1//===-- PdbIndex.cpp ------------------------------------------------------===//
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 "PdbIndex.h"
10#include "PdbUtil.h"
11
12#include "llvm/DebugInfo/CodeView/SymbolDeserializer.h"
13#include "llvm/DebugInfo/PDB/Native/DbiStream.h"
14#include "llvm/DebugInfo/PDB/Native/GlobalsStream.h"
15#include "llvm/DebugInfo/PDB/Native/ISectionContribVisitor.h"
16#include "llvm/DebugInfo/PDB/Native/PDBFile.h"
17#include "llvm/DebugInfo/PDB/Native/PublicsStream.h"
18#include "llvm/DebugInfo/PDB/Native/SymbolStream.h"
19#include "llvm/DebugInfo/PDB/Native/TpiStream.h"
20#include "llvm/Object/COFF.h"
21#include "llvm/Support/Error.h"
22
23#include "lldb/Utility/LLDBAssert.h"
24#include "lldb/lldb-defines.h"
25#include <optional>
26
27using namespace lldb_private;
28using namespace lldb_private::npdb;
29using namespace llvm::codeview;
30using namespace llvm::pdb;
31
32PdbIndex::PdbIndex() : m_cus(*this), m_va_to_modi(m_allocator) {}
33
34#define ASSIGN_PTR_OR_RETURN(result_ptr, expr)                                 \
35  {                                                                            \
36    auto expected_result = expr;                                               \
37    if (!expected_result)                                                      \
38      return expected_result.takeError();                                      \
39    result_ptr = &expected_result.get();                                       \
40  }
41
42llvm::Expected<std::unique_ptr<PdbIndex>>
43PdbIndex::create(llvm::pdb::PDBFile *file) {
44  lldbassert(file);
45
46  std::unique_ptr<PdbIndex> result(new PdbIndex());
47  ASSIGN_PTR_OR_RETURN(result->m_dbi, file->getPDBDbiStream());
48  ASSIGN_PTR_OR_RETURN(result->m_tpi, file->getPDBTpiStream());
49  ASSIGN_PTR_OR_RETURN(result->m_ipi, file->getPDBIpiStream());
50  ASSIGN_PTR_OR_RETURN(result->m_info, file->getPDBInfoStream());
51  ASSIGN_PTR_OR_RETURN(result->m_publics, file->getPDBPublicsStream());
52  ASSIGN_PTR_OR_RETURN(result->m_globals, file->getPDBGlobalsStream());
53  ASSIGN_PTR_OR_RETURN(result->m_symrecords, file->getPDBSymbolStream());
54
55  result->m_tpi->buildHashMap();
56
57  result->m_file = file;
58
59  return std::move(result);
60}
61
62lldb::addr_t PdbIndex::MakeVirtualAddress(uint16_t segment,
63                                          uint32_t offset) const {
64  uint32_t max_section = dbi().getSectionHeaders().size();
65  // Segment indices are 1-based.
66  // If this is an absolute symbol, it's indicated by the magic section index
67  // |max_section+1|.  In this case, the offset is meaningless, so just return.
68  if (segment == 0 || segment > max_section)
69    return LLDB_INVALID_ADDRESS;
70
71  const llvm::object::coff_section &cs = dbi().getSectionHeaders()[segment - 1];
72  return m_load_address + static_cast<lldb::addr_t>(cs.VirtualAddress) +
73         static_cast<lldb::addr_t>(offset);
74}
75
76std::optional<uint16_t> PdbIndex::GetModuleIndexForAddr(uint16_t segment,
77                                                        uint32_t offset) const {
78  return GetModuleIndexForVa(MakeVirtualAddress(segment, offset));
79}
80
81std::optional<uint16_t> PdbIndex::GetModuleIndexForVa(lldb::addr_t va) const {
82  auto iter = m_va_to_modi.find(va);
83  if (iter == m_va_to_modi.end())
84    return std::nullopt;
85
86  return iter.value();
87}
88
89void PdbIndex::ParseSectionContribs() {
90  class Visitor : public ISectionContribVisitor {
91    PdbIndex &m_ctx;
92    llvm::IntervalMap<uint64_t, uint16_t> &m_imap;
93
94  public:
95    Visitor(PdbIndex &ctx, llvm::IntervalMap<uint64_t, uint16_t> &imap)
96        : m_ctx(ctx), m_imap(imap) {}
97
98    void visit(const SectionContrib &C) override {
99      if (C.Size == 0)
100        return;
101
102      uint64_t va = m_ctx.MakeVirtualAddress(C.ISect, C.Off);
103      if (va == LLDB_INVALID_ADDRESS)
104        return;
105      uint64_t end = va + C.Size;
106      // IntervalMap's start and end represent a closed range, not a half-open
107      // range, so we have to subtract 1.
108      m_imap.insert(va, end - 1, C.Imod);
109    }
110    void visit(const SectionContrib2 &C) override { visit(C.Base); }
111  };
112  Visitor v(*this, m_va_to_modi);
113  dbi().visitSectionContributions(v);
114}
115
116void PdbIndex::BuildAddrToSymbolMap(CompilandIndexItem &cci) {
117  lldbassert(cci.m_symbols_by_va.empty() &&
118             "Addr to symbol map is already built!");
119  uint16_t modi = cci.m_id.modi;
120  const CVSymbolArray &syms = cci.m_debug_stream.getSymbolArray();
121  for (auto iter = syms.begin(); iter != syms.end(); ++iter) {
122    if (!SymbolHasAddress(*iter))
123      continue;
124
125    SegmentOffset so = GetSegmentAndOffset(*iter);
126    lldb::addr_t va = MakeVirtualAddress(so.segment, so.offset);
127    if (va == LLDB_INVALID_ADDRESS)
128      continue;
129
130    PdbCompilandSymId cu_sym_id(modi, iter.offset());
131
132    // It's rare, but we could have multiple symbols with the same address
133    // because of identical comdat folding.  Right now, the first one will win.
134    cci.m_symbols_by_va.insert(std::make_pair(va, PdbSymUid(cu_sym_id)));
135  }
136}
137
138std::vector<SymbolAndUid> PdbIndex::FindSymbolsByVa(lldb::addr_t va) {
139  std::vector<SymbolAndUid> result;
140
141  std::optional<uint16_t> modi = GetModuleIndexForVa(va);
142  if (!modi)
143    return result;
144
145  CompilandIndexItem &cci = compilands().GetOrCreateCompiland(*modi);
146  if (cci.m_symbols_by_va.empty())
147    BuildAddrToSymbolMap(cci);
148
149  // The map is sorted by starting address of the symbol.  So for example
150  // we could (in theory) have this situation
151  //
152  // [------------------]
153  //    [----------]
154  //      [-----------]
155  //          [-------------]
156  //            [----]
157  //               [-----]
158  //             ^ Address we're searching for
159  // In order to find this, we use the upper_bound of the key value which would
160  // be the first symbol whose starting address is higher than the element we're
161  // searching for.
162
163  auto ub = cci.m_symbols_by_va.upper_bound(va);
164
165  for (auto iter = cci.m_symbols_by_va.begin(); iter != ub; ++iter) {
166    PdbCompilandSymId cu_sym_id = iter->second.asCompilandSym();
167    CVSymbol sym = ReadSymbolRecord(cu_sym_id);
168
169    SegmentOffsetLength sol;
170    if (SymbolIsCode(sym))
171      sol = GetSegmentOffsetAndLength(sym);
172    else
173      sol.so = GetSegmentAndOffset(sym);
174
175    lldb::addr_t start = MakeVirtualAddress(sol.so.segment, sol.so.offset);
176    if (start == LLDB_INVALID_ADDRESS)
177      continue;
178
179    lldb::addr_t end = start + sol.length;
180    if (va >= start && va < end)
181      result.push_back({std::move(sym), iter->second});
182  }
183
184  return result;
185}
186
187CVSymbol PdbIndex::ReadSymbolRecord(PdbCompilandSymId cu_sym) const {
188  const CompilandIndexItem *cci = compilands().GetCompiland(cu_sym.modi);
189  auto iter = cci->m_debug_stream.getSymbolArray().at(cu_sym.offset);
190  lldbassert(iter != cci->m_debug_stream.getSymbolArray().end());
191  return *iter;
192}
193
194CVSymbol PdbIndex::ReadSymbolRecord(PdbGlobalSymId global) const {
195  return symrecords().readRecord(global.offset);
196}
197