1 //===-- DisassemblerLLVMC.cpp -----------------------------------*- C++ -*-===//
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
7 //===----------------------------------------------------------------------===//
9 #include "DisassemblerLLVMC.h"
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/Support/ErrorHandling.h"
24 #include "llvm/Support/ScopedPrinter.h"
25 #include "llvm/Support/TargetRegistry.h"
26 #include "llvm/Support/TargetSelect.h"
28 #include "lldb/Core/Address.h"
29 #include "lldb/Core/Module.h"
30 #include "lldb/Symbol/SymbolContext.h"
31 #include "lldb/Target/ExecutionContext.h"
32 #include "lldb/Target/Process.h"
33 #include "lldb/Target/RegisterContext.h"
34 #include "lldb/Target/SectionLoadList.h"
35 #include "lldb/Target/StackFrame.h"
36 #include "lldb/Target/Target.h"
37 #include "lldb/Utility/DataExtractor.h"
38 #include "lldb/Utility/Log.h"
39 #include "lldb/Utility/RegularExpression.h"
40 #include "lldb/Utility/Stream.h"
43 using namespace lldb_private;
45 class DisassemblerLLVMC::MCDisasmInstance {
47 static std::unique_ptr<MCDisasmInstance>
48 Create(const char *triple, const char *cpu, const char *features_str,
49 unsigned flavor, DisassemblerLLVMC &owner);
51 ~MCDisasmInstance() = default;
53 uint64_t GetMCInst(const uint8_t *opcode_data, size_t opcode_data_len,
54 lldb::addr_t pc, llvm::MCInst &mc_inst) const;
55 void PrintMCInst(llvm::MCInst &mc_inst, std::string &inst_string,
56 std::string &comments_string);
57 void SetStyle(bool use_hex_immed, HexImmediateStyle hex_style);
58 bool CanBranch(llvm::MCInst &mc_inst) const;
59 bool HasDelaySlot(llvm::MCInst &mc_inst) const;
60 bool IsCall(llvm::MCInst &mc_inst) const;
63 MCDisasmInstance(std::unique_ptr<llvm::MCInstrInfo> &&instr_info_up,
64 std::unique_ptr<llvm::MCRegisterInfo> &®_info_up,
65 std::unique_ptr<llvm::MCSubtargetInfo> &&subtarget_info_up,
66 std::unique_ptr<llvm::MCAsmInfo> &&asm_info_up,
67 std::unique_ptr<llvm::MCContext> &&context_up,
68 std::unique_ptr<llvm::MCDisassembler> &&disasm_up,
69 std::unique_ptr<llvm::MCInstPrinter> &&instr_printer_up);
71 std::unique_ptr<llvm::MCInstrInfo> m_instr_info_up;
72 std::unique_ptr<llvm::MCRegisterInfo> m_reg_info_up;
73 std::unique_ptr<llvm::MCSubtargetInfo> m_subtarget_info_up;
74 std::unique_ptr<llvm::MCAsmInfo> m_asm_info_up;
75 std::unique_ptr<llvm::MCContext> m_context_up;
76 std::unique_ptr<llvm::MCDisassembler> m_disasm_up;
77 std::unique_ptr<llvm::MCInstPrinter> m_instr_printer_up;
80 class InstructionLLVMC : public lldb_private::Instruction {
82 InstructionLLVMC(DisassemblerLLVMC &disasm,
83 const lldb_private::Address &address,
84 AddressClass addr_class)
85 : Instruction(address, addr_class),
86 m_disasm_wp(std::static_pointer_cast<DisassemblerLLVMC>(
87 disasm.shared_from_this())),
88 m_does_branch(eLazyBoolCalculate), m_has_delay_slot(eLazyBoolCalculate),
89 m_is_call(eLazyBoolCalculate), m_is_valid(false),
90 m_using_file_addr(false) {}
92 ~InstructionLLVMC() override = default;
94 bool DoesBranch() override {
95 if (m_does_branch == eLazyBoolCalculate) {
96 DisassemblerScope disasm(*this);
99 if (m_opcode.GetData(data)) {
100 bool is_alternate_isa;
101 lldb::addr_t pc = m_address.GetFileAddress();
103 DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
104 GetDisasmToUse(is_alternate_isa, disasm);
105 const uint8_t *opcode_data = data.GetDataStart();
106 const size_t opcode_data_len = data.GetByteSize();
108 const size_t inst_size =
109 mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
110 // Be conservative, if we didn't understand the instruction, say it
113 m_does_branch = eLazyBoolYes;
115 const bool can_branch = mc_disasm_ptr->CanBranch(inst);
117 m_does_branch = eLazyBoolYes;
119 m_does_branch = eLazyBoolNo;
124 return m_does_branch == eLazyBoolYes;
127 bool HasDelaySlot() override {
128 if (m_has_delay_slot == eLazyBoolCalculate) {
129 DisassemblerScope disasm(*this);
132 if (m_opcode.GetData(data)) {
133 bool is_alternate_isa;
134 lldb::addr_t pc = m_address.GetFileAddress();
136 DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
137 GetDisasmToUse(is_alternate_isa, disasm);
138 const uint8_t *opcode_data = data.GetDataStart();
139 const size_t opcode_data_len = data.GetByteSize();
141 const size_t inst_size =
142 mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
143 // if we didn't understand the instruction, say it doesn't have a
146 m_has_delay_slot = eLazyBoolNo;
148 const bool has_delay_slot = mc_disasm_ptr->HasDelaySlot(inst);
150 m_has_delay_slot = eLazyBoolYes;
152 m_has_delay_slot = eLazyBoolNo;
157 return m_has_delay_slot == eLazyBoolYes;
160 DisassemblerLLVMC::MCDisasmInstance *GetDisasmToUse(bool &is_alternate_isa) {
161 DisassemblerScope disasm(*this);
162 return GetDisasmToUse(is_alternate_isa, disasm);
165 size_t Decode(const lldb_private::Disassembler &disassembler,
166 const lldb_private::DataExtractor &data,
167 lldb::offset_t data_offset) override {
168 // All we have to do is read the opcode which can be easy for some
171 DisassemblerScope disasm(*this);
173 const ArchSpec &arch = disasm->GetArchitecture();
174 const lldb::ByteOrder byte_order = data.GetByteOrder();
176 const uint32_t min_op_byte_size = arch.GetMinimumOpcodeByteSize();
177 const uint32_t max_op_byte_size = arch.GetMaximumOpcodeByteSize();
178 if (min_op_byte_size == max_op_byte_size) {
179 // Fixed size instructions, just read that amount of data.
180 if (!data.ValidOffsetForDataOfSize(data_offset, min_op_byte_size))
183 switch (min_op_byte_size) {
185 m_opcode.SetOpcode8(data.GetU8(&data_offset), byte_order);
190 m_opcode.SetOpcode16(data.GetU16(&data_offset), byte_order);
195 m_opcode.SetOpcode32(data.GetU32(&data_offset), byte_order);
200 m_opcode.SetOpcode64(data.GetU64(&data_offset), byte_order);
205 m_opcode.SetOpcodeBytes(data.PeekData(data_offset, min_op_byte_size),
212 bool is_alternate_isa = false;
213 DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr =
214 GetDisasmToUse(is_alternate_isa, disasm);
216 const llvm::Triple::ArchType machine = arch.GetMachine();
217 if (machine == llvm::Triple::arm || machine == llvm::Triple::thumb) {
218 if (machine == llvm::Triple::thumb || is_alternate_isa) {
219 uint32_t thumb_opcode = data.GetU16(&data_offset);
220 if ((thumb_opcode & 0xe000) != 0xe000 ||
221 ((thumb_opcode & 0x1800u) == 0)) {
222 m_opcode.SetOpcode16(thumb_opcode, byte_order);
226 thumb_opcode |= data.GetU16(&data_offset);
227 m_opcode.SetOpcode16_2(thumb_opcode, byte_order);
231 m_opcode.SetOpcode32(data.GetU32(&data_offset), byte_order);
235 // The opcode isn't evenly sized, so we need to actually use the llvm
236 // disassembler to parse it and get the size.
237 uint8_t *opcode_data =
238 const_cast<uint8_t *>(data.PeekData(data_offset, 1));
239 const size_t opcode_data_len = data.BytesLeft(data_offset);
240 const addr_t pc = m_address.GetFileAddress();
243 const size_t inst_size =
244 mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
248 m_opcode.SetOpcodeBytes(opcode_data, inst_size);
253 return m_opcode.GetByteSize();
258 void AppendComment(std::string &description) {
259 if (m_comment.empty())
260 m_comment.swap(description);
262 m_comment.append(", ");
263 m_comment.append(description);
267 void CalculateMnemonicOperandsAndComment(
268 const lldb_private::ExecutionContext *exe_ctx) override {
270 const AddressClass address_class = GetAddressClass();
272 if (m_opcode.GetData(data)) {
273 std::string out_string;
274 std::string comment_string;
276 DisassemblerScope disasm(*this, exe_ctx);
278 DisassemblerLLVMC::MCDisasmInstance *mc_disasm_ptr;
280 if (address_class == AddressClass::eCodeAlternateISA)
281 mc_disasm_ptr = disasm->m_alternate_disasm_up.get();
283 mc_disasm_ptr = disasm->m_disasm_up.get();
285 lldb::addr_t pc = m_address.GetFileAddress();
286 m_using_file_addr = true;
288 const bool data_from_file = disasm->m_data_from_file;
289 bool use_hex_immediates = true;
290 Disassembler::HexImmediateStyle hex_style = Disassembler::eHexStyleC;
293 Target *target = exe_ctx->GetTargetPtr();
295 use_hex_immediates = target->GetUseHexImmediates();
296 hex_style = target->GetHexImmediateStyle();
298 if (!data_from_file) {
299 const lldb::addr_t load_addr = m_address.GetLoadAddress(target);
300 if (load_addr != LLDB_INVALID_ADDRESS) {
302 m_using_file_addr = false;
308 const uint8_t *opcode_data = data.GetDataStart();
309 const size_t opcode_data_len = data.GetByteSize();
312 mc_disasm_ptr->GetMCInst(opcode_data, opcode_data_len, pc, inst);
315 mc_disasm_ptr->SetStyle(use_hex_immediates, hex_style);
316 mc_disasm_ptr->PrintMCInst(inst, out_string, comment_string);
318 if (!comment_string.empty()) {
319 AppendComment(comment_string);
323 if (inst_size == 0) {
324 m_comment.assign("unknown opcode");
325 inst_size = m_opcode.GetByteSize();
326 StreamString mnemonic_strm;
327 lldb::offset_t offset = 0;
328 lldb::ByteOrder byte_order = data.GetByteOrder();
331 const uint8_t uval8 = data.GetU8(&offset);
332 m_opcode.SetOpcode8(uval8, byte_order);
333 m_opcode_name.assign(".byte");
334 mnemonic_strm.Printf("0x%2.2x", uval8);
337 const uint16_t uval16 = data.GetU16(&offset);
338 m_opcode.SetOpcode16(uval16, byte_order);
339 m_opcode_name.assign(".short");
340 mnemonic_strm.Printf("0x%4.4x", uval16);
343 const uint32_t uval32 = data.GetU32(&offset);
344 m_opcode.SetOpcode32(uval32, byte_order);
345 m_opcode_name.assign(".long");
346 mnemonic_strm.Printf("0x%8.8x", uval32);
349 const uint64_t uval64 = data.GetU64(&offset);
350 m_opcode.SetOpcode64(uval64, byte_order);
351 m_opcode_name.assign(".quad");
352 mnemonic_strm.Printf("0x%16.16" PRIx64, uval64);
358 const uint8_t *bytes = data.PeekData(offset, inst_size);
359 if (bytes == nullptr)
361 m_opcode_name.assign(".byte");
362 m_opcode.SetOpcodeBytes(bytes, inst_size);
363 mnemonic_strm.Printf("0x%2.2x", bytes[0]);
364 for (uint32_t i = 1; i < inst_size; ++i)
365 mnemonic_strm.Printf(" 0x%2.2x", bytes[i]);
369 m_mnemonics = mnemonic_strm.GetString();
372 if (m_does_branch == eLazyBoolCalculate) {
373 const bool can_branch = mc_disasm_ptr->CanBranch(inst);
375 m_does_branch = eLazyBoolYes;
377 m_does_branch = eLazyBoolNo;
381 static RegularExpression s_regex(
382 llvm::StringRef("[ \t]*([^ ^\t]+)[ \t]*([^ ^\t].*)?"));
384 RegularExpression::Match matches(3);
386 if (s_regex.Execute(out_string, &matches)) {
387 matches.GetMatchAtIndex(out_string.c_str(), 1, m_opcode_name);
388 matches.GetMatchAtIndex(out_string.c_str(), 2, m_mnemonics);
394 bool IsValid() const { return m_is_valid; }
396 bool UsingFileAddress() const { return m_using_file_addr; }
397 size_t GetByteSize() const { return m_opcode.GetByteSize(); }
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;
405 explicit DisassemblerScope(
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;
413 ~DisassemblerScope() { m_disasm->m_mutex.unlock(); }
415 /// Evaluates to true if this scope contains a valid disassembler.
416 operator bool() const { return static_cast<bool>(m_disasm); }
418 std::shared_ptr<DisassemblerLLVMC> operator->() { return m_disasm; }
421 static llvm::StringRef::const_iterator
422 ConsumeWhitespace(llvm::StringRef::const_iterator osi,
423 llvm::StringRef::const_iterator ose) {
438 static std::pair<bool, llvm::StringRef::const_iterator>
439 ConsumeChar(llvm::StringRef::const_iterator osi, const char c,
440 llvm::StringRef::const_iterator ose) {
443 osi = ConsumeWhitespace(osi, ose);
444 if (osi != ose && *osi == c) {
449 return std::make_pair(found, osi);
452 static std::pair<Operand, llvm::StringRef::const_iterator>
453 ParseRegisterName(llvm::StringRef::const_iterator osi,
454 llvm::StringRef::const_iterator ose) {
456 ret.m_type = Operand::Type::Register;
459 osi = ConsumeWhitespace(osi, ose);
462 if (*osi >= '0' && *osi <= '9') {
464 return std::make_pair(Operand(), osi);
468 } else if (*osi >= 'a' && *osi <= 'z') {
474 return std::make_pair(Operand(), osi);
476 ret.m_register = ConstString(str);
477 return std::make_pair(ret, osi);
481 return std::make_pair(Operand(), osi);
489 ret.m_register = ConstString(str);
490 return std::make_pair(ret, osi);
493 static std::pair<Operand, llvm::StringRef::const_iterator>
494 ParseImmediate(llvm::StringRef::const_iterator osi,
495 llvm::StringRef::const_iterator ose) {
497 ret.m_type = Operand::Type::Immediate;
501 osi = ConsumeWhitespace(osi, ose);
504 if (*osi >= '0' && *osi <= '9') {
506 } else if (*osi >= 'a' && *osi <= 'f') {
510 return std::make_pair(Operand(), osi);
516 return std::make_pair(Operand(), osi);
518 ret.m_immediate = strtoull(str.c_str(), nullptr, 0);
519 return std::make_pair(ret, osi);
522 if (!str.compare("0")) {
526 return std::make_pair(Operand(), osi);
532 return std::make_pair(Operand(), osi);
537 ret.m_negative = true;
539 return std::make_pair(Operand(), osi);
546 ret.m_immediate = strtoull(str.c_str(), nullptr, 0);
547 return std::make_pair(ret, osi);
551 static std::pair<Operand, llvm::StringRef::const_iterator>
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;
561 std::tie(found, osi) = ConsumeChar(osi, '(', ose);
563 return std::make_pair(Operand(), osi);
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;
571 return std::make_pair(Operand(), osi);
574 std::tie(found, osi) = ConsumeChar(osi, ',', ose);
576 return std::make_pair(Operand(), osi);
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;
584 return std::make_pair(Operand(), osi);
587 std::tie(found, osi) = ConsumeChar(osi, ',', ose);
589 return std::make_pair(Operand(), osi);
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;
597 return std::make_pair(Operand(), osi);
600 std::tie(found, osi) = ConsumeChar(osi, ')', ose);
602 return std::make_pair(Operand(), osi);
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);
611 index.m_type = Operand::Type::Sum;
612 index.m_children.push_back(base_and_iterator.first);
613 index.m_children.push_back(product);
615 if (offset_and_iterator.first.IsValid()) {
617 offset.m_type = Operand::Type::Sum;
618 offset.m_children.push_back(offset_and_iterator.first);
619 offset.m_children.push_back(index);
622 deref.m_type = Operand::Type::Dereference;
623 deref.m_children.push_back(offset);
624 return std::make_pair(deref, osi);
627 deref.m_type = Operand::Type::Dereference;
628 deref.m_children.push_back(index);
629 return std::make_pair(deref, osi);
634 static std::pair<Operand, llvm::StringRef::const_iterator>
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;
644 std::tie(found, osi) = ConsumeChar(osi, '(', ose);
646 return std::make_pair(Operand(), osi);
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;
654 return std::make_pair(Operand(), osi);
657 std::tie(found, osi) = ConsumeChar(osi, ')', ose);
659 return std::make_pair(Operand(), osi);
662 if (offset_and_iterator.first.IsValid()) {
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);
669 deref.m_type = Operand::Type::Dereference;
670 deref.m_children.push_back(offset);
671 return std::make_pair(deref, osi);
674 deref.m_type = Operand::Type::Dereference;
675 deref.m_children.push_back(base_and_iterator.first);
676 return std::make_pair(deref, osi);
681 static std::pair<Operand, llvm::StringRef::const_iterator>
682 ParseARMOffsetAccess(llvm::StringRef::const_iterator osi,
683 llvm::StringRef::const_iterator ose) {
685 std::tie(found, osi) = ConsumeChar(osi, '[', ose);
687 return std::make_pair(Operand(), osi);
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;
695 return std::make_pair(Operand(), osi);
698 std::tie(found, osi) = ConsumeChar(osi, ',', ose);
700 return std::make_pair(Operand(), osi);
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;
709 std::tie(found, osi) = ConsumeChar(osi, ']', ose);
711 return std::make_pair(Operand(), osi);
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);
720 deref.m_type = Operand::Type::Dereference;
721 deref.m_children.push_back(offset);
722 return std::make_pair(deref, osi);
726 static std::pair<Operand, llvm::StringRef::const_iterator>
727 ParseARMDerefAccess(llvm::StringRef::const_iterator osi,
728 llvm::StringRef::const_iterator ose) {
730 std::tie(found, osi) = ConsumeChar(osi, '[', ose);
732 return std::make_pair(Operand(), osi);
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;
740 return std::make_pair(Operand(), osi);
743 std::tie(found, osi) = ConsumeChar(osi, ']', ose);
745 return std::make_pair(Operand(), osi);
749 deref.m_type = Operand::Type::Dereference;
750 deref.m_children.push_back(base_and_iterator.first);
751 return std::make_pair(deref, osi);
754 static void DumpOperand(const Operand &op, Stream &s) {
756 case Operand::Type::Dereference:
758 DumpOperand(op.m_children[0], s);
760 case Operand::Type::Immediate:
764 s.PutCString(llvm::to_string(op.m_immediate));
766 case Operand::Type::Invalid:
767 s.PutCString("Invalid");
769 case Operand::Type::Product:
771 DumpOperand(op.m_children[0], s);
773 DumpOperand(op.m_children[1], s);
776 case Operand::Type::Register:
777 s.PutCString(op.m_register.AsCString());
779 case Operand::Type::Sum:
781 DumpOperand(op.m_children[0], s);
783 DumpOperand(op.m_children[1], s);
790 llvm::SmallVectorImpl<Instruction::Operand> &operands) override {
791 const char *operands_string = GetOperands(nullptr);
793 if (!operands_string) {
797 llvm::StringRef operands_ref(operands_string);
799 llvm::StringRef::const_iterator osi = operands_ref.begin();
800 llvm::StringRef::const_iterator ose = operands_ref.end();
804 llvm::StringRef::const_iterator iter;
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())) {
819 operands.push_back(operand);
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;
830 osi = ConsumeWhitespace(osi, ose);
833 DisassemblerSP disasm_sp = m_disasm_wp.lock();
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()) {
840 case llvm::Triple::x86:
841 case llvm::Triple::x86_64:
842 operands[operands.size() - 1].m_clobbered = true;
844 case llvm::Triple::arm:
845 operands[0].m_clobbered = true;
851 lldb_private::GetLogIfAllCategoriesSet(LIBLLDB_LOG_PROCESS)) {
854 ss.Printf("[%s] expands to %zu operands:\n", operands_string,
856 for (const Operand &operand : operands) {
858 DumpOperand(operand, ss);
862 log->PutString(ss.GetString());
868 bool IsCall() override {
869 if (m_is_call == eLazyBoolCalculate) {
870 DisassemblerScope disasm(*this);
873 if (m_opcode.GetData(data)) {
874 bool is_alternate_isa;
875 lldb::addr_t pc = m_address.GetFileAddress();
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();
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;
887 if (mc_disasm_ptr->IsCall(inst))
888 m_is_call = eLazyBoolYes;
890 m_is_call = eLazyBoolNo;
895 return m_is_call == eLazyBoolYes;
899 std::weak_ptr<DisassemblerLLVMC> m_disasm_wp;
900 LazyBool m_does_branch;
901 LazyBool m_has_delay_slot;
904 bool m_using_file_addr;
907 DisassemblerLLVMC::MCDisasmInstance *
908 GetDisasmToUse(bool &is_alternate_isa, DisassemblerScope &disasm) {
909 is_alternate_isa = false;
911 if (disasm->m_alternate_disasm_up) {
912 const AddressClass address_class = GetAddressClass();
914 if (address_class == AddressClass::eCodeAlternateISA) {
915 is_alternate_isa = true;
916 return disasm->m_alternate_disasm_up.get();
919 return disasm->m_disasm_up.get();
925 std::unique_ptr<DisassemblerLLVMC::MCDisasmInstance>
926 DisassemblerLLVMC::MCDisasmInstance::Create(const char *triple, const char *cpu,
927 const char *features_str,
929 DisassemblerLLVMC &owner) {
930 using Instance = std::unique_ptr<DisassemblerLLVMC::MCDisasmInstance>;
933 const llvm::Target *curr_target =
934 llvm::TargetRegistry::lookupTarget(triple, Status);
938 std::unique_ptr<llvm::MCInstrInfo> instr_info_up(
939 curr_target->createMCInstrInfo());
943 std::unique_ptr<llvm::MCRegisterInfo> reg_info_up(
944 curr_target->createMCRegInfo(triple));
948 std::unique_ptr<llvm::MCSubtargetInfo> subtarget_info_up(
949 curr_target->createMCSubtargetInfo(triple, cpu, features_str));
950 if (!subtarget_info_up)
953 std::unique_ptr<llvm::MCAsmInfo> asm_info_up(
954 curr_target->createMCAsmInfo(*reg_info_up, triple));
958 std::unique_ptr<llvm::MCContext> context_up(
959 new llvm::MCContext(asm_info_up.get(), reg_info_up.get(), nullptr));
963 std::unique_ptr<llvm::MCDisassembler> disasm_up(
964 curr_target->createMCDisassembler(*subtarget_info_up, *context_up));
968 std::unique_ptr<llvm::MCRelocationInfo> rel_info_up(
969 curr_target->createMCRelocationInfo(triple, *context_up));
973 std::unique_ptr<llvm::MCSymbolizer> symbolizer_up(
974 curr_target->createMCSymbolizer(
975 triple, nullptr, DisassemblerLLVMC::SymbolLookupCallback, &owner,
976 context_up.get(), std::move(rel_info_up)));
977 disasm_up->setSymbolizer(std::move(symbolizer_up));
979 unsigned asm_printer_variant =
980 flavor == ~0U ? asm_info_up->getAssemblerDialect() : flavor;
982 std::unique_ptr<llvm::MCInstPrinter> instr_printer_up(
983 curr_target->createMCInstPrinter(llvm::Triple{triple},
984 asm_printer_variant, *asm_info_up,
985 *instr_info_up, *reg_info_up));
986 if (!instr_printer_up)
990 new MCDisasmInstance(std::move(instr_info_up), std::move(reg_info_up),
991 std::move(subtarget_info_up), std::move(asm_info_up),
992 std::move(context_up), std::move(disasm_up),
993 std::move(instr_printer_up)));
996 DisassemblerLLVMC::MCDisasmInstance::MCDisasmInstance(
997 std::unique_ptr<llvm::MCInstrInfo> &&instr_info_up,
998 std::unique_ptr<llvm::MCRegisterInfo> &®_info_up,
999 std::unique_ptr<llvm::MCSubtargetInfo> &&subtarget_info_up,
1000 std::unique_ptr<llvm::MCAsmInfo> &&asm_info_up,
1001 std::unique_ptr<llvm::MCContext> &&context_up,
1002 std::unique_ptr<llvm::MCDisassembler> &&disasm_up,
1003 std::unique_ptr<llvm::MCInstPrinter> &&instr_printer_up)
1004 : m_instr_info_up(std::move(instr_info_up)),
1005 m_reg_info_up(std::move(reg_info_up)),
1006 m_subtarget_info_up(std::move(subtarget_info_up)),
1007 m_asm_info_up(std::move(asm_info_up)),
1008 m_context_up(std::move(context_up)), m_disasm_up(std::move(disasm_up)),
1009 m_instr_printer_up(std::move(instr_printer_up)) {
1010 assert(m_instr_info_up && m_reg_info_up && m_subtarget_info_up &&
1011 m_asm_info_up && m_context_up && m_disasm_up && m_instr_printer_up);
1014 uint64_t DisassemblerLLVMC::MCDisasmInstance::GetMCInst(
1015 const uint8_t *opcode_data, size_t opcode_data_len, lldb::addr_t pc,
1016 llvm::MCInst &mc_inst) const {
1017 llvm::ArrayRef<uint8_t> data(opcode_data, opcode_data_len);
1018 llvm::MCDisassembler::DecodeStatus status;
1020 uint64_t new_inst_size;
1021 status = m_disasm_up->getInstruction(mc_inst, new_inst_size, data, pc,
1022 llvm::nulls(), llvm::nulls());
1023 if (status == llvm::MCDisassembler::Success)
1024 return new_inst_size;
1029 void DisassemblerLLVMC::MCDisasmInstance::PrintMCInst(
1030 llvm::MCInst &mc_inst, std::string &inst_string,
1031 std::string &comments_string) {
1032 llvm::raw_string_ostream inst_stream(inst_string);
1033 llvm::raw_string_ostream comments_stream(comments_string);
1035 m_instr_printer_up->setCommentStream(comments_stream);
1036 m_instr_printer_up->printInst(&mc_inst, inst_stream, llvm::StringRef(),
1037 *m_subtarget_info_up);
1038 m_instr_printer_up->setCommentStream(llvm::nulls());
1039 comments_stream.flush();
1041 static std::string g_newlines("\r\n");
1043 for (size_t newline_pos = 0;
1044 (newline_pos = comments_string.find_first_of(g_newlines, newline_pos)) !=
1045 comments_string.npos;
1047 comments_string.replace(comments_string.begin() + newline_pos,
1048 comments_string.begin() + newline_pos + 1, 1, ' ');
1052 void DisassemblerLLVMC::MCDisasmInstance::SetStyle(
1053 bool use_hex_immed, HexImmediateStyle hex_style) {
1054 m_instr_printer_up->setPrintImmHex(use_hex_immed);
1055 switch (hex_style) {
1057 m_instr_printer_up->setPrintHexStyle(llvm::HexStyle::C);
1060 m_instr_printer_up->setPrintHexStyle(llvm::HexStyle::Asm);
1065 bool DisassemblerLLVMC::MCDisasmInstance::CanBranch(
1066 llvm::MCInst &mc_inst) const {
1067 return m_instr_info_up->get(mc_inst.getOpcode())
1068 .mayAffectControlFlow(mc_inst, *m_reg_info_up);
1071 bool DisassemblerLLVMC::MCDisasmInstance::HasDelaySlot(
1072 llvm::MCInst &mc_inst) const {
1073 return m_instr_info_up->get(mc_inst.getOpcode()).hasDelaySlot();
1076 bool DisassemblerLLVMC::MCDisasmInstance::IsCall(llvm::MCInst &mc_inst) const {
1077 return m_instr_info_up->get(mc_inst.getOpcode()).isCall();
1080 DisassemblerLLVMC::DisassemblerLLVMC(const ArchSpec &arch,
1081 const char *flavor_string)
1082 : Disassembler(arch, flavor_string), m_exe_ctx(nullptr), m_inst(nullptr),
1083 m_data_from_file(false) {
1084 if (!FlavorValidForArchSpec(arch, m_flavor.c_str())) {
1085 m_flavor.assign("default");
1088 unsigned flavor = ~0U;
1089 llvm::Triple triple = arch.GetTriple();
1091 // So far the only supported flavor is "intel" on x86. The base class will
1092 // set this correctly coming in.
1093 if (triple.getArch() == llvm::Triple::x86 ||
1094 triple.getArch() == llvm::Triple::x86_64) {
1095 if (m_flavor == "intel") {
1097 } else if (m_flavor == "att") {
1102 ArchSpec thumb_arch(arch);
1103 if (triple.getArch() == llvm::Triple::arm) {
1104 std::string thumb_arch_name(thumb_arch.GetTriple().getArchName().str());
1105 // Replace "arm" with "thumb" so we get all thumb variants correct
1106 if (thumb_arch_name.size() > 3) {
1107 thumb_arch_name.erase(0, 3);
1108 thumb_arch_name.insert(0, "thumb");
1110 thumb_arch_name = "thumbv8.2a";
1112 thumb_arch.GetTriple().setArchName(llvm::StringRef(thumb_arch_name));
1115 // If no sub architecture specified then use the most recent arm architecture
1116 // so the disassembler will return all instruction. Without it we will see a
1117 // lot of unknow opcode in case the code uses instructions which are not
1118 // available in the oldest arm version (used when no sub architecture is
1120 if (triple.getArch() == llvm::Triple::arm &&
1121 triple.getSubArch() == llvm::Triple::NoSubArch)
1122 triple.setArchName("armv8.2a");
1124 std::string features_str = "";
1125 const char *triple_str = triple.getTriple().c_str();
1127 // ARM Cortex M0-M7 devices only execute thumb instructions
1128 if (arch.IsAlwaysThumbInstructions()) {
1129 triple_str = thumb_arch.GetTriple().getTriple().c_str();
1130 features_str += "+fp-armv8,";
1133 const char *cpu = "";
1135 switch (arch.GetCore()) {
1136 case ArchSpec::eCore_mips32:
1137 case ArchSpec::eCore_mips32el:
1140 case ArchSpec::eCore_mips32r2:
1141 case ArchSpec::eCore_mips32r2el:
1144 case ArchSpec::eCore_mips32r3:
1145 case ArchSpec::eCore_mips32r3el:
1148 case ArchSpec::eCore_mips32r5:
1149 case ArchSpec::eCore_mips32r5el:
1152 case ArchSpec::eCore_mips32r6:
1153 case ArchSpec::eCore_mips32r6el:
1156 case ArchSpec::eCore_mips64:
1157 case ArchSpec::eCore_mips64el:
1160 case ArchSpec::eCore_mips64r2:
1161 case ArchSpec::eCore_mips64r2el:
1164 case ArchSpec::eCore_mips64r3:
1165 case ArchSpec::eCore_mips64r3el:
1168 case ArchSpec::eCore_mips64r5:
1169 case ArchSpec::eCore_mips64r5el:
1172 case ArchSpec::eCore_mips64r6:
1173 case ArchSpec::eCore_mips64r6el:
1181 if (arch.IsMIPS()) {
1182 uint32_t arch_flags = arch.GetFlags();
1183 if (arch_flags & ArchSpec::eMIPSAse_msa)
1184 features_str += "+msa,";
1185 if (arch_flags & ArchSpec::eMIPSAse_dsp)
1186 features_str += "+dsp,";
1187 if (arch_flags & ArchSpec::eMIPSAse_dspr2)
1188 features_str += "+dspr2,";
1191 // If any AArch64 variant, enable the ARMv8.5 ISA with SVE extensions so we
1192 // can disassemble newer instructions.
1193 if (triple.getArch() == llvm::Triple::aarch64)
1194 features_str += "+v8.5a,+sve2";
1196 if (triple.getArch() == llvm::Triple::aarch64
1197 && triple.getVendor() == llvm::Triple::Apple) {
1198 cpu = "apple-latest";
1201 // We use m_disasm_up.get() to tell whether we are valid or not, so if this
1202 // isn't good for some reason, we won't be valid and FindPlugin will fail and
1203 // we won't get used.
1204 m_disasm_up = MCDisasmInstance::Create(triple_str, cpu, features_str.c_str(),
1207 llvm::Triple::ArchType llvm_arch = triple.getArch();
1209 // For arm CPUs that can execute arm or thumb instructions, also create a
1210 // thumb instruction disassembler.
1211 if (llvm_arch == llvm::Triple::arm) {
1212 std::string thumb_triple(thumb_arch.GetTriple().getTriple());
1213 m_alternate_disasm_up =
1214 MCDisasmInstance::Create(thumb_triple.c_str(), "", features_str.c_str(),
1216 if (!m_alternate_disasm_up)
1217 m_disasm_up.reset();
1219 } else if (arch.IsMIPS()) {
1220 /* Create alternate disassembler for MIPS16 and microMIPS */
1221 uint32_t arch_flags = arch.GetFlags();
1222 if (arch_flags & ArchSpec::eMIPSAse_mips16)
1223 features_str += "+mips16,";
1224 else if (arch_flags & ArchSpec::eMIPSAse_micromips)
1225 features_str += "+micromips,";
1227 m_alternate_disasm_up = MCDisasmInstance::Create(
1228 triple_str, cpu, features_str.c_str(), flavor, *this);
1229 if (!m_alternate_disasm_up)
1230 m_disasm_up.reset();
1234 DisassemblerLLVMC::~DisassemblerLLVMC() = default;
1236 Disassembler *DisassemblerLLVMC::CreateInstance(const ArchSpec &arch,
1237 const char *flavor) {
1238 if (arch.GetTriple().getArch() != llvm::Triple::UnknownArch) {
1239 std::unique_ptr<DisassemblerLLVMC> disasm_up(
1240 new DisassemblerLLVMC(arch, flavor));
1242 if (disasm_up.get() && disasm_up->IsValid())
1243 return disasm_up.release();
1248 size_t DisassemblerLLVMC::DecodeInstructions(const Address &base_addr,
1249 const DataExtractor &data,
1250 lldb::offset_t data_offset,
1251 size_t num_instructions,
1252 bool append, bool data_from_file) {
1254 m_instruction_list.Clear();
1259 m_data_from_file = data_from_file;
1260 uint32_t data_cursor = data_offset;
1261 const size_t data_byte_size = data.GetByteSize();
1262 uint32_t instructions_parsed = 0;
1263 Address inst_addr(base_addr);
1265 while (data_cursor < data_byte_size &&
1266 instructions_parsed < num_instructions) {
1268 AddressClass address_class = AddressClass::eCode;
1270 if (m_alternate_disasm_up)
1271 address_class = inst_addr.GetAddressClass();
1273 InstructionSP inst_sp(
1274 new InstructionLLVMC(*this, inst_addr, address_class));
1279 uint32_t inst_size = inst_sp->Decode(*this, data, data_cursor);
1284 m_instruction_list.Append(inst_sp);
1285 data_cursor += inst_size;
1286 inst_addr.Slide(inst_size);
1287 instructions_parsed++;
1290 return data_cursor - data_offset;
1293 void DisassemblerLLVMC::Initialize() {
1294 PluginManager::RegisterPlugin(GetPluginNameStatic(),
1295 "Disassembler that uses LLVM MC to disassemble "
1296 "i386, x86_64, ARM, and ARM64.",
1299 llvm::InitializeAllTargetInfos();
1300 llvm::InitializeAllTargetMCs();
1301 llvm::InitializeAllAsmParsers();
1302 llvm::InitializeAllDisassemblers();
1305 void DisassemblerLLVMC::Terminate() {
1306 PluginManager::UnregisterPlugin(CreateInstance);
1309 ConstString DisassemblerLLVMC::GetPluginNameStatic() {
1310 static ConstString g_name("llvm-mc");
1314 int DisassemblerLLVMC::OpInfoCallback(void *disassembler, uint64_t pc,
1315 uint64_t offset, uint64_t size,
1316 int tag_type, void *tag_bug) {
1317 return static_cast<DisassemblerLLVMC *>(disassembler)
1318 ->OpInfo(pc, offset, size, tag_type, tag_bug);
1321 const char *DisassemblerLLVMC::SymbolLookupCallback(void *disassembler,
1323 uint64_t *type, uint64_t pc,
1324 const char **name) {
1325 return static_cast<DisassemblerLLVMC *>(disassembler)
1326 ->SymbolLookup(value, type, pc, name);
1329 bool DisassemblerLLVMC::FlavorValidForArchSpec(
1330 const lldb_private::ArchSpec &arch, const char *flavor) {
1331 llvm::Triple triple = arch.GetTriple();
1332 if (flavor == nullptr || strcmp(flavor, "default") == 0)
1335 if (triple.getArch() == llvm::Triple::x86 ||
1336 triple.getArch() == llvm::Triple::x86_64) {
1337 return strcmp(flavor, "intel") == 0 || strcmp(flavor, "att") == 0;
1342 bool DisassemblerLLVMC::IsValid() const { return m_disasm_up.operator bool(); }
1344 int DisassemblerLLVMC::OpInfo(uint64_t PC, uint64_t Offset, uint64_t Size,
1345 int tag_type, void *tag_bug) {
1350 memset(tag_bug, 0, sizeof(::LLVMOpInfo1));
1356 const char *DisassemblerLLVMC::SymbolLookup(uint64_t value, uint64_t *type_ptr,
1357 uint64_t pc, const char **name) {
1359 if (m_exe_ctx && m_inst) {
1360 // std::string remove_this_prior_to_checkin;
1361 Target *target = m_exe_ctx ? m_exe_ctx->GetTargetPtr() : nullptr;
1362 Address value_so_addr;
1364 if (m_inst->UsingFileAddress()) {
1365 ModuleSP module_sp(m_inst->GetAddress().GetModule());
1367 module_sp->ResolveFileAddress(value, value_so_addr);
1368 module_sp->ResolveFileAddress(pc, pc_so_addr);
1370 } else if (target && !target->GetSectionLoadList().IsEmpty()) {
1371 target->GetSectionLoadList().ResolveLoadAddress(value, value_so_addr);
1372 target->GetSectionLoadList().ResolveLoadAddress(pc, pc_so_addr);
1375 SymbolContext sym_ctx;
1376 const SymbolContextItem resolve_scope =
1377 eSymbolContextFunction | eSymbolContextSymbol;
1378 if (pc_so_addr.IsValid() && pc_so_addr.GetModule()) {
1379 pc_so_addr.GetModule()->ResolveSymbolContextForAddress(
1380 pc_so_addr, resolve_scope, sym_ctx);
1383 if (value_so_addr.IsValid() && value_so_addr.GetSection()) {
1386 bool format_omitting_current_func_name = false;
1387 if (sym_ctx.symbol || sym_ctx.function) {
1389 if (sym_ctx.GetAddressRange(resolve_scope, 0, false, range) &&
1390 range.GetBaseAddress().IsValid() &&
1391 range.ContainsLoadAddress(value_so_addr, target)) {
1392 format_omitting_current_func_name = true;
1396 // If the "value" address (the target address we're symbolicating) is
1397 // inside the same SymbolContext as the current instruction pc
1398 // (pc_so_addr), don't print the full function name - just print it
1399 // with DumpStyleNoFunctionName style, e.g. "<+36>".
1400 if (format_omitting_current_func_name) {
1401 value_so_addr.Dump(&ss, target, Address::DumpStyleNoFunctionName,
1402 Address::DumpStyleSectionNameOffset);
1406 Address::DumpStyleResolvedDescriptionNoFunctionArguments,
1407 Address::DumpStyleSectionNameOffset);
1410 if (!ss.GetString().empty()) {
1411 // If Address::Dump returned a multi-line description, most commonly
1412 // seen when we have multiple levels of inlined functions at an
1413 // address, only show the first line.
1414 std::string str = ss.GetString();
1415 size_t first_eol_char = str.find_first_of("\r\n");
1416 if (first_eol_char != std::string::npos) {
1417 str.erase(first_eol_char);
1419 m_inst->AppendComment(str);
1425 *type_ptr = LLVMDisassembler_ReferenceType_InOut_None;
1430 // PluginInterface protocol
1431 ConstString DisassemblerLLVMC::GetPluginName() { return GetPluginNameStatic(); }
1433 uint32_t DisassemblerLLVMC::GetPluginVersion() { return 1; }