blob: 7130e0e17498a0a9897aee92d9b072596cf5a43c [file] [log] [blame]
Jim Grosbache0934be2012-01-16 23:50:58 +00001//===-- RuntimeDyldMachO.cpp - Run-time dynamic linker for MC-JIT -*- C++ -*-=//
Danil Malyshevcf852dc2011-07-13 07:57:58 +00002//
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// Implementation of the MC-JIT runtime dynamic linker.
11//
12//===----------------------------------------------------------------------===//
13
14#define DEBUG_TYPE "dyld"
15#include "llvm/ADT/OwningPtr.h"
16#include "llvm/ADT/StringRef.h"
17#include "llvm/ADT/STLExtras.h"
Eli Bendersky76463fd2012-01-22 07:05:02 +000018#include "RuntimeDyldMachO.h"
Danil Malyshevcf852dc2011-07-13 07:57:58 +000019using namespace llvm;
20using namespace llvm::object;
21
22namespace llvm {
23
Chandler Carruth3e296712012-03-22 05:44:06 +000024bool RuntimeDyldMachO::
25resolveRelocation(uint8_t *LocalAddress,
26 uint64_t FinalAddress,
27 uint64_t Value,
28 bool isPCRel,
29 unsigned Type,
30 unsigned Size,
31 int64_t Addend) {
Danil Malyshevcf852dc2011-07-13 07:57:58 +000032 // This just dispatches to the proper target specific routine.
Chandler Carruth3e296712012-03-22 05:44:06 +000033 switch (CPUType) {
Craig Topper85814382012-02-07 05:05:23 +000034 default: llvm_unreachable("Unsupported CPU type!");
Sean Callananb38aae42012-03-26 20:45:52 +000035 case mach::CTM_i386:
36 return resolveI386Relocation(LocalAddress,
37 FinalAddress,
38 (uintptr_t)Value,
39 isPCRel,
40 Type,
41 Size,
42 Addend);
Chandler Carruth3e296712012-03-22 05:44:06 +000043 case mach::CTM_x86_64:
44 return resolveX86_64Relocation(LocalAddress,
45 FinalAddress,
46 (uintptr_t)Value,
47 isPCRel,
48 Type,
49 Size,
50 Addend);
51 case mach::CTM_ARM:
52 return resolveARMRelocation(LocalAddress,
53 FinalAddress,
54 (uintptr_t)Value,
55 isPCRel,
56 Type,
57 Size,
58 Addend);
Danil Malyshevcf852dc2011-07-13 07:57:58 +000059 }
Danil Malyshevcf852dc2011-07-13 07:57:58 +000060}
61
62bool RuntimeDyldMachO::
Sean Callananb38aae42012-03-26 20:45:52 +000063resolveI386Relocation(uint8_t *LocalAddress,
64 uint64_t FinalAddress,
65 uint64_t Value,
66 bool isPCRel,
67 unsigned Type,
68 unsigned Size,
69 int64_t Addend) {
70 if (isPCRel)
71 Value -= FinalAddress + 4; // see resolveX86_64Relocation
72
73 switch (Type) {
74 default:
75 llvm_unreachable("Invalid relocation type!");
76 case macho::RIT_Vanilla: {
77 uint8_t *p = LocalAddress;
78 uint64_t ValueToWrite = Value + Addend;
79 for (unsigned i = 0; i < Size; ++i) {
80 *p++ = (uint8_t)(ValueToWrite & 0xff);
81 ValueToWrite >>= 8;
82 }
83 }
84 case macho::RIT_Difference:
85 case macho::RIT_Generic_LocalDifference:
86 case macho::RIT_Generic_PreboundLazyPointer:
87 return Error("Relocation type not implemented yet!");
88 }
89}
90
91bool RuntimeDyldMachO::
Sean Callanan61dfa772012-03-07 23:05:25 +000092resolveX86_64Relocation(uint8_t *LocalAddress,
93 uint64_t FinalAddress,
94 uint64_t Value,
95 bool isPCRel,
96 unsigned Type,
97 unsigned Size,
98 int64_t Addend) {
Danil Malyshevcf852dc2011-07-13 07:57:58 +000099 // If the relocation is PC-relative, the value to be encoded is the
100 // pointer difference.
101 if (isPCRel)
102 // FIXME: It seems this value needs to be adjusted by 4 for an effective PC
103 // address. Is that expected? Only for branches, perhaps?
Sean Callanan61dfa772012-03-07 23:05:25 +0000104 Value -= FinalAddress + 4;
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000105
106 switch(Type) {
107 default:
108 llvm_unreachable("Invalid relocation type!");
Jim Grosbach652ca2f2012-01-16 23:50:49 +0000109 case macho::RIT_X86_64_Signed1:
110 case macho::RIT_X86_64_Signed2:
111 case macho::RIT_X86_64_Signed4:
112 case macho::RIT_X86_64_Signed:
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000113 case macho::RIT_X86_64_Unsigned:
114 case macho::RIT_X86_64_Branch: {
Jim Grosbach652ca2f2012-01-16 23:50:49 +0000115 Value += Addend;
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000116 // Mask in the target value a byte at a time (we don't have an alignment
117 // guarantee for the target address, so this is safest).
Sean Callanan61dfa772012-03-07 23:05:25 +0000118 uint8_t *p = (uint8_t*)LocalAddress;
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000119 for (unsigned i = 0; i < Size; ++i) {
120 *p++ = (uint8_t)Value;
121 Value >>= 8;
122 }
123 return false;
124 }
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000125 case macho::RIT_X86_64_GOTLoad:
126 case macho::RIT_X86_64_GOT:
127 case macho::RIT_X86_64_Subtractor:
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000128 case macho::RIT_X86_64_TLV:
129 return Error("Relocation type not implemented yet!");
130 }
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000131}
132
Jim Grosbach61425c02012-01-16 22:26:39 +0000133bool RuntimeDyldMachO::
Sean Callanan61dfa772012-03-07 23:05:25 +0000134resolveARMRelocation(uint8_t *LocalAddress,
135 uint64_t FinalAddress,
136 uint64_t Value,
137 bool isPCRel,
138 unsigned Type,
139 unsigned Size,
140 int64_t Addend) {
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000141 // If the relocation is PC-relative, the value to be encoded is the
142 // pointer difference.
143 if (isPCRel) {
Sean Callanan61dfa772012-03-07 23:05:25 +0000144 Value -= FinalAddress;
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000145 // ARM PCRel relocations have an effective-PC offset of two instructions
146 // (four bytes in Thumb mode, 8 bytes in ARM mode).
147 // FIXME: For now, assume ARM mode.
148 Value -= 8;
149 }
150
151 switch(Type) {
152 default:
153 llvm_unreachable("Invalid relocation type!");
154 case macho::RIT_Vanilla: {
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000155 // Mask in the target value a byte at a time (we don't have an alignment
156 // guarantee for the target address, so this is safest).
Sean Callanan61dfa772012-03-07 23:05:25 +0000157 uint8_t *p = (uint8_t*)LocalAddress;
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000158 for (unsigned i = 0; i < Size; ++i) {
159 *p++ = (uint8_t)Value;
160 Value >>= 8;
161 }
162 break;
163 }
164 case macho::RIT_ARM_Branch24Bit: {
165 // Mask the value into the target address. We know instructions are
166 // 32-bit aligned, so we can do it all at once.
Sean Callanan61dfa772012-03-07 23:05:25 +0000167 uint32_t *p = (uint32_t*)LocalAddress;
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000168 // The low two bits of the value are not encoded.
169 Value >>= 2;
170 // Mask the value to 24 bits.
171 Value &= 0xffffff;
172 // FIXME: If the destination is a Thumb function (and the instruction
173 // is a non-predicated BL instruction), we need to change it to a BLX
174 // instruction instead.
175
176 // Insert the value into the instruction.
177 *p = (*p & ~0xffffff) | Value;
178 break;
179 }
180 case macho::RIT_ARM_ThumbBranch22Bit:
181 case macho::RIT_ARM_ThumbBranch32Bit:
182 case macho::RIT_ARM_Half:
183 case macho::RIT_ARM_HalfDifference:
184 case macho::RIT_Pair:
185 case macho::RIT_Difference:
186 case macho::RIT_ARM_LocalDifference:
187 case macho::RIT_ARM_PreboundLazyPointer:
188 return Error("Relocation type not implemented yet!");
189 }
190 return false;
191}
192
Chandler Carruth3e296712012-03-22 05:44:06 +0000193bool RuntimeDyldMachO::
194loadSegment32(const MachOObject *Obj,
195 const MachOObject::LoadCommandInfo *SegmentLCI,
196 const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
197 // FIXME: This should really be combined w/ loadSegment64. Templatized
198 // function on the 32/64 datatypes maybe?
199 InMemoryStruct<macho::SegmentLoadCommand> SegmentLC;
200 Obj->ReadSegmentLoadCommand(*SegmentLCI, SegmentLC);
201 if (!SegmentLC)
202 return Error("unable to load segment load command");
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000203
Jim Grosbach61425c02012-01-16 22:26:39 +0000204
Chandler Carruth3e296712012-03-22 05:44:06 +0000205 SmallVector<unsigned, 16> SectionMap;
206 for (unsigned SectNum = 0; SectNum != SegmentLC->NumSections; ++SectNum) {
207 InMemoryStruct<macho::Section> Sect;
208 Obj->ReadSection(*SegmentLCI, SectNum, Sect);
209 if (!Sect)
210 return Error("unable to load section: '" + Twine(SectNum) + "'");
211
212 // Allocate memory via the MM for the section.
213 uint8_t *Buffer;
214 uint32_t SectionID = Sections.size();
215 if (Sect->Flags == 0x80000400)
216 Buffer = MemMgr->allocateCodeSection(Sect->Size, Sect->Align, SectionID);
217 else
218 Buffer = MemMgr->allocateDataSection(Sect->Size, Sect->Align, SectionID);
219
220 DEBUG(dbgs() << "Loading "
221 << ((Sect->Flags == 0x80000400) ? "text" : "data")
222 << " (ID #" << SectionID << ")"
223 << " '" << Sect->SegmentName << ","
224 << Sect->Name << "' of size " << Sect->Size
225 << " to address " << Buffer << ".\n");
226
227 // Copy the payload from the object file into the allocated buffer.
228 uint8_t *Base = (uint8_t*)Obj->getData(SegmentLC->FileOffset,
229 SegmentLC->FileSize).data();
230 memcpy(Buffer, Base + Sect->Address, Sect->Size);
231
232 // Remember what got allocated for this SectionID.
233 Sections.push_back(sys::MemoryBlock(Buffer, Sect->Size));
234 SectionLocalMemToID[Buffer] = SectionID;
235
236 // By default, the load address of a section is its memory buffer.
237 SectionLoadAddress.push_back((uint64_t)Buffer);
238
239 // Keep a map of object file section numbers to corresponding SectionIDs
240 // while processing the file.
241 SectionMap.push_back(SectionID);
242 }
243
244 // Process the symbol table.
245 SmallVector<StringRef, 64> SymbolNames;
246 processSymbols32(Obj, SectionMap, SymbolNames, SymtabLC);
247
248 // Process the relocations for each section we're loading.
249 Relocations.grow(Relocations.size() + SegmentLC->NumSections);
250 Referrers.grow(Referrers.size() + SegmentLC->NumSections);
251 for (unsigned SectNum = 0; SectNum != SegmentLC->NumSections; ++SectNum) {
252 InMemoryStruct<macho::Section> Sect;
253 Obj->ReadSection(*SegmentLCI, SectNum, Sect);
254 if (!Sect)
255 return Error("unable to load section: '" + Twine(SectNum) + "'");
256 for (unsigned j = 0; j != Sect->NumRelocationTableEntries; ++j) {
257 InMemoryStruct<macho::RelocationEntry> RE;
258 Obj->ReadRelocationEntry(Sect->RelocationTableOffset, j, RE);
259 if (RE->Word0 & macho::RF_Scattered)
260 return Error("NOT YET IMPLEMENTED: scattered relocations.");
261 // Word0 of the relocation is the offset into the section where the
262 // relocation should be applied. We need to translate that into an
263 // offset into a function since that's our atom.
264 uint32_t Offset = RE->Word0;
265 bool isExtern = (RE->Word1 >> 27) & 1;
266
267 // FIXME: Get the relocation addend from the target address.
268 // FIXME: VERY imporant for internal relocations.
269
270 // Figure out the source symbol of the relocation. If isExtern is true,
271 // this relocation references the symbol table, otherwise it references
272 // a section in the same object, numbered from 1 through NumSections
273 // (SectionBases is [0, NumSections-1]).
274 uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value
275 if (!isExtern) {
276 assert(SourceNum > 0 && "Invalid relocation section number!");
277 unsigned SectionID = SectionMap[SourceNum - 1];
278 unsigned TargetID = SectionMap[SectNum];
279 DEBUG(dbgs() << "Internal relocation at Section #"
280 << TargetID << " + " << Offset
281 << " from Section #"
282 << SectionID << " (Word1: "
283 << format("0x%x", RE->Word1) << ")\n");
284
285 // Store the relocation information. It will get resolved when
286 // the section addresses are assigned.
287 uint32_t RelocationIndex = Relocations[SectionID].size();
288 Relocations[SectionID].push_back(RelocationEntry(TargetID,
289 Offset,
290 RE->Word1,
291 0 /*Addend*/));
292 Referrers[TargetID].push_back(Referrer(SectionID, RelocationIndex));
293 } else {
294 StringRef SourceName = SymbolNames[SourceNum];
295
296 // Now store the relocation information. Associate it with the source
297 // symbol. Just add it to the unresolved list and let the general
298 // path post-load resolve it if we know where the symbol is.
299 UnresolvedRelocations[SourceName].push_back(RelocationEntry(SectNum,
300 Offset,
301 RE->Word1,
302 0 /*Addend*/));
303 DEBUG(dbgs() << "Relocation at Section #" << SectNum << " + " << Offset
304 << " from '" << SourceName << "(Word1: "
305 << format("0x%x", RE->Word1) << ")\n");
306 }
Danil Malyshev799184d2012-03-21 21:06:29 +0000307 }
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000308 }
309
Chandler Carruth3e296712012-03-22 05:44:06 +0000310 // Resolve the addresses of any symbols that were defined in this segment.
311 for (int i = 0, e = SymbolNames.size(); i != e; ++i)
312 resolveSymbol(SymbolNames[i]);
Jim Grosbach61425c02012-01-16 22:26:39 +0000313
Chandler Carruth3e296712012-03-22 05:44:06 +0000314 return false;
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000315}
316
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000317
Chandler Carruth3e296712012-03-22 05:44:06 +0000318bool RuntimeDyldMachO::
319loadSegment64(const MachOObject *Obj,
320 const MachOObject::LoadCommandInfo *SegmentLCI,
321 const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
322 InMemoryStruct<macho::Segment64LoadCommand> Segment64LC;
323 Obj->ReadSegment64LoadCommand(*SegmentLCI, Segment64LC);
324 if (!Segment64LC)
325 return Error("unable to load segment load command");
326
327
328 SmallVector<unsigned, 16> SectionMap;
329 for (unsigned SectNum = 0; SectNum != Segment64LC->NumSections; ++SectNum) {
330 InMemoryStruct<macho::Section64> Sect;
331 Obj->ReadSection64(*SegmentLCI, SectNum, Sect);
332 if (!Sect)
333 return Error("unable to load section: '" + Twine(SectNum) + "'");
334
335 // Allocate memory via the MM for the section.
336 uint8_t *Buffer;
337 uint32_t SectionID = Sections.size();
338 unsigned Align = 1 << Sect->Align; // .o file has log2 alignment.
339 if (Sect->Flags == 0x80000400)
340 Buffer = MemMgr->allocateCodeSection(Sect->Size, Align, SectionID);
341 else
342 Buffer = MemMgr->allocateDataSection(Sect->Size, Align, SectionID);
343
344 DEBUG(dbgs() << "Loading "
345 << ((Sect->Flags == 0x80000400) ? "text" : "data")
346 << " (ID #" << SectionID << ")"
347 << " '" << Sect->SegmentName << ","
348 << Sect->Name << "' of size " << Sect->Size
349 << " (align " << Align << ")"
350 << " to address " << Buffer << ".\n");
351
352 // Copy the payload from the object file into the allocated buffer.
353 uint8_t *Base = (uint8_t*)Obj->getData(Segment64LC->FileOffset,
354 Segment64LC->FileSize).data();
355 memcpy(Buffer, Base + Sect->Address, Sect->Size);
356
357 // Remember what got allocated for this SectionID.
358 Sections.push_back(sys::MemoryBlock(Buffer, Sect->Size));
359 SectionLocalMemToID[Buffer] = SectionID;
360
361 // By default, the load address of a section is its memory buffer.
362 SectionLoadAddress.push_back((uint64_t)Buffer);
363
364 // Keep a map of object file section numbers to corresponding SectionIDs
365 // while processing the file.
366 SectionMap.push_back(SectionID);
367 }
368
369 // Process the symbol table.
370 SmallVector<StringRef, 64> SymbolNames;
371 processSymbols64(Obj, SectionMap, SymbolNames, SymtabLC);
372
373 // Process the relocations for each section we're loading.
374 Relocations.grow(Relocations.size() + Segment64LC->NumSections);
375 Referrers.grow(Referrers.size() + Segment64LC->NumSections);
376 for (unsigned SectNum = 0; SectNum != Segment64LC->NumSections; ++SectNum) {
377 InMemoryStruct<macho::Section64> Sect;
378 Obj->ReadSection64(*SegmentLCI, SectNum, Sect);
379 if (!Sect)
380 return Error("unable to load section: '" + Twine(SectNum) + "'");
381 for (unsigned j = 0; j != Sect->NumRelocationTableEntries; ++j) {
382 InMemoryStruct<macho::RelocationEntry> RE;
383 Obj->ReadRelocationEntry(Sect->RelocationTableOffset, j, RE);
384 if (RE->Word0 & macho::RF_Scattered)
385 return Error("NOT YET IMPLEMENTED: scattered relocations.");
386 // Word0 of the relocation is the offset into the section where the
387 // relocation should be applied. We need to translate that into an
388 // offset into a function since that's our atom.
389 uint32_t Offset = RE->Word0;
390 bool isExtern = (RE->Word1 >> 27) & 1;
391
392 // FIXME: Get the relocation addend from the target address.
393 // FIXME: VERY imporant for internal relocations.
394
395 // Figure out the source symbol of the relocation. If isExtern is true,
396 // this relocation references the symbol table, otherwise it references
397 // a section in the same object, numbered from 1 through NumSections
398 // (SectionBases is [0, NumSections-1]).
399 uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value
400 if (!isExtern) {
401 assert(SourceNum > 0 && "Invalid relocation section number!");
402 unsigned SectionID = SectionMap[SourceNum - 1];
403 unsigned TargetID = SectionMap[SectNum];
404 DEBUG(dbgs() << "Internal relocation at Section #"
405 << TargetID << " + " << Offset
406 << " from Section #"
407 << SectionID << " (Word1: "
408 << format("0x%x", RE->Word1) << ")\n");
409
410 // Store the relocation information. It will get resolved when
411 // the section addresses are assigned.
412 uint32_t RelocationIndex = Relocations[SectionID].size();
413 Relocations[SectionID].push_back(RelocationEntry(TargetID,
414 Offset,
415 RE->Word1,
416 0 /*Addend*/));
417 Referrers[TargetID].push_back(Referrer(SectionID, RelocationIndex));
418 } else {
419 StringRef SourceName = SymbolNames[SourceNum];
420
421 // Now store the relocation information. Associate it with the source
422 // symbol. Just add it to the unresolved list and let the general
423 // path post-load resolve it if we know where the symbol is.
424 UnresolvedRelocations[SourceName].push_back(RelocationEntry(SectNum,
425 Offset,
426 RE->Word1,
427 0 /*Addend*/));
428 DEBUG(dbgs() << "Relocation at Section #" << SectNum << " + " << Offset
429 << " from '" << SourceName << "(Word1: "
430 << format("0x%x", RE->Word1) << ")\n");
431 }
432 }
433 }
434
435 // Resolve the addresses of any symbols that were defined in this segment.
436 for (int i = 0, e = SymbolNames.size(); i != e; ++i)
437 resolveSymbol(SymbolNames[i]);
438
439 return false;
440}
441
442bool RuntimeDyldMachO::
443processSymbols32(const MachOObject *Obj,
444 SmallVectorImpl<unsigned> &SectionMap,
445 SmallVectorImpl<StringRef> &SymbolNames,
446 const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
447 // FIXME: Combine w/ processSymbols64. Factor 64/32 datatype and such.
448 for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) {
449 InMemoryStruct<macho::SymbolTableEntry> STE;
450 Obj->ReadSymbolTableEntry(SymtabLC->SymbolTableOffset, i, STE);
451 if (!STE)
452 return Error("unable to read symbol: '" + Twine(i) + "'");
453 // Get the symbol name.
454 StringRef Name = Obj->getStringAtIndex(STE->StringIndex);
455 SymbolNames.push_back(Name);
456
457 // FIXME: Check the symbol type and flags.
458 if (STE->Type != 0xF) // external, defined in this segment.
459 continue;
460 // Flags in the upper nibble we don't care about.
461 if ((STE->Flags & 0xf) != 0x0)
462 continue;
463
464 // Remember the symbol.
465 uint32_t SectionID = SectionMap[STE->SectionIndex - 1];
466 SymbolTable[Name] = SymbolLoc(SectionID, STE->Value);
467
468 DEBUG(dbgs() << "Symbol: '" << Name << "' @ "
469 << (getSectionAddress(SectionID) + STE->Value)
470 << "\n");
471 }
472 return false;
473}
474
475bool RuntimeDyldMachO::
476processSymbols64(const MachOObject *Obj,
477 SmallVectorImpl<unsigned> &SectionMap,
478 SmallVectorImpl<StringRef> &SymbolNames,
479 const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) {
480 for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) {
481 InMemoryStruct<macho::Symbol64TableEntry> STE;
482 Obj->ReadSymbol64TableEntry(SymtabLC->SymbolTableOffset, i, STE);
483 if (!STE)
484 return Error("unable to read symbol: '" + Twine(i) + "'");
485 // Get the symbol name.
486 StringRef Name = Obj->getStringAtIndex(STE->StringIndex);
487 SymbolNames.push_back(Name);
488
489 // FIXME: Check the symbol type and flags.
490 if (STE->Type != 0xF) // external, defined in this segment.
491 continue;
492 // Flags in the upper nibble we don't care about.
493 if ((STE->Flags & 0xf) != 0x0)
494 continue;
495
496 // Remember the symbol.
497 uint32_t SectionID = SectionMap[STE->SectionIndex - 1];
498 SymbolTable[Name] = SymbolLoc(SectionID, STE->Value);
499
500 DEBUG(dbgs() << "Symbol: '" << Name << "' @ "
501 << (getSectionAddress(SectionID) + STE->Value)
502 << "\n");
503 }
504 return false;
505}
506
507// resolveSymbol - Resolve any relocations to the specified symbol if
508// we know where it lives.
509void RuntimeDyldMachO::resolveSymbol(StringRef Name) {
510 StringMap<SymbolLoc>::const_iterator Loc = SymbolTable.find(Name);
511 if (Loc == SymbolTable.end())
512 return;
513
514 RelocationList &Relocs = UnresolvedRelocations[Name];
515 DEBUG(dbgs() << "Resolving symbol '" << Name << "'\n");
516 for (int i = 0, e = Relocs.size(); i != e; ++i) {
517 // Change the relocation to be section relative rather than symbol
518 // relative and move it to the resolved relocation list.
519 RelocationEntry Entry = Relocs[i];
520 Entry.Addend += Loc->second.second;
521 uint32_t RelocationIndex = Relocations[Loc->second.first].size();
522 Relocations[Loc->second.first].push_back(Entry);
523 Referrers[Entry.SectionID].push_back(Referrer(Loc->second.first, RelocationIndex));
524 }
525 // FIXME: Keep a worklist of the relocations we've added so that we can
526 // resolve more selectively later.
527 Relocs.clear();
528}
529
530bool RuntimeDyldMachO::loadObject(MemoryBuffer *InputBuffer) {
531 // If the linker is in an error state, don't do anything.
532 if (hasError())
533 return true;
534 // Load the Mach-O wrapper object.
535 std::string ErrorStr;
536 OwningPtr<MachOObject> Obj(
537 MachOObject::LoadFromBuffer(InputBuffer, &ErrorStr));
538 if (!Obj)
539 return Error("unable to load object: '" + ErrorStr + "'");
540
541 // Get the CPU type information from the header.
542 const macho::Header &Header = Obj->getHeader();
543
544 // FIXME: Error checking that the loaded object is compatible with
545 // the system we're running on.
546 CPUType = Header.CPUType;
547 CPUSubtype = Header.CPUSubtype;
548
549 // Validate that the load commands match what we expect.
550 const MachOObject::LoadCommandInfo *SegmentLCI = 0, *SymtabLCI = 0,
551 *DysymtabLCI = 0;
552 for (unsigned i = 0; i != Header.NumLoadCommands; ++i) {
553 const MachOObject::LoadCommandInfo &LCI = Obj->getLoadCommandInfo(i);
554 switch (LCI.Command.Type) {
555 case macho::LCT_Segment:
556 case macho::LCT_Segment64:
557 if (SegmentLCI)
558 return Error("unexpected input object (multiple segments)");
559 SegmentLCI = &LCI;
560 break;
561 case macho::LCT_Symtab:
562 if (SymtabLCI)
563 return Error("unexpected input object (multiple symbol tables)");
564 SymtabLCI = &LCI;
565 break;
566 case macho::LCT_Dysymtab:
567 if (DysymtabLCI)
568 return Error("unexpected input object (multiple symbol tables)");
569 DysymtabLCI = &LCI;
570 break;
571 default:
572 return Error("unexpected input object (unexpected load command");
573 }
574 }
575
576 if (!SymtabLCI)
577 return Error("no symbol table found in object");
578 if (!SegmentLCI)
579 return Error("no segments found in object");
580
581 // Read and register the symbol table data.
582 InMemoryStruct<macho::SymtabLoadCommand> SymtabLC;
583 Obj->ReadSymtabLoadCommand(*SymtabLCI, SymtabLC);
584 if (!SymtabLC)
585 return Error("unable to load symbol table load command");
586 Obj->RegisterStringTable(*SymtabLC);
587
588 // Read the dynamic link-edit information, if present (not present in static
589 // objects).
590 if (DysymtabLCI) {
591 InMemoryStruct<macho::DysymtabLoadCommand> DysymtabLC;
592 Obj->ReadDysymtabLoadCommand(*DysymtabLCI, DysymtabLC);
593 if (!DysymtabLC)
594 return Error("unable to load dynamic link-exit load command");
595
596 // FIXME: We don't support anything interesting yet.
597// if (DysymtabLC->LocalSymbolsIndex != 0)
598// return Error("NOT YET IMPLEMENTED: local symbol entries");
599// if (DysymtabLC->ExternalSymbolsIndex != 0)
600// return Error("NOT YET IMPLEMENTED: non-external symbol entries");
601// if (DysymtabLC->UndefinedSymbolsIndex != SymtabLC->NumSymbolTableEntries)
602// return Error("NOT YET IMPLEMENTED: undefined symbol entries");
603 }
604
605 // Load the segment load command.
606 if (SegmentLCI->Command.Type == macho::LCT_Segment) {
607 if (loadSegment32(Obj.get(), SegmentLCI, SymtabLC))
608 return true;
609 } else {
610 if (loadSegment64(Obj.get(), SegmentLCI, SymtabLC))
611 return true;
612 }
613
614 // Assign the addresses of the sections from the object so that any
615 // relocations to them get set properly.
616 // FIXME: This is done directly from the client at the moment. We should
617 // default the values to the local storage, at least when the target arch
618 // is the same as the host arch.
619
620 return false;
621}
622
623// Assign an address to a symbol name and resolve all the relocations
624// associated with it.
625void RuntimeDyldMachO::reassignSectionAddress(unsigned SectionID,
626 uint64_t Addr) {
627 // The address to use for relocation resolution is not
628 // the address of the local section buffer. We must be doing
629 // a remote execution environment of some sort. Re-apply any
630 // relocations referencing this section with the given address.
631 //
632 // Addr is a uint64_t because we can't assume the pointer width
633 // of the target is the same as that of the host. Just use a generic
634 // "big enough" type.
635
636 SectionLoadAddress[SectionID] = Addr;
637
638 RelocationList &Relocs = Relocations[SectionID];
639 for (unsigned i = 0, e = Relocs.size(); i != e; ++i) {
640 RelocationEntry &RE = Relocs[i];
641 uint8_t *Target = (uint8_t*)Sections[RE.SectionID].base() + RE.Offset;
642 uint64_t FinalTarget = (uint64_t)SectionLoadAddress[RE.SectionID] + RE.Offset;
643 bool isPCRel = (RE.Data >> 24) & 1;
644 unsigned Type = (RE.Data >> 28) & 0xf;
645 unsigned Size = 1 << ((RE.Data >> 25) & 3);
646
647 DEBUG(dbgs() << "Resolving relocation at Section #" << RE.SectionID
648 << " + " << RE.Offset << " (" << format("%p", Target) << ")"
649 << " from Section #" << SectionID << " (" << format("%p", Addr) << ")"
650 << "(" << (isPCRel ? "pcrel" : "absolute")
651 << ", type: " << Type << ", Size: " << Size << ", Addend: "
652 << RE.Addend << ").\n");
653
654 resolveRelocation(Target,
655 FinalTarget,
656 Addr,
657 isPCRel,
658 Type,
659 Size,
660 RE.Addend);
661 }
662 ReferrerList &Refers = Referrers[SectionID];
663 for (unsigned i = 0, e = Refers.size(); i != e; ++i) {
664 Referrer &R = Refers[i];
665 RelocationEntry &RE = Relocations[R.SectionID][R.Index];
666 uint8_t *Target = (uint8_t*)Sections[RE.SectionID].base() + RE.Offset;
667 uint64_t FinalTarget = (uint64_t)SectionLoadAddress[RE.SectionID] + RE.Offset;
668 bool isPCRel = (RE.Data >> 24) & 1;
669 unsigned Type = (RE.Data >> 28) & 0xf;
670 unsigned Size = 1 << ((RE.Data >> 25) & 3);
671
672 DEBUG(dbgs() << "Resolving relocation at Section #" << RE.SectionID
673 << " + " << RE.Offset << " (" << format("%p", Target) << ")"
674 << " from Section #" << SectionID << " (" << format("%p", Addr) << ")"
675 << "(" << (isPCRel ? "pcrel" : "absolute")
676 << ", type: " << Type << ", Size: " << Size << ", Addend: "
677 << RE.Addend << ").\n");
678
679 resolveRelocation(Target,
680 FinalTarget,
681 Addr,
682 isPCRel,
683 Type,
684 Size,
685 RE.Addend);
686 }
687}
688
689bool RuntimeDyldMachO::isKnownFormat(const MemoryBuffer *InputBuffer) {
Danil Malyshevcf852dc2011-07-13 07:57:58 +0000690 StringRef Magic = InputBuffer->getBuffer().slice(0, 4);
691 if (Magic == "\xFE\xED\xFA\xCE") return true;
692 if (Magic == "\xCE\xFA\xED\xFE") return true;
693 if (Magic == "\xFE\xED\xFA\xCF") return true;
694 if (Magic == "\xCF\xFA\xED\xFE") return true;
695 return false;
696}
697
698} // end namespace llvm