget_info_up, *context_up));
969 std::unique_ptr<llvm::MCRelocationInfo> rel_info_up(
970 curr_target->createMCRelocationInfo(triple, *context_up));
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));
980 unsigned asm_printer_variant =
981 flavor == ~0U ? asm_info_up->getAssemblerDialect() : flavor;
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)
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)));
997 DisassemblerLLVMC::MCDisasmInstance::MCDisasmInstance(
998 std::unique_ptr<llvm::MCInstrInfo> &&instr_info_up,
999 std::unique_ptr<llvm::MCRegisterInfo> &®_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);
1015 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;
1021 uint64_t new_inst_size;
1022 status = m_disasm_up->getInstruction(mc_inst, new_inst_size, data, pc,
1023 llvm::nulls(), llvm::nulls());
1024 if (status == llvm::MCDisassembler::Success)
1025 return new_inst_size;
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);
1036 m_instr_printer_up->setCommentStream(comments_stream);
1037 m_instr_printer_up->printInst(&mc_inst, inst_stream, llvm::StringRef(),
1038 *m_subtarget_info_up);
1039 m_instr_printer_up->setCommentStream(llvm::nulls());
1040 comments_stream.flush();
1042 static std::string g_newlines("\r\n");
1044 for (size_t newline_pos = 0;
1045 (newline_pos = comments_string.find_first_of(g_newlines, newline_pos)) !=
1046 comments_string.npos;
1048 comments_string.replace(comments_string.begin() + newline_pos,
1049 comments_string.begin() + newline_pos + 1, 1, ' ');
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) {
1058 m_instr_printer_up->setPrintHexStyle(llvm::HexStyle::C);
1061 m_instr_printer_up->setPrintHexStyle(llvm::HexStyle::Asm);
1066 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);
1072 bool DisassemblerLLVMC::MCDisasmInstance::HasDelaySlot(
1073 llvm::MCInst &mc_inst) const {
1074 return m_instr_info_up->get(mc_inst.getOpcode()).hasDelaySlot();
1077 bool DisassemblerLLVMC::MCDisasmInstance::IsCall(llvm::MCInst &mc_inst) const {
1078 return m_instr_info_up->get(mc_inst.getOpcode()).isCall();
1081 DisassemblerLLVMC::DisassemblerLLVMC(const ArchSpec &arch,
1082 const char *flavor_string)
1083 : Disassembler(arch, flavor_string), m_exe_ctx(NULL), m_inst(NULL),
1084 m_data_from_file(false) {
1085 if (!FlavorValidForArchSpec(arch, m_flavor.c_str())) {
1086 m_flavor.assign("default");
1089 unsigned flavor = ~0U;
1090 llvm::Triple triple = arch.GetTriple();
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") {
1098 } else if (m_flavor == "att") {
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");
1111 thumb_arch_name = "thumbv8.2a";
1113 thumb_arch.GetTriple().setArchName(llvm::StringRef(thumb_arch_name));
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
1121 if (triple.getArch() == llvm::Triple::arm &&
1122 triple.getSubArch() == llvm::Triple::NoSubArch)
1123 triple.setArchName("armv8.2a");
1125 std::string features_str = "";
1126 const char *triple_str = triple.getTriple().c_str();
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,";
1134 const char *cpu = "";
1136 switch (arch.GetCore()) {
1137 case ArchSpec::eCore_mips32:
1138 case ArchSpec::eCore_mips32el:
1141 case ArchSpec::eCore_mips32r2:
1142 case ArchSpec::eCore_mips32r2el:
1145 case ArchSpec::eCore_mips32r3:
1146 case ArchSpec::eCore_mips32r3el:
1149 case ArchSpec::eCore_mips32r5:
1150 case ArchSpec::eCore_mips32r5el:
1153 case ArchSpec::eCore_mips32r6:
1154 case ArchSpec::eCore_mips32r6el:
1157 case ArchSpec::eCore_mips64:
1158 case ArchSpec::eCore_mips64el:
1161 case ArchSpec::eCore_mips64r2:
1162 case ArchSpec::eCore_mips64r2el:
1165 case ArchSpec::eCore_mips64r3:
1166 case ArchSpec::eCore_mips64r3el:
1169 case ArchSpec::eCore_mips64r5:
1170 case ArchSpec::eCore_mips64r5el:
1173 case ArchSpec::eCore_mips64r6:
1174 case ArchSpec::eCore_mips64r6el:
1182 if (triple.getArch() == llvm::Triple::mips ||
1183 triple.getArch() == llvm::Triple::mipsel ||
1184 triple.getArch() == llvm::Triple::mips64 ||
1185 triple.getArch() == llvm::Triple::mips64el) {
1186 uint32_t arch_flags = arch.GetFlags();
1187 if (arch_flags & ArchSpec::eMIPSAse_msa)
1188 features_str += "+msa,";
1189 if (arch_flags & ArchSpec::eMIPSAse_dsp)
1190 features_str += "+dsp,";
1191 if (arch_flags & ArchSpec::eMIPSAse_dspr2)
1192 features_str += "+dspr2,";
1195 // If any AArch64 variant, enable the ARMv8.2 ISA extensions so we can
1196 // disassemble newer instructions.
1197 if (triple.getArch() == llvm::Triple::aarch64)
1198 features_str += "+v8.2a";
1200 // We use m_disasm_ap.get() to tell whether we are valid or not, so if this
1201 // isn't good for some reason, we won't be valid and FindPlugin will fail and
1202 // we won't get used.
1203 m_disasm_up = MCDisasmInstance::Create(triple_str, cpu, features_str.c_str(),
1206 llvm::Triple::ArchType llvm_arch = triple.getArch();
1208 // For arm CPUs that can execute arm or thumb instructions, also create a
1209 // thumb instruction disassembler.
1210 if (llvm_arch == llvm::Triple::arm) {
1211 std::string thumb_triple(thumb_arch.GetTriple().getTriple());
1212 m_alternate_disasm_up =
1213 MCDisasmInstance::Create(thumb_triple.c_str(), "", features_str.c_str(),
1215 if (!m_alternate_disasm_up)
1216 m_disasm_up.reset();
1218 } else if (llvm_arch == llvm::Triple::mips ||
1219 llvm_arch == llvm::Triple::mipsel ||
1220 llvm_arch == llvm::Triple::mips64 ||
1221 llvm_arch == llvm::Triple::mips64el) {
1222 /* Create alternate disassembler for MIPS16 and microMIPS */
1223 uint32_t arch_flags = arch.GetFlags();
1224 if (arch_flags & ArchSpec::eMIPSAse_mips16)
1225 features_str += "+mips16,";
1226 else if (arch_flags & ArchSpec::eMIPSAse_micromips)
1227 features_str += "+micromips,";
1229 m_alternate_disasm_up = MCDisasmInstance::Create(
1230 triple_str, cpu, features_str.c_str(), flavor, *this);
1231 if (!m_alternate_disasm_up)
1232 m_disasm_up.reset();
1236 DisassemblerLLVMC::~DisassemblerLLVMC() = default;
1238 Disassembler *DisassemblerLLVMC::CreateInstance(const ArchSpec &arch,
1239 const char *flavor) {
1240 if (arch.GetTriple().getArch() != llvm::Triple::UnknownArch) {
1241 std::unique_ptr<DisassemblerLLVMC> disasm_ap(
1242 new DisassemblerLLVMC(arch, flavor));
1244 if (disasm_ap.get() && disasm_ap->IsValid())
1245 return disasm_ap.release();
1250 size_t DisassemblerLLVMC::DecodeInstructions(const Address &base_addr,
1251 const DataExtractor &data,
1252 lldb::offset_t data_offset,
1253 size_t num_instructions,
1254 bool append, bool data_from_file) {
1256 m_instruction_list.Clear();
1261 m_data_from_file = data_from_file;
1262 uint32_t data_cursor = data_offset;
1263 const size_t data_byte_size = data.GetByteSize();
1264 uint32_t instructions_parsed = 0;
1265 Address inst_addr(base_addr);
1267 while (data_cursor < data_byte_size &&
1268 instructions_parsed < num_instructions) {
1270 AddressClass address_class = AddressClass::eCode;
1272 if (m_alternate_disasm_up)
1273 address_class = inst_addr.GetAddressClass();
1275 InstructionSP inst_sp(
1276 new InstructionLLVMC(*this, inst_addr, address_class));
1281 uint32_t inst_size = inst_sp->Decode(*this, data, data_cursor);
1286 m_instruction_list.Append(inst_sp);
1287 data_cursor += inst_size;
1288 inst_addr.Slide(inst_size);
1289 instructions_parsed++;
1292 return data_cursor - data_offset;
1295 void DisassemblerLLVMC::Initialize() {
1296 PluginManager::RegisterPlugin(GetPluginNameStatic(),
1297 "Disassembler that uses LLVM MC to disassemble "
1298 "i386, x86_64, ARM, and ARM64.",
1301 llvm::InitializeAllTargetInfos();
1302 llvm::InitializeAllTargetMCs();
1303 llvm::InitializeAllAsmParsers();
1304 llvm::InitializeAllDisassemblers();
1307 void DisassemblerLLVMC::Terminate() {
1308 PluginManager::UnregisterPlugin(CreateInstance);
1311 ConstString DisassemblerLLVMC::GetPluginNameStatic() {
1312 static ConstString g_name("llvm-mc");
1316 int DisassemblerLLVMC::OpInfoCallback(void *disassembler, uint64_t pc,
1317 uint64_t offset, uint64_t size,
1318 int tag_type, void *tag_bug) {
1319 return static_cast<DisassemblerLLVMC *>(disassembler)
1320 ->OpInfo(pc, offset, size, tag_type, tag_bug);
1323 const char *DisassemblerLLVMC::SymbolLookupCallback(void *disassembler,
1325 uint64_t *type, uint64_t pc,
1326 const char **name) {
1327 return static_cast<DisassemblerLLVMC *>(disassembler)
1328 ->SymbolLookup(value, type, pc, name);
1331 bool DisassemblerLLVMC::FlavorValidForArchSpec(
1332 const lldb_private::ArchSpec &arch, const char *flavor) {
1333 llvm::Triple triple = arch.GetTriple();
1334 if (flavor == NULL || strcmp(flavor, "default") == 0)
1337 if (triple.getArch() == llvm::Triple::x86 ||
1338 triple.getArch() == llvm::Triple::x86_64) {
1339 return strcmp(flavor, "intel") == 0 || strcmp(flavor, "att") == 0;
1344 bool DisassemblerLLVMC::IsValid() const { return m_disasm_up.operator bool(); }
1346 int DisassemblerLLVMC::OpInfo(uint64_t PC, uint64_t Offset, uint64_t Size,
1347 int tag_type, void *tag_bug) {
1352 memset(tag_bug, 0, sizeof(::LLVMOpInfo1));
1358 const char *DisassemblerLLVMC::SymbolLookup(uint64_t value, uint64_t *type_ptr,
1359 uint64_t pc, const char **name) {
1361 if (m_exe_ctx && m_inst) {
1362 // std::string remove_this_prior_to_checkin;
1363 Target *target = m_exe_ctx ? m_exe_ctx->GetTargetPtr() : NULL;
1364 Address value_so_addr;
1366 if (m_inst->UsingFileAddress()) {
1367 ModuleSP module_sp(m_inst->GetAddress().GetModule());
1369 module_sp->ResolveFileAddress(value, value_so_addr);
1370 module_sp->ResolveFileAddress(pc, pc_so_addr);
1372 } else if (target && !target->GetSectionLoadList().IsEmpty()) {
1373 target->GetSectionLoadList().ResolveLoadAddress(value, value_so_addr);
1374 target->GetSectionLoadList().ResolveLoadAddress(pc, pc_so_addr);
1377 SymbolContext sym_ctx;
1378 const SymbolContextItem resolve_scope =
1379 eSymbolContextFunction | eSymbolContextSymbol;
1380 if (pc_so_addr.IsValid() && pc_so_addr.GetModule()) {
1381 pc_so_addr.GetModule()->ResolveSymbolContextForAddress(
1382 pc_so_addr, resolve_scope, sym_ctx);
1385 if (value_so_addr.IsValid() && value_so_addr.GetSection()) {
1388 bool format_omitting_current_func_name = false;
1389 if (sym_ctx.symbol || sym_ctx.function) {
1391 if (sym_ctx.GetAddressRange(resolve_scope, 0, false, range) &&
1392 range.GetBaseAddress().IsValid() &&
1393 range.ContainsLoadAddress(value_so_addr, target)) {
1394 format_omitting_current_func_name = true;
1398 // If the "value" address (the target address we're symbolicating) is
1399 // inside the same SymbolContext as the current instruction pc
1400 // (pc_so_addr), don't print the full function name - just print it
1401 // with DumpStyleNoFunctionName style, e.g. "<+36>".
1402 if (format_omitting_current_func_name) {
1403 value_so_addr.Dump(&ss, target, Address::DumpStyleNoFunctionName,
1404 Address::DumpStyleSectionNameOffset);
1408 Address::DumpStyleResolvedDescriptionNoFunctionArguments,
1409 Address::DumpStyleSectionNameOffset);
1412 if (!ss.GetString().empty()) {
1413 // If Address::Dump returned a multi-line description, most commonly
1414 // seen when we have multiple levels of inlined functions at an
1415 // address, only show the first line.
1416 std::string str = ss.GetString();
1417 size_t first_eol_char = str.find_first_of("\r\n");
1418 if (first_eol_char != std::string::npos) {
1419 str.erase(first_eol_char);
1421 m_inst->AppendComment(str);
1427 *type_ptr = LLVMDisassembler_ReferenceType_InOut_None;
1432 //------------------------------------------------------------------
1433 // PluginInterface protocol
1434 //------------------------------------------------------------------
1435 ConstString DisassemblerLLVMC::GetPluginName() { return GetPluginNameStatic(); }
1437 uint32_t DisassemblerLLVMC::GetPluginVersion() { return 1; }