MCObjectSymbolizer.cpp revision 263508 
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/MachO.h"
18#include "llvm/Object/ELFObjectFile.h"
19#include "llvm/Support/raw_ostream.h"
20#include <algorithm>
21
22using namespace llvm;
23using namespace object;
24
25//===- MCMachObjectSymbolizer ---------------------------------------------===//
26
27namespace {
28class 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
36public:
37  MCMachObjectSymbolizer(MCContext &Ctx, OwningPtr<MCRelocationInfo> &RelInfo,
38                         const MachOObjectFile *MOOF);
39
40  StringRef findExternalFunctionAt(uint64_t Addr) LLVM_OVERRIDE;
41
42  void tryAddingPcLoadReferenceComment(raw_ostream &cStream,
43                                       int64_t Value,
44                                       uint64_t Address) LLVM_OVERRIDE;
45};
46} // End unnamed namespace
47
48
49MCMachObjectSymbolizer::
50MCMachObjectSymbolizer(MCContext &Ctx, OwningPtr<MCRelocationInfo> &RelInfo,
51                       const MachOObjectFile *MOOF)
52    : MCObjectSymbolizer(Ctx, RelInfo, MOOF), MOOF(MOOF),
53      StubsStart(0), StubsCount(0), StubSize(0), StubsIndSymIndex(0) {
54
55  error_code ec;
56  for (section_iterator SI = MOOF->begin_sections(), SE = MOOF->end_sections();
57       SI != SE; SI.increment(ec)) {
58    if (ec) break;
59    StringRef Name; SI->getName(Name);
60    if (Name == "__stubs") {
61      SectionRef StubsSec = *SI;
62      if (MOOF->is64Bit()) {
63        MachO::section_64 S = MOOF->getSection64(StubsSec.getRawDataRefImpl());
64        StubsIndSymIndex = S.reserved1;
65        StubSize = S.reserved2;
66      } else {
67        MachO::section S = MOOF->getSection(StubsSec.getRawDataRefImpl());
68        StubsIndSymIndex = S.reserved1;
69        StubSize = S.reserved2;
70      }
71      assert(StubSize && "Mach-O stub entry size can't be zero!");
72      StubsSec.getAddress(StubsStart);
73      StubsSec.getSize(StubsCount);
74      StubsCount /= StubSize;
75    }
76  }
77}
78
79StringRef MCMachObjectSymbolizer::findExternalFunctionAt(uint64_t Addr) {
80  // FIXME: also, this can all be done at the very beginning, by iterating over
81  // all stubs and creating the calls to outside functions. Is it worth it
82  // though?
83  if (!StubSize)
84    return StringRef();
85  uint64_t StubIdx = (Addr - StubsStart) / StubSize;
86  if (StubIdx >= StubsCount)
87    return StringRef();
88
89  uint32_t SymtabIdx =
90    MOOF->getIndirectSymbolTableEntry(MOOF->getDysymtabLoadCommand(), StubIdx);
91
92  StringRef SymName;
93  symbol_iterator SI = MOOF->begin_symbols();
94  error_code ec;
95  for (uint32_t i = 0; i != SymtabIdx; ++i) {
96    SI.increment(ec);
97  }
98  SI->getName(SymName);
99  assert(SI != MOOF->end_symbols() && "Stub wasn't found in the symbol table!");
100  assert(SymName.front() == '_' && "Mach-O symbol doesn't start with '_'!");
101  return SymName.substr(1);
102}
103
104void MCMachObjectSymbolizer::
105tryAddingPcLoadReferenceComment(raw_ostream &cStream, int64_t Value,
106                                uint64_t Address) {
107  if (const RelocationRef *R = findRelocationAt(Address)) {
108    const MCExpr *RelExpr = RelInfo->createExprForRelocation(*R);
109    if (!RelExpr || RelExpr->EvaluateAsAbsolute(Value) == false)
110      return;
111  }
112  uint64_t Addr = Value;
113  if (const SectionRef *S = findSectionContaining(Addr)) {
114    StringRef Name; S->getName(Name);
115    uint64_t SAddr; S->getAddress(SAddr);
116    if (Name == "__cstring") {
117      StringRef Contents;
118      S->getContents(Contents);
119      Contents = Contents.substr(Addr - SAddr);
120      cStream << " ## literal pool for: "
121              << Contents.substr(0, Contents.find_first_of(0));
122    }
123  }
124}
125
126//===- MCObjectSymbolizer -------------------------------------------------===//
127
128MCObjectSymbolizer::MCObjectSymbolizer(MCContext &Ctx,
129                                       OwningPtr<MCRelocationInfo> &RelInfo,
130                                       const ObjectFile *Obj)
131    : MCSymbolizer(Ctx, RelInfo), Obj(Obj), SortedSections(), AddrToReloc() {
132}
133
134bool MCObjectSymbolizer::
135tryAddingSymbolicOperand(MCInst &MI, raw_ostream &cStream,
136                         int64_t Value, uint64_t Address, bool IsBranch,
137                         uint64_t Offset, uint64_t InstSize) {
138  if (IsBranch) {
139    StringRef ExtFnName = findExternalFunctionAt((uint64_t)Value);
140    if (!ExtFnName.empty()) {
141      MCSymbol *Sym = Ctx.GetOrCreateSymbol(ExtFnName);
142      const MCExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
143      MI.addOperand(MCOperand::CreateExpr(Expr));
144      return true;
145    }
146  }
147
148  if (const RelocationRef *R = findRelocationAt(Address + Offset)) {
149    if (const MCExpr *RelExpr = RelInfo->createExprForRelocation(*R)) {
150      MI.addOperand(MCOperand::CreateExpr(RelExpr));
151      return true;
152    }
153    // Only try to create a symbol+offset expression if there is no relocation.
154    return false;
155  }
156
157  // Interpret Value as a branch target.
158  if (IsBranch == false)
159    return false;
160  uint64_t UValue = Value;
161  // FIXME: map instead of looping each time?
162  error_code ec;
163  for (symbol_iterator SI = Obj->begin_symbols(), SE = Obj->end_symbols();
164       SI != SE; SI.increment(ec)) {
165    if (ec) break;
166    uint64_t SymAddr; SI->getAddress(SymAddr);
167    uint64_t SymSize; SI->getSize(SymSize);
168    StringRef SymName; SI->getName(SymName);
169    SymbolRef::Type SymType; SI->getType(SymType);
170    if (SymAddr == UnknownAddressOrSize || SymSize == UnknownAddressOrSize
171        || SymName.empty() || SymType != SymbolRef::ST_Function)
172      continue;
173
174    if ( SymAddr == UValue ||
175        (SymAddr <= UValue && SymAddr + SymSize > UValue)) {
176      MCSymbol *Sym = Ctx.GetOrCreateSymbol(SymName);
177      const MCExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
178      if (SymAddr != UValue) {
179        const MCExpr *Off = MCConstantExpr::Create(UValue - SymAddr, Ctx);
180        Expr = MCBinaryExpr::CreateAdd(Expr, Off, Ctx);
181      }
182      MI.addOperand(MCOperand::CreateExpr(Expr));
183      return true;
184    }
185  }
186  return false;
187}
188
189void MCObjectSymbolizer::
190tryAddingPcLoadReferenceComment(raw_ostream &cStream,
191                                int64_t Value, uint64_t Address) {
192}
193
194StringRef MCObjectSymbolizer::findExternalFunctionAt(uint64_t Addr) {
195  return StringRef();
196}
197
198MCObjectSymbolizer *
199MCObjectSymbolizer::createObjectSymbolizer(MCContext &Ctx,
200                                           OwningPtr<MCRelocationInfo> &RelInfo,
201                                           const ObjectFile *Obj) {
202  if (const MachOObjectFile *MOOF = dyn_cast<MachOObjectFile>(Obj))
203    return new MCMachObjectSymbolizer(Ctx, RelInfo, MOOF);
204  return new MCObjectSymbolizer(Ctx, RelInfo, Obj);
205}
206
207// SortedSections implementation.
208
209static bool SectionStartsBefore(const SectionRef &S, uint64_t Addr) {
210  uint64_t SAddr; S.getAddress(SAddr);
211  return SAddr < Addr;
212}
213
214const SectionRef *MCObjectSymbolizer::findSectionContaining(uint64_t Addr) {
215  if (SortedSections.empty())
216    buildSectionList();
217
218  SortedSectionList::iterator
219    EndIt = SortedSections.end(),
220    It = std::lower_bound(SortedSections.begin(), EndIt,
221                          Addr, SectionStartsBefore);
222  if (It == EndIt)
223    return 0;
224  uint64_t SAddr; It->getAddress(SAddr);
225  uint64_t SSize; It->getSize(SSize);
226  if (Addr >= SAddr + SSize)
227    return 0;
228  return &*It;
229}
230
231const RelocationRef *MCObjectSymbolizer::findRelocationAt(uint64_t Addr) {
232  if (AddrToReloc.empty())
233    buildRelocationByAddrMap();
234
235  AddrToRelocMap::const_iterator RI = AddrToReloc.find(Addr);
236  if (RI == AddrToReloc.end())
237    return 0;
238  return &RI->second;
239}
240
241void MCObjectSymbolizer::buildSectionList() {
242  error_code ec;
243  for (section_iterator SI = Obj->begin_sections(), SE = Obj->end_sections();
244                        SI != SE; SI.increment(ec)) {
245    if (ec) break;
246
247    bool RequiredForExec; SI->isRequiredForExecution(RequiredForExec);
248    if (RequiredForExec == false)
249      continue;
250    uint64_t SAddr; SI->getAddress(SAddr);
251    uint64_t SSize; SI->getSize(SSize);
252    SortedSectionList::iterator It = std::lower_bound(SortedSections.begin(),
253                                                      SortedSections.end(),
254                                                      SAddr,
255                                                      SectionStartsBefore);
256    if (It != SortedSections.end()) {
257      uint64_t FoundSAddr; It->getAddress(FoundSAddr);
258      if (FoundSAddr < SAddr + SSize)
259        llvm_unreachable("Inserting overlapping sections");
260    }
261    SortedSections.insert(It, *SI);
262  }
263}
264
265void MCObjectSymbolizer::buildRelocationByAddrMap() {
266  error_code ec;
267  for (section_iterator SI = Obj->begin_sections(), SE = Obj->end_sections();
268                        SI != SE; SI.increment(ec)) {
269    if (ec) break;
270
271    section_iterator RelSecI = SI->getRelocatedSection();
272    if (RelSecI == Obj->end_sections())
273      continue;
274
275    uint64_t StartAddr; RelSecI->getAddress(StartAddr);
276    uint64_t Size; RelSecI->getSize(Size);
277    bool RequiredForExec; RelSecI->isRequiredForExecution(RequiredForExec);
278    if (RequiredForExec == false || Size == 0)
279      continue;
280    for (relocation_iterator RI = SI->begin_relocations(),
281                             RE = SI->end_relocations();
282                             RI != RE;
283                             RI.increment(ec)) {
284      if (ec) break;
285      // FIXME: libObject is inconsistent regarding error handling. The
286      // overwhelming majority of methods always return object_error::success,
287      // and assert for simple errors.. Here, ELFObjectFile::getRelocationOffset
288      // asserts when the file type isn't ET_REL.
289      // This workaround handles x86-64 elf, the only one that has a relocinfo.
290      uint64_t Offset;
291      if (Obj->isELF()) {
292        const ELF64LEObjectFile *ELFObj = dyn_cast<ELF64LEObjectFile>(Obj);
293        if (ELFObj == 0)
294          break;
295        if (ELFObj->getELFFile()->getHeader()->e_type == ELF::ET_REL) {
296          RI->getOffset(Offset);
297          Offset += StartAddr;
298        } else {
299          RI->getAddress(Offset);
300        }
301      } else {
302        RI->getOffset(Offset);
303        Offset += StartAddr;
304      }
305      // At a specific address, only keep the first relocation.
306      if (AddrToReloc.find(Offset) == AddrToReloc.end())
307        AddrToReloc[Offset] = *RI;
308    }
309  }
310}
311