1 //===-- lib/MC/MCObjectSymbolizer.cpp -------------------------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9
10 #include "llvm/MC/MCObjectSymbolizer.h"
11 #include "llvm/ADT/SmallString.h"
12 #include "llvm/MC/MCContext.h"
13 #include "llvm/MC/MCExpr.h"
14 #include "llvm/MC/MCInst.h"
15 #include "llvm/MC/MCRelocationInfo.h"
16 #include "llvm/MC/MCSymbol.h"
17 #include "llvm/Object/ELFObjectFile.h"
18 #include "llvm/Object/MachO.h"
19 #include "llvm/Support/raw_ostream.h"
20 #include <algorithm>
21
22 using namespace llvm;
23 using namespace object;
24
25 //===- MCMachObjectSymbolizer ---------------------------------------------===//
26
27 namespace {
28 class MCMachObjectSymbolizer : public MCObjectSymbolizer {
29 const MachOObjectFile *MOOF;
30 // __TEXT;__stubs support.
31 uint64_t StubsStart;
32 uint64_t StubsCount;
33 uint64_t StubSize;
34 uint64_t StubsIndSymIndex;
35
36 public:
37 MCMachObjectSymbolizer(MCContext &Ctx,
38 std::unique_ptr<MCRelocationInfo> RelInfo,
39 const MachOObjectFile *MOOF);
40
41 StringRef findExternalFunctionAt(uint64_t Addr) override;
42
43 void tryAddingPcLoadReferenceComment(raw_ostream &cStream, int64_t Value,
44 uint64_t Address) override;
45 };
46 } // End unnamed namespace
47
MCMachObjectSymbolizer(MCContext & Ctx,std::unique_ptr<MCRelocationInfo> RelInfo,const MachOObjectFile * MOOF)48 MCMachObjectSymbolizer::MCMachObjectSymbolizer(
49 MCContext &Ctx, std::unique_ptr<MCRelocationInfo> RelInfo,
50 const MachOObjectFile *MOOF)
51 : MCObjectSymbolizer(Ctx, std::move(RelInfo), MOOF), MOOF(MOOF),
52 StubsStart(0), StubsCount(0), StubSize(0), StubsIndSymIndex(0) {
53
54 for (const SectionRef &Section : MOOF->sections()) {
55 StringRef Name;
56 Section.getName(Name);
57 if (Name == "__stubs") {
58 SectionRef StubsSec = Section;
59 if (MOOF->is64Bit()) {
60 MachO::section_64 S = MOOF->getSection64(StubsSec.getRawDataRefImpl());
61 StubsIndSymIndex = S.reserved1;
62 StubSize = S.reserved2;
63 } else {
64 MachO::section S = MOOF->getSection(StubsSec.getRawDataRefImpl());
65 StubsIndSymIndex = S.reserved1;
66 StubSize = S.reserved2;
67 }
68 assert(StubSize && "Mach-O stub entry size can't be zero!");
69 StubsSec.getAddress(StubsStart);
70 StubsSec.getSize(StubsCount);
71 StubsCount /= StubSize;
72 }
73 }
74 }
75
findExternalFunctionAt(uint64_t Addr)76 StringRef MCMachObjectSymbolizer::findExternalFunctionAt(uint64_t Addr) {
77 // FIXME: also, this can all be done at the very beginning, by iterating over
78 // all stubs and creating the calls to outside functions. Is it worth it
79 // though?
80 if (!StubSize)
81 return StringRef();
82 uint64_t StubIdx = (Addr - StubsStart) / StubSize;
83 if (StubIdx >= StubsCount)
84 return StringRef();
85
86 uint32_t SymtabIdx =
87 MOOF->getIndirectSymbolTableEntry(MOOF->getDysymtabLoadCommand(), StubIdx);
88
89 StringRef SymName;
90 symbol_iterator SI = MOOF->symbol_begin();
91 for (uint32_t i = 0; i != SymtabIdx; ++i)
92 ++SI;
93 SI->getName(SymName);
94 assert(SI != MOOF->symbol_end() && "Stub wasn't found in the symbol table!");
95 assert(SymName.front() == '_' && "Mach-O symbol doesn't start with '_'!");
96 return SymName.substr(1);
97 }
98
99 void MCMachObjectSymbolizer::
tryAddingPcLoadReferenceComment(raw_ostream & cStream,int64_t Value,uint64_t Address)100 tryAddingPcLoadReferenceComment(raw_ostream &cStream, int64_t Value,
101 uint64_t Address) {
102 if (const RelocationRef *R = findRelocationAt(Address)) {
103 const MCExpr *RelExpr = RelInfo->createExprForRelocation(*R);
104 if (!RelExpr || RelExpr->EvaluateAsAbsolute(Value) == false)
105 return;
106 }
107 uint64_t Addr = Value;
108 if (const SectionRef *S = findSectionContaining(Addr)) {
109 StringRef Name; S->getName(Name);
110 uint64_t SAddr; S->getAddress(SAddr);
111 if (Name == "__cstring") {
112 StringRef Contents;
113 S->getContents(Contents);
114 Contents = Contents.substr(Addr - SAddr);
115 cStream << " ## literal pool for: "
116 << Contents.substr(0, Contents.find_first_of(0));
117 }
118 }
119 }
120
121 //===- MCObjectSymbolizer -------------------------------------------------===//
122
MCObjectSymbolizer(MCContext & Ctx,std::unique_ptr<MCRelocationInfo> RelInfo,const ObjectFile * Obj)123 MCObjectSymbolizer::MCObjectSymbolizer(
124 MCContext &Ctx, std::unique_ptr<MCRelocationInfo> RelInfo,
125 const ObjectFile *Obj)
126 : MCSymbolizer(Ctx, std::move(RelInfo)), Obj(Obj), SortedSections(),
127 AddrToReloc() {}
128
129 bool MCObjectSymbolizer::
tryAddingSymbolicOperand(MCInst & MI,raw_ostream & cStream,int64_t Value,uint64_t Address,bool IsBranch,uint64_t Offset,uint64_t InstSize)130 tryAddingSymbolicOperand(MCInst &MI, raw_ostream &cStream,
131 int64_t Value, uint64_t Address, bool IsBranch,
132 uint64_t Offset, uint64_t InstSize) {
133 if (IsBranch) {
134 StringRef ExtFnName = findExternalFunctionAt((uint64_t)Value);
135 if (!ExtFnName.empty()) {
136 MCSymbol *Sym = Ctx.GetOrCreateSymbol(ExtFnName);
137 const MCExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
138 MI.addOperand(MCOperand::CreateExpr(Expr));
139 return true;
140 }
141 }
142
143 if (const RelocationRef *R = findRelocationAt(Address + Offset)) {
144 if (const MCExpr *RelExpr = RelInfo->createExprForRelocation(*R)) {
145 MI.addOperand(MCOperand::CreateExpr(RelExpr));
146 return true;
147 }
148 // Only try to create a symbol+offset expression if there is no relocation.
149 return false;
150 }
151
152 // Interpret Value as a branch target.
153 if (IsBranch == false)
154 return false;
155 uint64_t UValue = Value;
156 // FIXME: map instead of looping each time?
157 for (const SymbolRef &Symbol : Obj->symbols()) {
158 uint64_t SymAddr;
159 Symbol.getAddress(SymAddr);
160 uint64_t SymSize;
161 Symbol.getSize(SymSize);
162 StringRef SymName;
163 Symbol.getName(SymName);
164 SymbolRef::Type SymType;
165 Symbol.getType(SymType);
166 if (SymAddr == UnknownAddressOrSize || SymSize == UnknownAddressOrSize ||
167 SymName.empty() || SymType != SymbolRef::ST_Function)
168 continue;
169
170 if ( SymAddr == UValue ||
171 (SymAddr <= UValue && SymAddr + SymSize > UValue)) {
172 MCSymbol *Sym = Ctx.GetOrCreateSymbol(SymName);
173 const MCExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
174 if (SymAddr != UValue) {
175 const MCExpr *Off = MCConstantExpr::Create(UValue - SymAddr, Ctx);
176 Expr = MCBinaryExpr::CreateAdd(Expr, Off, Ctx);
177 }
178 MI.addOperand(MCOperand::CreateExpr(Expr));
179 return true;
180 }
181 }
182 return false;
183 }
184
185 void MCObjectSymbolizer::
tryAddingPcLoadReferenceComment(raw_ostream & cStream,int64_t Value,uint64_t Address)186 tryAddingPcLoadReferenceComment(raw_ostream &cStream,
187 int64_t Value, uint64_t Address) {
188 }
189
findExternalFunctionAt(uint64_t Addr)190 StringRef MCObjectSymbolizer::findExternalFunctionAt(uint64_t Addr) {
191 return StringRef();
192 }
193
createObjectSymbolizer(MCContext & Ctx,std::unique_ptr<MCRelocationInfo> RelInfo,const ObjectFile * Obj)194 MCObjectSymbolizer *MCObjectSymbolizer::createObjectSymbolizer(
195 MCContext &Ctx, std::unique_ptr<MCRelocationInfo> RelInfo,
196 const ObjectFile *Obj) {
197 if (const MachOObjectFile *MOOF = dyn_cast<MachOObjectFile>(Obj))
198 return new MCMachObjectSymbolizer(Ctx, std::move(RelInfo), MOOF);
199 return new MCObjectSymbolizer(Ctx, std::move(RelInfo), Obj);
200 }
201
202 // SortedSections implementation.
203
SectionStartsBefore(const SectionRef & S,uint64_t Addr)204 static bool SectionStartsBefore(const SectionRef &S, uint64_t Addr) {
205 uint64_t SAddr; S.getAddress(SAddr);
206 return SAddr < Addr;
207 }
208
findSectionContaining(uint64_t Addr)209 const SectionRef *MCObjectSymbolizer::findSectionContaining(uint64_t Addr) {
210 if (SortedSections.empty())
211 buildSectionList();
212
213 SortedSectionList::iterator
214 EndIt = SortedSections.end(),
215 It = std::lower_bound(SortedSections.begin(), EndIt,
216 Addr, SectionStartsBefore);
217 if (It == EndIt)
218 return nullptr;
219 uint64_t SAddr; It->getAddress(SAddr);
220 uint64_t SSize; It->getSize(SSize);
221 if (Addr >= SAddr + SSize)
222 return nullptr;
223 return &*It;
224 }
225
findRelocationAt(uint64_t Addr)226 const RelocationRef *MCObjectSymbolizer::findRelocationAt(uint64_t Addr) {
227 if (AddrToReloc.empty())
228 buildRelocationByAddrMap();
229
230 AddrToRelocMap::const_iterator RI = AddrToReloc.find(Addr);
231 if (RI == AddrToReloc.end())
232 return nullptr;
233 return &RI->second;
234 }
235
buildSectionList()236 void MCObjectSymbolizer::buildSectionList() {
237 for (const SectionRef &Section : Obj->sections()) {
238 bool RequiredForExec;
239 Section.isRequiredForExecution(RequiredForExec);
240 if (RequiredForExec == false)
241 continue;
242 uint64_t SAddr;
243 Section.getAddress(SAddr);
244 uint64_t SSize;
245 Section.getSize(SSize);
246 SortedSectionList::iterator It =
247 std::lower_bound(SortedSections.begin(), SortedSections.end(), SAddr,
248 SectionStartsBefore);
249 if (It != SortedSections.end()) {
250 uint64_t FoundSAddr; It->getAddress(FoundSAddr);
251 if (FoundSAddr < SAddr + SSize)
252 llvm_unreachable("Inserting overlapping sections");
253 }
254 SortedSections.insert(It, Section);
255 }
256 }
257
buildRelocationByAddrMap()258 void MCObjectSymbolizer::buildRelocationByAddrMap() {
259 for (const SectionRef &Section : Obj->sections()) {
260 for (const RelocationRef &Reloc : Section.relocations()) {
261 uint64_t Address;
262 Reloc.getAddress(Address);
263 // At a specific address, only keep the first relocation.
264 if (AddrToReloc.find(Address) == AddrToReloc.end())
265 AddrToReloc[Address] = Reloc;
266 }
267 }
268 }
269