1 //===-- RuntimeDyldImpl.h - Run-time dynamic linker for MC-JIT --*- C++ -*-===//
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 // Interface for the implementations of runtime dynamic linker facilities.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #ifndef LLVM_LIB_EXECUTIONENGINE_RUNTIMEDYLD_RUNTIMEDYLDIMPL_H
15 #define LLVM_LIB_EXECUTIONENGINE_RUNTIMEDYLD_RUNTIMEDYLDIMPL_H
16
17 #include "llvm/ADT/DenseMap.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/ADT/StringMap.h"
20 #include "llvm/ADT/Triple.h"
21 #include "llvm/ExecutionEngine/RTDyldMemoryManager.h"
22 #include "llvm/ExecutionEngine/RuntimeDyld.h"
23 #include "llvm/ExecutionEngine/RuntimeDyldChecker.h"
24 #include "llvm/Object/ObjectFile.h"
25 #include "llvm/Support/Debug.h"
26 #include "llvm/Support/ErrorHandling.h"
27 #include "llvm/Support/Format.h"
28 #include "llvm/Support/Host.h"
29 #include "llvm/Support/Mutex.h"
30 #include "llvm/Support/SwapByteOrder.h"
31 #include "llvm/Support/raw_ostream.h"
32 #include <map>
33 #include <unordered_map>
34 #include <system_error>
35
36 using namespace llvm;
37 using namespace llvm::object;
38
39 namespace llvm {
40
41 // Helper for extensive error checking in debug builds.
Check(std::error_code Err)42 inline std::error_code Check(std::error_code Err) {
43 if (Err) {
44 report_fatal_error(Err.message());
45 }
46 return Err;
47 }
48
49 class Twine;
50
51 /// SectionEntry - represents a section emitted into memory by the dynamic
52 /// linker.
53 class SectionEntry {
54 /// Name - section name.
55 std::string Name;
56
57 /// Address - address in the linker's memory where the section resides.
58 uint8_t *Address;
59
60 /// Size - section size. Doesn't include the stubs.
61 size_t Size;
62
63 /// LoadAddress - the address of the section in the target process's memory.
64 /// Used for situations in which JIT-ed code is being executed in the address
65 /// space of a separate process. If the code executes in the same address
66 /// space where it was JIT-ed, this just equals Address.
67 uint64_t LoadAddress;
68
69 /// StubOffset - used for architectures with stub functions for far
70 /// relocations (like ARM).
71 uintptr_t StubOffset;
72
73 /// The total amount of space allocated for this section. This includes the
74 /// section size and the maximum amount of space that the stubs can occupy.
75 size_t AllocationSize;
76
77 /// ObjAddress - address of the section in the in-memory object file. Used
78 /// for calculating relocations in some object formats (like MachO).
79 uintptr_t ObjAddress;
80
81 public:
SectionEntry(StringRef name,uint8_t * address,size_t size,size_t allocationSize,uintptr_t objAddress)82 SectionEntry(StringRef name, uint8_t *address, size_t size,
83 size_t allocationSize, uintptr_t objAddress)
84 : Name(name), Address(address), Size(size),
85 LoadAddress(reinterpret_cast<uintptr_t>(address)), StubOffset(size),
86 AllocationSize(allocationSize), ObjAddress(objAddress) {
87 // AllocationSize is used only in asserts, prevent an "unused private field"
88 // warning:
89 (void)AllocationSize;
90 }
91
getName()92 StringRef getName() const { return Name; }
93
getAddress()94 uint8_t *getAddress() const { return Address; }
95
96 /// \brief Return the address of this section with an offset.
getAddressWithOffset(unsigned OffsetBytes)97 uint8_t *getAddressWithOffset(unsigned OffsetBytes) const {
98 assert(OffsetBytes <= AllocationSize && "Offset out of bounds!");
99 return Address + OffsetBytes;
100 }
101
getSize()102 size_t getSize() const { return Size; }
103
getLoadAddress()104 uint64_t getLoadAddress() const { return LoadAddress; }
setLoadAddress(uint64_t LA)105 void setLoadAddress(uint64_t LA) { LoadAddress = LA; }
106
107 /// \brief Return the load address of this section with an offset.
getLoadAddressWithOffset(unsigned OffsetBytes)108 uint64_t getLoadAddressWithOffset(unsigned OffsetBytes) const {
109 assert(OffsetBytes <= AllocationSize && "Offset out of bounds!");
110 return LoadAddress + OffsetBytes;
111 }
112
getStubOffset()113 uintptr_t getStubOffset() const { return StubOffset; }
114
advanceStubOffset(unsigned StubSize)115 void advanceStubOffset(unsigned StubSize) {
116 StubOffset += StubSize;
117 assert(StubOffset <= AllocationSize && "Not enough space allocated!");
118 }
119
getObjAddress()120 uintptr_t getObjAddress() const { return ObjAddress; }
121 };
122
123 /// RelocationEntry - used to represent relocations internally in the dynamic
124 /// linker.
125 class RelocationEntry {
126 public:
127 /// SectionID - the section this relocation points to.
128 unsigned SectionID;
129
130 /// Offset - offset into the section.
131 uint64_t Offset;
132
133 /// RelType - relocation type.
134 uint32_t RelType;
135
136 /// Addend - the relocation addend encoded in the instruction itself. Also
137 /// used to make a relocation section relative instead of symbol relative.
138 int64_t Addend;
139
140 struct SectionPair {
141 uint32_t SectionA;
142 uint32_t SectionB;
143 };
144
145 /// SymOffset - Section offset of the relocation entry's symbol (used for GOT
146 /// lookup).
147 union {
148 uint64_t SymOffset;
149 SectionPair Sections;
150 };
151
152 /// True if this is a PCRel relocation (MachO specific).
153 bool IsPCRel;
154
155 /// The size of this relocation (MachO specific).
156 unsigned Size;
157
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend)158 RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend)
159 : SectionID(id), Offset(offset), RelType(type), Addend(addend),
160 SymOffset(0), IsPCRel(false), Size(0) {}
161
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend,uint64_t symoffset)162 RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
163 uint64_t symoffset)
164 : SectionID(id), Offset(offset), RelType(type), Addend(addend),
165 SymOffset(symoffset), IsPCRel(false), Size(0) {}
166
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend,bool IsPCRel,unsigned Size)167 RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
168 bool IsPCRel, unsigned Size)
169 : SectionID(id), Offset(offset), RelType(type), Addend(addend),
170 SymOffset(0), IsPCRel(IsPCRel), Size(Size) {}
171
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend,unsigned SectionA,uint64_t SectionAOffset,unsigned SectionB,uint64_t SectionBOffset,bool IsPCRel,unsigned Size)172 RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
173 unsigned SectionA, uint64_t SectionAOffset, unsigned SectionB,
174 uint64_t SectionBOffset, bool IsPCRel, unsigned Size)
175 : SectionID(id), Offset(offset), RelType(type),
176 Addend(SectionAOffset - SectionBOffset + addend), IsPCRel(IsPCRel),
177 Size(Size) {
178 Sections.SectionA = SectionA;
179 Sections.SectionB = SectionB;
180 }
181 };
182
183 class RelocationValueRef {
184 public:
185 unsigned SectionID;
186 uint64_t Offset;
187 int64_t Addend;
188 const char *SymbolName;
RelocationValueRef()189 RelocationValueRef() : SectionID(0), Offset(0), Addend(0),
190 SymbolName(nullptr) {}
191
192 inline bool operator==(const RelocationValueRef &Other) const {
193 return SectionID == Other.SectionID && Offset == Other.Offset &&
194 Addend == Other.Addend && SymbolName == Other.SymbolName;
195 }
196 inline bool operator<(const RelocationValueRef &Other) const {
197 if (SectionID != Other.SectionID)
198 return SectionID < Other.SectionID;
199 if (Offset != Other.Offset)
200 return Offset < Other.Offset;
201 if (Addend != Other.Addend)
202 return Addend < Other.Addend;
203 return SymbolName < Other.SymbolName;
204 }
205 };
206
207 /// @brief Symbol info for RuntimeDyld.
208 class SymbolTableEntry : public JITSymbolBase {
209 public:
SymbolTableEntry()210 SymbolTableEntry()
211 : JITSymbolBase(JITSymbolFlags::None), Offset(0), SectionID(0) {}
212
SymbolTableEntry(unsigned SectionID,uint64_t Offset,JITSymbolFlags Flags)213 SymbolTableEntry(unsigned SectionID, uint64_t Offset, JITSymbolFlags Flags)
214 : JITSymbolBase(Flags), Offset(Offset), SectionID(SectionID) {}
215
getSectionID()216 unsigned getSectionID() const { return SectionID; }
getOffset()217 uint64_t getOffset() const { return Offset; }
218
219 private:
220 uint64_t Offset;
221 unsigned SectionID;
222 };
223
224 typedef StringMap<SymbolTableEntry> RTDyldSymbolTable;
225
226 class RuntimeDyldImpl {
227 friend class RuntimeDyld::LoadedObjectInfo;
228 friend class RuntimeDyldCheckerImpl;
229 protected:
230 static const unsigned AbsoluteSymbolSection = ~0U;
231
232 // The MemoryManager to load objects into.
233 RuntimeDyld::MemoryManager &MemMgr;
234
235 // The symbol resolver to use for external symbols.
236 RuntimeDyld::SymbolResolver &Resolver;
237
238 // Attached RuntimeDyldChecker instance. Null if no instance attached.
239 RuntimeDyldCheckerImpl *Checker;
240
241 // A list of all sections emitted by the dynamic linker. These sections are
242 // referenced in the code by means of their index in this list - SectionID.
243 typedef SmallVector<SectionEntry, 64> SectionList;
244 SectionList Sections;
245
246 typedef unsigned SID; // Type for SectionIDs
247 #define RTDYLD_INVALID_SECTION_ID ((RuntimeDyldImpl::SID)(-1))
248
249 // Keep a map of sections from object file to the SectionID which
250 // references it.
251 typedef std::map<SectionRef, unsigned> ObjSectionToIDMap;
252
253 // A global symbol table for symbols from all loaded modules.
254 RTDyldSymbolTable GlobalSymbolTable;
255
256 // Keep a map of common symbols to their info pairs
257 typedef std::vector<SymbolRef> CommonSymbolList;
258
259 // For each symbol, keep a list of relocations based on it. Anytime
260 // its address is reassigned (the JIT re-compiled the function, e.g.),
261 // the relocations get re-resolved.
262 // The symbol (or section) the relocation is sourced from is the Key
263 // in the relocation list where it's stored.
264 typedef SmallVector<RelocationEntry, 64> RelocationList;
265 // Relocations to sections already loaded. Indexed by SectionID which is the
266 // source of the address. The target where the address will be written is
267 // SectionID/Offset in the relocation itself.
268 std::unordered_map<unsigned, RelocationList> Relocations;
269
270 // Relocations to external symbols that are not yet resolved. Symbols are
271 // external when they aren't found in the global symbol table of all loaded
272 // modules. This map is indexed by symbol name.
273 StringMap<RelocationList> ExternalSymbolRelocations;
274
275
276 typedef std::map<RelocationValueRef, uintptr_t> StubMap;
277
278 Triple::ArchType Arch;
279 bool IsTargetLittleEndian;
280 bool IsMipsO32ABI;
281 bool IsMipsN64ABI;
282
283 // True if all sections should be passed to the memory manager, false if only
284 // sections containing relocations should be. Defaults to 'false'.
285 bool ProcessAllSections;
286
287 // This mutex prevents simultaneously loading objects from two different
288 // threads. This keeps us from having to protect individual data structures
289 // and guarantees that section allocation requests to the memory manager
290 // won't be interleaved between modules. It is also used in mapSectionAddress
291 // and resolveRelocations to protect write access to internal data structures.
292 //
293 // loadObject may be called on the same thread during the handling of of
294 // processRelocations, and that's OK. The handling of the relocation lists
295 // is written in such a way as to work correctly if new elements are added to
296 // the end of the list while the list is being processed.
297 sys::Mutex lock;
298
299 virtual unsigned getMaxStubSize() = 0;
300 virtual unsigned getStubAlignment() = 0;
301
302 bool HasError;
303 std::string ErrorStr;
304
305 // Set the error state and record an error string.
Error(const Twine & Msg)306 bool Error(const Twine &Msg) {
307 ErrorStr = Msg.str();
308 HasError = true;
309 return true;
310 }
311
getSectionLoadAddress(unsigned SectionID)312 uint64_t getSectionLoadAddress(unsigned SectionID) const {
313 return Sections[SectionID].getLoadAddress();
314 }
315
getSectionAddress(unsigned SectionID)316 uint8_t *getSectionAddress(unsigned SectionID) const {
317 return Sections[SectionID].getAddress();
318 }
319
writeInt16BE(uint8_t * Addr,uint16_t Value)320 void writeInt16BE(uint8_t *Addr, uint16_t Value) {
321 if (IsTargetLittleEndian)
322 sys::swapByteOrder(Value);
323 *Addr = (Value >> 8) & 0xFF;
324 *(Addr + 1) = Value & 0xFF;
325 }
326
writeInt32BE(uint8_t * Addr,uint32_t Value)327 void writeInt32BE(uint8_t *Addr, uint32_t Value) {
328 if (IsTargetLittleEndian)
329 sys::swapByteOrder(Value);
330 *Addr = (Value >> 24) & 0xFF;
331 *(Addr + 1) = (Value >> 16) & 0xFF;
332 *(Addr + 2) = (Value >> 8) & 0xFF;
333 *(Addr + 3) = Value & 0xFF;
334 }
335
writeInt64BE(uint8_t * Addr,uint64_t Value)336 void writeInt64BE(uint8_t *Addr, uint64_t Value) {
337 if (IsTargetLittleEndian)
338 sys::swapByteOrder(Value);
339 *Addr = (Value >> 56) & 0xFF;
340 *(Addr + 1) = (Value >> 48) & 0xFF;
341 *(Addr + 2) = (Value >> 40) & 0xFF;
342 *(Addr + 3) = (Value >> 32) & 0xFF;
343 *(Addr + 4) = (Value >> 24) & 0xFF;
344 *(Addr + 5) = (Value >> 16) & 0xFF;
345 *(Addr + 6) = (Value >> 8) & 0xFF;
346 *(Addr + 7) = Value & 0xFF;
347 }
348
setMipsABI(const ObjectFile & Obj)349 virtual void setMipsABI(const ObjectFile &Obj) {
350 IsMipsO32ABI = false;
351 IsMipsN64ABI = false;
352 }
353
354 /// Endian-aware read Read the least significant Size bytes from Src.
355 uint64_t readBytesUnaligned(uint8_t *Src, unsigned Size) const;
356
357 /// Endian-aware write. Write the least significant Size bytes from Value to
358 /// Dst.
359 void writeBytesUnaligned(uint64_t Value, uint8_t *Dst, unsigned Size) const;
360
361 /// \brief Given the common symbols discovered in the object file, emit a
362 /// new section for them and update the symbol mappings in the object and
363 /// symbol table.
364 void emitCommonSymbols(const ObjectFile &Obj, CommonSymbolList &CommonSymbols);
365
366 /// \brief Emits section data from the object file to the MemoryManager.
367 /// \param IsCode if it's true then allocateCodeSection() will be
368 /// used for emits, else allocateDataSection() will be used.
369 /// \return SectionID.
370 unsigned emitSection(const ObjectFile &Obj, const SectionRef &Section,
371 bool IsCode);
372
373 /// \brief Find Section in LocalSections. If the secton is not found - emit
374 /// it and store in LocalSections.
375 /// \param IsCode if it's true then allocateCodeSection() will be
376 /// used for emmits, else allocateDataSection() will be used.
377 /// \return SectionID.
378 unsigned findOrEmitSection(const ObjectFile &Obj, const SectionRef &Section,
379 bool IsCode, ObjSectionToIDMap &LocalSections);
380
381 // \brief Add a relocation entry that uses the given section.
382 void addRelocationForSection(const RelocationEntry &RE, unsigned SectionID);
383
384 // \brief Add a relocation entry that uses the given symbol. This symbol may
385 // be found in the global symbol table, or it may be external.
386 void addRelocationForSymbol(const RelocationEntry &RE, StringRef SymbolName);
387
388 /// \brief Emits long jump instruction to Addr.
389 /// \return Pointer to the memory area for emitting target address.
390 uint8_t *createStubFunction(uint8_t *Addr, unsigned AbiVariant = 0);
391
392 /// \brief Resolves relocations from Relocs list with address from Value.
393 void resolveRelocationList(const RelocationList &Relocs, uint64_t Value);
394
395 /// \brief A object file specific relocation resolver
396 /// \param RE The relocation to be resolved
397 /// \param Value Target symbol address to apply the relocation action
398 virtual void resolveRelocation(const RelocationEntry &RE, uint64_t Value) = 0;
399
400 /// \brief Parses one or more object file relocations (some object files use
401 /// relocation pairs) and stores it to Relocations or SymbolRelocations
402 /// (this depends on the object file type).
403 /// \return Iterator to the next relocation that needs to be parsed.
404 virtual relocation_iterator
405 processRelocationRef(unsigned SectionID, relocation_iterator RelI,
406 const ObjectFile &Obj, ObjSectionToIDMap &ObjSectionToID,
407 StubMap &Stubs) = 0;
408
409 /// \brief Resolve relocations to external symbols.
410 void resolveExternalSymbols();
411
412 // \brief Compute an upper bound of the memory that is required to load all
413 // sections
414 void computeTotalAllocSize(const ObjectFile &Obj, uint64_t &CodeSize,
415 uint64_t &DataSizeRO, uint64_t &DataSizeRW);
416
417 // \brief Compute the stub buffer size required for a section
418 unsigned computeSectionStubBufSize(const ObjectFile &Obj,
419 const SectionRef &Section);
420
421 // \brief Implementation of the generic part of the loadObject algorithm.
422 ObjSectionToIDMap loadObjectImpl(const object::ObjectFile &Obj);
423
424 // \brief Return true if the relocation R may require allocating a stub.
relocationNeedsStub(const RelocationRef & R)425 virtual bool relocationNeedsStub(const RelocationRef &R) const {
426 return true; // Conservative answer
427 }
428
429 public:
RuntimeDyldImpl(RuntimeDyld::MemoryManager & MemMgr,RuntimeDyld::SymbolResolver & Resolver)430 RuntimeDyldImpl(RuntimeDyld::MemoryManager &MemMgr,
431 RuntimeDyld::SymbolResolver &Resolver)
432 : MemMgr(MemMgr), Resolver(Resolver), Checker(nullptr),
433 ProcessAllSections(false), HasError(false) {
434 }
435
436 virtual ~RuntimeDyldImpl();
437
setProcessAllSections(bool ProcessAllSections)438 void setProcessAllSections(bool ProcessAllSections) {
439 this->ProcessAllSections = ProcessAllSections;
440 }
441
setRuntimeDyldChecker(RuntimeDyldCheckerImpl * Checker)442 void setRuntimeDyldChecker(RuntimeDyldCheckerImpl *Checker) {
443 this->Checker = Checker;
444 }
445
446 virtual std::unique_ptr<RuntimeDyld::LoadedObjectInfo>
447 loadObject(const object::ObjectFile &Obj) = 0;
448
getSymbolLocalAddress(StringRef Name)449 uint8_t* getSymbolLocalAddress(StringRef Name) const {
450 // FIXME: Just look up as a function for now. Overly simple of course.
451 // Work in progress.
452 RTDyldSymbolTable::const_iterator pos = GlobalSymbolTable.find(Name);
453 if (pos == GlobalSymbolTable.end())
454 return nullptr;
455 const auto &SymInfo = pos->second;
456 // Absolute symbols do not have a local address.
457 if (SymInfo.getSectionID() == AbsoluteSymbolSection)
458 return nullptr;
459 return getSectionAddress(SymInfo.getSectionID()) + SymInfo.getOffset();
460 }
461
getSymbol(StringRef Name)462 RuntimeDyld::SymbolInfo getSymbol(StringRef Name) const {
463 // FIXME: Just look up as a function for now. Overly simple of course.
464 // Work in progress.
465 RTDyldSymbolTable::const_iterator pos = GlobalSymbolTable.find(Name);
466 if (pos == GlobalSymbolTable.end())
467 return nullptr;
468 const auto &SymEntry = pos->second;
469 uint64_t SectionAddr = 0;
470 if (SymEntry.getSectionID() != AbsoluteSymbolSection)
471 SectionAddr = getSectionLoadAddress(SymEntry.getSectionID());
472 uint64_t TargetAddr = SectionAddr + SymEntry.getOffset();
473 return RuntimeDyld::SymbolInfo(TargetAddr, SymEntry.getFlags());
474 }
475
476 void resolveRelocations();
477
478 void reassignSectionAddress(unsigned SectionID, uint64_t Addr);
479
480 void mapSectionAddress(const void *LocalAddress, uint64_t TargetAddress);
481
482 // Is the linker in an error state?
hasError()483 bool hasError() { return HasError; }
484
485 // Mark the error condition as handled and continue.
clearError()486 void clearError() { HasError = false; }
487
488 // Get the error message.
getErrorString()489 StringRef getErrorString() { return ErrorStr; }
490
491 virtual bool isCompatibleFile(const ObjectFile &Obj) const = 0;
492
493 virtual void registerEHFrames();
494
495 virtual void deregisterEHFrames();
496
finalizeLoad(const ObjectFile & ObjImg,ObjSectionToIDMap & SectionMap)497 virtual void finalizeLoad(const ObjectFile &ObjImg,
498 ObjSectionToIDMap &SectionMap) {}
499 };
500
501 } // end namespace llvm
502
503 #endif
504