1 | //===- lib/MC/MachObjectWriter.cpp - Mach-O File Writer -------------------===// |
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 "llvm/ADT/DenseMap.h" |
10 | #include "llvm/ADT/Twine.h" |
11 | #include "llvm/ADT/iterator_range.h" |
12 | #include "llvm/BinaryFormat/MachO.h" |
13 | #include "llvm/MC/MCAsmBackend.h" |
14 | #include "llvm/MC/MCAsmInfoDarwin.h" |
15 | #include "llvm/MC/MCAssembler.h" |
16 | #include "llvm/MC/MCContext.h" |
17 | #include "llvm/MC/MCDirectives.h" |
18 | #include "llvm/MC/MCExpr.h" |
19 | #include "llvm/MC/MCFixupKindInfo.h" |
20 | #include "llvm/MC/MCFragment.h" |
21 | #include "llvm/MC/MCMachObjectWriter.h" |
22 | #include "llvm/MC/MCObjectFileInfo.h" |
23 | #include "llvm/MC/MCObjectWriter.h" |
24 | #include "llvm/MC/MCSection.h" |
25 | #include "llvm/MC/MCSectionMachO.h" |
26 | #include "llvm/MC/MCSymbol.h" |
27 | #include "llvm/MC/MCSymbolMachO.h" |
28 | #include "llvm/MC/MCValue.h" |
29 | #include "llvm/Support/Alignment.h" |
30 | #include "llvm/Support/Casting.h" |
31 | #include "llvm/Support/Debug.h" |
32 | #include "llvm/Support/ErrorHandling.h" |
33 | #include "llvm/Support/LEB128.h" |
34 | #include "llvm/Support/MathExtras.h" |
35 | #include "llvm/Support/raw_ostream.h" |
36 | #include <algorithm> |
37 | #include <cassert> |
38 | #include <cstdint> |
39 | #include <string> |
40 | #include <utility> |
41 | #include <vector> |
42 | |
43 | using namespace llvm; |
44 | |
45 | #define DEBUG_TYPE "mc" |
46 | |
47 | void MachObjectWriter::reset() { |
48 | Relocations.clear(); |
49 | IndirectSymBase.clear(); |
50 | IndirectSymbols.clear(); |
51 | DataRegions.clear(); |
52 | SectionAddress.clear(); |
53 | SectionOrder.clear(); |
54 | StringTable.clear(); |
55 | LocalSymbolData.clear(); |
56 | ExternalSymbolData.clear(); |
57 | UndefinedSymbolData.clear(); |
58 | LOHContainer.reset(); |
59 | VersionInfo.Major = 0; |
60 | VersionInfo.SDKVersion = VersionTuple(); |
61 | TargetVariantVersionInfo.Major = 0; |
62 | TargetVariantVersionInfo.SDKVersion = VersionTuple(); |
63 | LinkerOptions.clear(); |
64 | MCObjectWriter::reset(); |
65 | } |
66 | |
67 | bool MachObjectWriter::doesSymbolRequireExternRelocation(const MCSymbol &S) { |
68 | // Undefined symbols are always extern. |
69 | if (S.isUndefined()) |
70 | return true; |
71 | |
72 | // References to weak definitions require external relocation entries; the |
73 | // definition may not always be the one in the same object file. |
74 | if (cast<MCSymbolMachO>(Val: S).isWeakDefinition()) |
75 | return true; |
76 | |
77 | // Otherwise, we can use an internal relocation. |
78 | return false; |
79 | } |
80 | |
81 | bool MachObjectWriter:: |
82 | MachSymbolData::operator<(const MachSymbolData &RHS) const { |
83 | return Symbol->getName() < RHS.Symbol->getName(); |
84 | } |
85 | |
86 | bool MachObjectWriter::isFixupKindPCRel(const MCAssembler &Asm, unsigned Kind) { |
87 | const MCFixupKindInfo &FKI = Asm.getBackend().getFixupKindInfo( |
88 | Kind: (MCFixupKind) Kind); |
89 | |
90 | return FKI.Flags & MCFixupKindInfo::FKF_IsPCRel; |
91 | } |
92 | |
93 | uint64_t |
94 | MachObjectWriter::getFragmentAddress(const MCAssembler &Asm, |
95 | const MCFragment *Fragment) const { |
96 | return getSectionAddress(Sec: Fragment->getParent()) + |
97 | Asm.getFragmentOffset(F: *Fragment); |
98 | } |
99 | |
100 | uint64_t MachObjectWriter::getSymbolAddress(const MCSymbol &S, |
101 | const MCAssembler &Asm) const { |
102 | // If this is a variable, then recursively evaluate now. |
103 | if (S.isVariable()) { |
104 | if (const MCConstantExpr *C = |
105 | dyn_cast<const MCConstantExpr>(Val: S.getVariableValue())) |
106 | return C->getValue(); |
107 | |
108 | MCValue Target; |
109 | if (!S.getVariableValue()->evaluateAsRelocatable(Res&: Target, Asm: &Asm, Fixup: nullptr)) |
110 | report_fatal_error(reason: "unable to evaluate offset for variable '" + |
111 | S.getName() + "'" ); |
112 | |
113 | // Verify that any used symbols are defined. |
114 | if (Target.getSymA() && Target.getSymA()->getSymbol().isUndefined()) |
115 | report_fatal_error(reason: "unable to evaluate offset to undefined symbol '" + |
116 | Target.getSymA()->getSymbol().getName() + "'" ); |
117 | if (Target.getSymB() && Target.getSymB()->getSymbol().isUndefined()) |
118 | report_fatal_error(reason: "unable to evaluate offset to undefined symbol '" + |
119 | Target.getSymB()->getSymbol().getName() + "'" ); |
120 | |
121 | uint64_t Address = Target.getConstant(); |
122 | if (Target.getSymA()) |
123 | Address += getSymbolAddress(S: Target.getSymA()->getSymbol(), Asm); |
124 | if (Target.getSymB()) |
125 | Address += getSymbolAddress(S: Target.getSymB()->getSymbol(), Asm); |
126 | return Address; |
127 | } |
128 | |
129 | return getSectionAddress(Sec: S.getFragment()->getParent()) + |
130 | Asm.getSymbolOffset(S); |
131 | } |
132 | |
133 | uint64_t MachObjectWriter::getPaddingSize(const MCAssembler &Asm, |
134 | const MCSection *Sec) const { |
135 | uint64_t EndAddr = getSectionAddress(Sec) + Asm.getSectionAddressSize(Sec: *Sec); |
136 | unsigned Next = cast<MCSectionMachO>(Val: Sec)->getLayoutOrder() + 1; |
137 | if (Next >= SectionOrder.size()) |
138 | return 0; |
139 | |
140 | const MCSection &NextSec = *SectionOrder[Next]; |
141 | if (NextSec.isVirtualSection()) |
142 | return 0; |
143 | return offsetToAlignment(Value: EndAddr, Alignment: NextSec.getAlign()); |
144 | } |
145 | |
146 | static bool isSymbolLinkerVisible(const MCSymbol &Symbol) { |
147 | // Non-temporary labels should always be visible to the linker. |
148 | if (!Symbol.isTemporary()) |
149 | return true; |
150 | |
151 | if (Symbol.isUsedInReloc()) |
152 | return true; |
153 | |
154 | return false; |
155 | } |
156 | |
157 | const MCSymbol *MachObjectWriter::getAtom(const MCSymbol &S) const { |
158 | // Linker visible symbols define atoms. |
159 | if (isSymbolLinkerVisible(Symbol: S)) |
160 | return &S; |
161 | |
162 | // Absolute and undefined symbols have no defining atom. |
163 | if (!S.isInSection()) |
164 | return nullptr; |
165 | |
166 | // Non-linker visible symbols in sections which can't be atomized have no |
167 | // defining atom. |
168 | if (!MCAsmInfoDarwin::isSectionAtomizableBySymbols( |
169 | Section: *S.getFragment()->getParent())) |
170 | return nullptr; |
171 | |
172 | // Otherwise, return the atom for the containing fragment. |
173 | return S.getFragment()->getAtom(); |
174 | } |
175 | |
176 | void MachObjectWriter::(MachO::HeaderFileType Type, |
177 | unsigned NumLoadCommands, |
178 | unsigned LoadCommandsSize, |
179 | bool SubsectionsViaSymbols) { |
180 | uint32_t Flags = 0; |
181 | |
182 | if (SubsectionsViaSymbols) |
183 | Flags |= MachO::MH_SUBSECTIONS_VIA_SYMBOLS; |
184 | |
185 | // struct mach_header (28 bytes) or |
186 | // struct mach_header_64 (32 bytes) |
187 | |
188 | uint64_t Start = W.OS.tell(); |
189 | (void) Start; |
190 | |
191 | W.write<uint32_t>(Val: is64Bit() ? MachO::MH_MAGIC_64 : MachO::MH_MAGIC); |
192 | |
193 | W.write<uint32_t>(Val: TargetObjectWriter->getCPUType()); |
194 | W.write<uint32_t>(Val: TargetObjectWriter->getCPUSubtype()); |
195 | |
196 | W.write<uint32_t>(Val: Type); |
197 | W.write<uint32_t>(Val: NumLoadCommands); |
198 | W.write<uint32_t>(Val: LoadCommandsSize); |
199 | W.write<uint32_t>(Val: Flags); |
200 | if (is64Bit()) |
201 | W.write<uint32_t>(Val: 0); // reserved |
202 | |
203 | assert(W.OS.tell() - Start == (is64Bit() ? sizeof(MachO::mach_header_64) |
204 | : sizeof(MachO::mach_header))); |
205 | } |
206 | |
207 | void MachObjectWriter::writeWithPadding(StringRef Str, uint64_t Size) { |
208 | assert(Size >= Str.size()); |
209 | W.OS << Str; |
210 | W.OS.write_zeros(NumZeros: Size - Str.size()); |
211 | } |
212 | |
213 | /// writeSegmentLoadCommand - Write a segment load command. |
214 | /// |
215 | /// \param NumSections The number of sections in this segment. |
216 | /// \param SectionDataSize The total size of the sections. |
217 | void MachObjectWriter::writeSegmentLoadCommand( |
218 | StringRef Name, unsigned NumSections, uint64_t VMAddr, uint64_t VMSize, |
219 | uint64_t SectionDataStartOffset, uint64_t SectionDataSize, uint32_t MaxProt, |
220 | uint32_t InitProt) { |
221 | // struct segment_command (56 bytes) or |
222 | // struct segment_command_64 (72 bytes) |
223 | |
224 | uint64_t Start = W.OS.tell(); |
225 | (void) Start; |
226 | |
227 | unsigned SegmentLoadCommandSize = |
228 | is64Bit() ? sizeof(MachO::segment_command_64): |
229 | sizeof(MachO::segment_command); |
230 | W.write<uint32_t>(Val: is64Bit() ? MachO::LC_SEGMENT_64 : MachO::LC_SEGMENT); |
231 | W.write<uint32_t>(Val: SegmentLoadCommandSize + |
232 | NumSections * (is64Bit() ? sizeof(MachO::section_64) : |
233 | sizeof(MachO::section))); |
234 | |
235 | writeWithPadding(Str: Name, Size: 16); |
236 | if (is64Bit()) { |
237 | W.write<uint64_t>(Val: VMAddr); // vmaddr |
238 | W.write<uint64_t>(Val: VMSize); // vmsize |
239 | W.write<uint64_t>(Val: SectionDataStartOffset); // file offset |
240 | W.write<uint64_t>(Val: SectionDataSize); // file size |
241 | } else { |
242 | W.write<uint32_t>(Val: VMAddr); // vmaddr |
243 | W.write<uint32_t>(Val: VMSize); // vmsize |
244 | W.write<uint32_t>(Val: SectionDataStartOffset); // file offset |
245 | W.write<uint32_t>(Val: SectionDataSize); // file size |
246 | } |
247 | // maxprot |
248 | W.write<uint32_t>(Val: MaxProt); |
249 | // initprot |
250 | W.write<uint32_t>(Val: InitProt); |
251 | W.write<uint32_t>(Val: NumSections); |
252 | W.write<uint32_t>(Val: 0); // flags |
253 | |
254 | assert(W.OS.tell() - Start == SegmentLoadCommandSize); |
255 | } |
256 | |
257 | void MachObjectWriter::writeSection(const MCAssembler &Asm, |
258 | const MCSection &Sec, uint64_t VMAddr, |
259 | uint64_t FileOffset, unsigned Flags, |
260 | uint64_t RelocationsStart, |
261 | unsigned NumRelocations) { |
262 | uint64_t SectionSize = Asm.getSectionAddressSize(Sec); |
263 | const MCSectionMachO &Section = cast<MCSectionMachO>(Val: Sec); |
264 | |
265 | // The offset is unused for virtual sections. |
266 | if (Section.isVirtualSection()) { |
267 | assert(Asm.getSectionFileSize(Sec) == 0 && "Invalid file size!" ); |
268 | FileOffset = 0; |
269 | } |
270 | |
271 | // struct section (68 bytes) or |
272 | // struct section_64 (80 bytes) |
273 | |
274 | uint64_t Start = W.OS.tell(); |
275 | (void) Start; |
276 | |
277 | writeWithPadding(Str: Section.getName(), Size: 16); |
278 | writeWithPadding(Str: Section.getSegmentName(), Size: 16); |
279 | if (is64Bit()) { |
280 | W.write<uint64_t>(Val: VMAddr); // address |
281 | W.write<uint64_t>(Val: SectionSize); // size |
282 | } else { |
283 | W.write<uint32_t>(Val: VMAddr); // address |
284 | W.write<uint32_t>(Val: SectionSize); // size |
285 | } |
286 | assert(isUInt<32>(FileOffset) && "Cannot encode offset of section" ); |
287 | W.write<uint32_t>(Val: FileOffset); |
288 | |
289 | W.write<uint32_t>(Val: Log2(A: Section.getAlign())); |
290 | assert((!NumRelocations || isUInt<32>(RelocationsStart)) && |
291 | "Cannot encode offset of relocations" ); |
292 | W.write<uint32_t>(Val: NumRelocations ? RelocationsStart : 0); |
293 | W.write<uint32_t>(Val: NumRelocations); |
294 | W.write<uint32_t>(Val: Flags); |
295 | W.write<uint32_t>(Val: IndirectSymBase.lookup(Val: &Sec)); // reserved1 |
296 | W.write<uint32_t>(Val: Section.getStubSize()); // reserved2 |
297 | if (is64Bit()) |
298 | W.write<uint32_t>(Val: 0); // reserved3 |
299 | |
300 | assert(W.OS.tell() - Start == |
301 | (is64Bit() ? sizeof(MachO::section_64) : sizeof(MachO::section))); |
302 | } |
303 | |
304 | void MachObjectWriter::writeSymtabLoadCommand(uint32_t SymbolOffset, |
305 | uint32_t NumSymbols, |
306 | uint32_t StringTableOffset, |
307 | uint32_t StringTableSize) { |
308 | // struct symtab_command (24 bytes) |
309 | |
310 | uint64_t Start = W.OS.tell(); |
311 | (void) Start; |
312 | |
313 | W.write<uint32_t>(Val: MachO::LC_SYMTAB); |
314 | W.write<uint32_t>(Val: sizeof(MachO::symtab_command)); |
315 | W.write<uint32_t>(Val: SymbolOffset); |
316 | W.write<uint32_t>(Val: NumSymbols); |
317 | W.write<uint32_t>(Val: StringTableOffset); |
318 | W.write<uint32_t>(Val: StringTableSize); |
319 | |
320 | assert(W.OS.tell() - Start == sizeof(MachO::symtab_command)); |
321 | } |
322 | |
323 | void MachObjectWriter::writeDysymtabLoadCommand(uint32_t FirstLocalSymbol, |
324 | uint32_t NumLocalSymbols, |
325 | uint32_t FirstExternalSymbol, |
326 | uint32_t NumExternalSymbols, |
327 | uint32_t FirstUndefinedSymbol, |
328 | uint32_t NumUndefinedSymbols, |
329 | uint32_t IndirectSymbolOffset, |
330 | uint32_t NumIndirectSymbols) { |
331 | // struct dysymtab_command (80 bytes) |
332 | |
333 | uint64_t Start = W.OS.tell(); |
334 | (void) Start; |
335 | |
336 | W.write<uint32_t>(Val: MachO::LC_DYSYMTAB); |
337 | W.write<uint32_t>(Val: sizeof(MachO::dysymtab_command)); |
338 | W.write<uint32_t>(Val: FirstLocalSymbol); |
339 | W.write<uint32_t>(Val: NumLocalSymbols); |
340 | W.write<uint32_t>(Val: FirstExternalSymbol); |
341 | W.write<uint32_t>(Val: NumExternalSymbols); |
342 | W.write<uint32_t>(Val: FirstUndefinedSymbol); |
343 | W.write<uint32_t>(Val: NumUndefinedSymbols); |
344 | W.write<uint32_t>(Val: 0); // tocoff |
345 | W.write<uint32_t>(Val: 0); // ntoc |
346 | W.write<uint32_t>(Val: 0); // modtaboff |
347 | W.write<uint32_t>(Val: 0); // nmodtab |
348 | W.write<uint32_t>(Val: 0); // extrefsymoff |
349 | W.write<uint32_t>(Val: 0); // nextrefsyms |
350 | W.write<uint32_t>(Val: IndirectSymbolOffset); |
351 | W.write<uint32_t>(Val: NumIndirectSymbols); |
352 | W.write<uint32_t>(Val: 0); // extreloff |
353 | W.write<uint32_t>(Val: 0); // nextrel |
354 | W.write<uint32_t>(Val: 0); // locreloff |
355 | W.write<uint32_t>(Val: 0); // nlocrel |
356 | |
357 | assert(W.OS.tell() - Start == sizeof(MachO::dysymtab_command)); |
358 | } |
359 | |
360 | MachObjectWriter::MachSymbolData * |
361 | MachObjectWriter::findSymbolData(const MCSymbol &Sym) { |
362 | for (auto *SymbolData : |
363 | {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData}) |
364 | for (MachSymbolData &Entry : *SymbolData) |
365 | if (Entry.Symbol == &Sym) |
366 | return &Entry; |
367 | |
368 | return nullptr; |
369 | } |
370 | |
371 | const MCSymbol &MachObjectWriter::findAliasedSymbol(const MCSymbol &Sym) const { |
372 | const MCSymbol *S = &Sym; |
373 | while (S->isVariable()) { |
374 | const MCExpr *Value = S->getVariableValue(); |
375 | const auto *Ref = dyn_cast<MCSymbolRefExpr>(Val: Value); |
376 | if (!Ref) |
377 | return *S; |
378 | S = &Ref->getSymbol(); |
379 | } |
380 | return *S; |
381 | } |
382 | |
383 | void MachObjectWriter::writeNlist(MachSymbolData &MSD, const MCAssembler &Asm) { |
384 | const MCSymbol *Symbol = MSD.Symbol; |
385 | const auto &Data = cast<MCSymbolMachO>(Val: *Symbol); |
386 | const MCSymbol *AliasedSymbol = &findAliasedSymbol(Sym: *Symbol); |
387 | uint8_t SectionIndex = MSD.SectionIndex; |
388 | uint8_t Type = 0; |
389 | uint64_t Address = 0; |
390 | bool IsAlias = Symbol != AliasedSymbol; |
391 | |
392 | const MCSymbol &OrigSymbol = *Symbol; |
393 | MachSymbolData *AliaseeInfo; |
394 | if (IsAlias) { |
395 | AliaseeInfo = findSymbolData(Sym: *AliasedSymbol); |
396 | if (AliaseeInfo) |
397 | SectionIndex = AliaseeInfo->SectionIndex; |
398 | Symbol = AliasedSymbol; |
399 | // FIXME: Should this update Data as well? |
400 | } |
401 | |
402 | // Set the N_TYPE bits. See <mach-o/nlist.h>. |
403 | // |
404 | // FIXME: Are the prebound or indirect fields possible here? |
405 | if (IsAlias && Symbol->isUndefined()) |
406 | Type = MachO::N_INDR; |
407 | else if (Symbol->isUndefined()) |
408 | Type = MachO::N_UNDF; |
409 | else if (Symbol->isAbsolute()) |
410 | Type = MachO::N_ABS; |
411 | else |
412 | Type = MachO::N_SECT; |
413 | |
414 | // FIXME: Set STAB bits. |
415 | |
416 | if (Data.isPrivateExtern()) |
417 | Type |= MachO::N_PEXT; |
418 | |
419 | // Set external bit. |
420 | if (Data.isExternal() || (!IsAlias && Symbol->isUndefined())) |
421 | Type |= MachO::N_EXT; |
422 | |
423 | // Compute the symbol address. |
424 | if (IsAlias && Symbol->isUndefined()) |
425 | Address = AliaseeInfo->StringIndex; |
426 | else if (Symbol->isDefined()) |
427 | Address = getSymbolAddress(S: OrigSymbol, Asm); |
428 | else if (Symbol->isCommon()) { |
429 | // Common symbols are encoded with the size in the address |
430 | // field, and their alignment in the flags. |
431 | Address = Symbol->getCommonSize(); |
432 | } |
433 | |
434 | // struct nlist (12 bytes) |
435 | |
436 | W.write<uint32_t>(Val: MSD.StringIndex); |
437 | W.OS << char(Type); |
438 | W.OS << char(SectionIndex); |
439 | |
440 | // The Mach-O streamer uses the lowest 16-bits of the flags for the 'desc' |
441 | // value. |
442 | bool EncodeAsAltEntry = |
443 | IsAlias && cast<MCSymbolMachO>(Val: OrigSymbol).isAltEntry(); |
444 | W.write<uint16_t>(Val: cast<MCSymbolMachO>(Val: Symbol)->getEncodedFlags(EncodeAsAltEntry)); |
445 | if (is64Bit()) |
446 | W.write<uint64_t>(Val: Address); |
447 | else |
448 | W.write<uint32_t>(Val: Address); |
449 | } |
450 | |
451 | void MachObjectWriter::writeLinkeditLoadCommand(uint32_t Type, |
452 | uint32_t DataOffset, |
453 | uint32_t DataSize) { |
454 | uint64_t Start = W.OS.tell(); |
455 | (void) Start; |
456 | |
457 | W.write<uint32_t>(Val: Type); |
458 | W.write<uint32_t>(Val: sizeof(MachO::linkedit_data_command)); |
459 | W.write<uint32_t>(Val: DataOffset); |
460 | W.write<uint32_t>(Val: DataSize); |
461 | |
462 | assert(W.OS.tell() - Start == sizeof(MachO::linkedit_data_command)); |
463 | } |
464 | |
465 | static unsigned ComputeLinkerOptionsLoadCommandSize( |
466 | const std::vector<std::string> &Options, bool is64Bit) |
467 | { |
468 | unsigned Size = sizeof(MachO::linker_option_command); |
469 | for (const std::string &Option : Options) |
470 | Size += Option.size() + 1; |
471 | return alignTo(Value: Size, Align: is64Bit ? 8 : 4); |
472 | } |
473 | |
474 | void MachObjectWriter::writeLinkerOptionsLoadCommand( |
475 | const std::vector<std::string> &Options) |
476 | { |
477 | unsigned Size = ComputeLinkerOptionsLoadCommandSize(Options, is64Bit: is64Bit()); |
478 | uint64_t Start = W.OS.tell(); |
479 | (void) Start; |
480 | |
481 | W.write<uint32_t>(Val: MachO::LC_LINKER_OPTION); |
482 | W.write<uint32_t>(Val: Size); |
483 | W.write<uint32_t>(Val: Options.size()); |
484 | uint64_t BytesWritten = sizeof(MachO::linker_option_command); |
485 | for (const std::string &Option : Options) { |
486 | // Write each string, including the null byte. |
487 | W.OS << Option << '\0'; |
488 | BytesWritten += Option.size() + 1; |
489 | } |
490 | |
491 | // Pad to a multiple of the pointer size. |
492 | W.OS.write_zeros( |
493 | NumZeros: offsetToAlignment(Value: BytesWritten, Alignment: is64Bit() ? Align(8) : Align(4))); |
494 | |
495 | assert(W.OS.tell() - Start == Size); |
496 | } |
497 | |
498 | static bool isFixupTargetValid(const MCValue &Target) { |
499 | // Target is (LHS - RHS + cst). |
500 | // We don't support the form where LHS is null: -RHS + cst |
501 | if (!Target.getSymA() && Target.getSymB()) |
502 | return false; |
503 | return true; |
504 | } |
505 | |
506 | void MachObjectWriter::recordRelocation(MCAssembler &Asm, |
507 | const MCFragment *Fragment, |
508 | const MCFixup &Fixup, MCValue Target, |
509 | uint64_t &FixedValue) { |
510 | if (!isFixupTargetValid(Target)) { |
511 | Asm.getContext().reportError(L: Fixup.getLoc(), |
512 | Msg: "unsupported relocation expression" ); |
513 | return; |
514 | } |
515 | |
516 | TargetObjectWriter->recordRelocation(Writer: this, Asm, Fragment, Fixup, Target, |
517 | FixedValue); |
518 | } |
519 | |
520 | void MachObjectWriter::bindIndirectSymbols(MCAssembler &Asm) { |
521 | // This is the point where 'as' creates actual symbols for indirect symbols |
522 | // (in the following two passes). It would be easier for us to do this sooner |
523 | // when we see the attribute, but that makes getting the order in the symbol |
524 | // table much more complicated than it is worth. |
525 | // |
526 | // FIXME: Revisit this when the dust settles. |
527 | |
528 | // Report errors for use of .indirect_symbol not in a symbol pointer section |
529 | // or stub section. |
530 | for (IndirectSymbolData &ISD : IndirectSymbols) { |
531 | const MCSectionMachO &Section = cast<MCSectionMachO>(Val&: *ISD.Section); |
532 | |
533 | if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS && |
534 | Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS && |
535 | Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS && |
536 | Section.getType() != MachO::S_SYMBOL_STUBS) { |
537 | MCSymbol &Symbol = *ISD.Symbol; |
538 | report_fatal_error(reason: "indirect symbol '" + Symbol.getName() + |
539 | "' not in a symbol pointer or stub section" ); |
540 | } |
541 | } |
542 | |
543 | // Bind non-lazy symbol pointers first. |
544 | for (auto [IndirectIndex, ISD] : enumerate(First&: IndirectSymbols)) { |
545 | const auto &Section = cast<MCSectionMachO>(Val&: *ISD.Section); |
546 | |
547 | if (Section.getType() != MachO::S_NON_LAZY_SYMBOL_POINTERS && |
548 | Section.getType() != MachO::S_THREAD_LOCAL_VARIABLE_POINTERS) |
549 | continue; |
550 | |
551 | // Initialize the section indirect symbol base, if necessary. |
552 | IndirectSymBase.insert(KV: std::make_pair(x&: ISD.Section, y&: IndirectIndex)); |
553 | |
554 | Asm.registerSymbol(Symbol: *ISD.Symbol); |
555 | } |
556 | |
557 | // Then lazy symbol pointers and symbol stubs. |
558 | for (auto [IndirectIndex, ISD] : enumerate(First&: IndirectSymbols)) { |
559 | const auto &Section = cast<MCSectionMachO>(Val&: *ISD.Section); |
560 | |
561 | if (Section.getType() != MachO::S_LAZY_SYMBOL_POINTERS && |
562 | Section.getType() != MachO::S_SYMBOL_STUBS) |
563 | continue; |
564 | |
565 | // Initialize the section indirect symbol base, if necessary. |
566 | IndirectSymBase.insert(KV: std::make_pair(x&: ISD.Section, y&: IndirectIndex)); |
567 | |
568 | // Set the symbol type to undefined lazy, but only on construction. |
569 | // |
570 | // FIXME: Do not hardcode. |
571 | if (Asm.registerSymbol(Symbol: *ISD.Symbol)) |
572 | cast<MCSymbolMachO>(Val: ISD.Symbol)->setReferenceTypeUndefinedLazy(true); |
573 | } |
574 | } |
575 | |
576 | /// computeSymbolTable - Compute the symbol table data |
577 | void MachObjectWriter::computeSymbolTable( |
578 | MCAssembler &Asm, std::vector<MachSymbolData> &LocalSymbolData, |
579 | std::vector<MachSymbolData> &ExternalSymbolData, |
580 | std::vector<MachSymbolData> &UndefinedSymbolData) { |
581 | // Build section lookup table. |
582 | DenseMap<const MCSection*, uint8_t> SectionIndexMap; |
583 | unsigned Index = 1; |
584 | for (MCSection &Sec : Asm) |
585 | SectionIndexMap[&Sec] = Index++; |
586 | assert(Index <= 256 && "Too many sections!" ); |
587 | |
588 | // Build the string table. |
589 | for (const MCSymbol &Symbol : Asm.symbols()) { |
590 | if (!cast<MCSymbolMachO>(Val: Symbol).isSymbolLinkerVisible()) |
591 | continue; |
592 | |
593 | StringTable.add(S: Symbol.getName()); |
594 | } |
595 | StringTable.finalize(); |
596 | |
597 | // Build the symbol arrays but only for non-local symbols. |
598 | // |
599 | // The particular order that we collect and then sort the symbols is chosen to |
600 | // match 'as'. Even though it doesn't matter for correctness, this is |
601 | // important for letting us diff .o files. |
602 | for (const MCSymbol &Symbol : Asm.symbols()) { |
603 | // Ignore non-linker visible symbols. |
604 | if (!cast<MCSymbolMachO>(Val: Symbol).isSymbolLinkerVisible()) |
605 | continue; |
606 | |
607 | if (!Symbol.isExternal() && !Symbol.isUndefined()) |
608 | continue; |
609 | |
610 | MachSymbolData MSD; |
611 | MSD.Symbol = &Symbol; |
612 | MSD.StringIndex = StringTable.getOffset(S: Symbol.getName()); |
613 | |
614 | if (Symbol.isUndefined()) { |
615 | MSD.SectionIndex = 0; |
616 | UndefinedSymbolData.push_back(x: MSD); |
617 | } else if (Symbol.isAbsolute()) { |
618 | MSD.SectionIndex = 0; |
619 | ExternalSymbolData.push_back(x: MSD); |
620 | } else { |
621 | MSD.SectionIndex = SectionIndexMap.lookup(Val: &Symbol.getSection()); |
622 | assert(MSD.SectionIndex && "Invalid section index!" ); |
623 | ExternalSymbolData.push_back(x: MSD); |
624 | } |
625 | } |
626 | |
627 | // Now add the data for local symbols. |
628 | for (const MCSymbol &Symbol : Asm.symbols()) { |
629 | // Ignore non-linker visible symbols. |
630 | if (!cast<MCSymbolMachO>(Val: Symbol).isSymbolLinkerVisible()) |
631 | continue; |
632 | |
633 | if (Symbol.isExternal() || Symbol.isUndefined()) |
634 | continue; |
635 | |
636 | MachSymbolData MSD; |
637 | MSD.Symbol = &Symbol; |
638 | MSD.StringIndex = StringTable.getOffset(S: Symbol.getName()); |
639 | |
640 | if (Symbol.isAbsolute()) { |
641 | MSD.SectionIndex = 0; |
642 | LocalSymbolData.push_back(x: MSD); |
643 | } else { |
644 | MSD.SectionIndex = SectionIndexMap.lookup(Val: &Symbol.getSection()); |
645 | assert(MSD.SectionIndex && "Invalid section index!" ); |
646 | LocalSymbolData.push_back(x: MSD); |
647 | } |
648 | } |
649 | |
650 | // External and undefined symbols are required to be in lexicographic order. |
651 | llvm::sort(C&: ExternalSymbolData); |
652 | llvm::sort(C&: UndefinedSymbolData); |
653 | |
654 | // Set the symbol indices. |
655 | Index = 0; |
656 | for (auto *SymbolData : |
657 | {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData}) |
658 | for (MachSymbolData &Entry : *SymbolData) |
659 | Entry.Symbol->setIndex(Index++); |
660 | |
661 | for (const MCSection &Section : Asm) { |
662 | for (RelAndSymbol &Rel : Relocations[&Section]) { |
663 | if (!Rel.Sym) |
664 | continue; |
665 | |
666 | // Set the Index and the IsExtern bit. |
667 | unsigned Index = Rel.Sym->getIndex(); |
668 | assert(isInt<24>(Index)); |
669 | if (W.Endian == llvm::endianness::little) |
670 | Rel.MRE.r_word1 = (Rel.MRE.r_word1 & (~0U << 24)) | Index | (1 << 27); |
671 | else |
672 | Rel.MRE.r_word1 = (Rel.MRE.r_word1 & 0xff) | Index << 8 | (1 << 4); |
673 | } |
674 | } |
675 | } |
676 | |
677 | void MachObjectWriter::computeSectionAddresses(const MCAssembler &Asm) { |
678 | // Assign layout order indices to sections. |
679 | unsigned i = 0; |
680 | // Compute the section layout order. Virtual sections must go last. |
681 | for (MCSection &Sec : Asm) { |
682 | if (!Sec.isVirtualSection()) { |
683 | SectionOrder.push_back(Elt: &Sec); |
684 | cast<MCSectionMachO>(Val&: Sec).setLayoutOrder(i++); |
685 | } |
686 | } |
687 | for (MCSection &Sec : Asm) { |
688 | if (Sec.isVirtualSection()) { |
689 | SectionOrder.push_back(Elt: &Sec); |
690 | cast<MCSectionMachO>(Val&: Sec).setLayoutOrder(i++); |
691 | } |
692 | } |
693 | |
694 | uint64_t StartAddress = 0; |
695 | for (const MCSection *Sec : SectionOrder) { |
696 | StartAddress = alignTo(Size: StartAddress, A: Sec->getAlign()); |
697 | SectionAddress[Sec] = StartAddress; |
698 | StartAddress += Asm.getSectionAddressSize(Sec: *Sec); |
699 | |
700 | // Explicitly pad the section to match the alignment requirements of the |
701 | // following one. This is for 'gas' compatibility, it shouldn't |
702 | /// strictly be necessary. |
703 | StartAddress += getPaddingSize(Asm, Sec); |
704 | } |
705 | } |
706 | |
707 | void MachObjectWriter::executePostLayoutBinding(MCAssembler &Asm) { |
708 | computeSectionAddresses(Asm); |
709 | |
710 | // Create symbol data for any indirect symbols. |
711 | bindIndirectSymbols(Asm); |
712 | } |
713 | |
714 | bool MachObjectWriter::isSymbolRefDifferenceFullyResolvedImpl( |
715 | const MCAssembler &Asm, const MCSymbol &SymA, const MCFragment &FB, |
716 | bool InSet, bool IsPCRel) const { |
717 | if (InSet) |
718 | return true; |
719 | |
720 | // The effective address is |
721 | // addr(atom(A)) + offset(A) |
722 | // - addr(atom(B)) - offset(B) |
723 | // and the offsets are not relocatable, so the fixup is fully resolved when |
724 | // addr(atom(A)) - addr(atom(B)) == 0. |
725 | const MCSymbol &SA = findAliasedSymbol(Sym: SymA); |
726 | const MCSection &SecA = SA.getSection(); |
727 | const MCSection &SecB = *FB.getParent(); |
728 | |
729 | if (IsPCRel) { |
730 | // The simple (Darwin, except on x86_64) way of dealing with this was to |
731 | // assume that any reference to a temporary symbol *must* be a temporary |
732 | // symbol in the same atom, unless the sections differ. Therefore, any PCrel |
733 | // relocation to a temporary symbol (in the same section) is fully |
734 | // resolved. This also works in conjunction with absolutized .set, which |
735 | // requires the compiler to use .set to absolutize the differences between |
736 | // symbols which the compiler knows to be assembly time constants, so we |
737 | // don't need to worry about considering symbol differences fully resolved. |
738 | // |
739 | // If the file isn't using sub-sections-via-symbols, we can make the |
740 | // same assumptions about any symbol that we normally make about |
741 | // assembler locals. |
742 | |
743 | bool hasReliableSymbolDifference = isX86_64(); |
744 | if (!hasReliableSymbolDifference) { |
745 | if (!SA.isInSection() || &SecA != &SecB || |
746 | (!SA.isTemporary() && FB.getAtom() != SA.getFragment()->getAtom() && |
747 | SubsectionsViaSymbols)) |
748 | return false; |
749 | return true; |
750 | } |
751 | } |
752 | |
753 | // If they are not in the same section, we can't compute the diff. |
754 | if (&SecA != &SecB) |
755 | return false; |
756 | |
757 | // If the atoms are the same, they are guaranteed to have the same address. |
758 | return SA.getFragment()->getAtom() == FB.getAtom(); |
759 | } |
760 | |
761 | static MachO::LoadCommandType getLCFromMCVM(MCVersionMinType Type) { |
762 | switch (Type) { |
763 | case MCVM_OSXVersionMin: return MachO::LC_VERSION_MIN_MACOSX; |
764 | case MCVM_IOSVersionMin: return MachO::LC_VERSION_MIN_IPHONEOS; |
765 | case MCVM_TvOSVersionMin: return MachO::LC_VERSION_MIN_TVOS; |
766 | case MCVM_WatchOSVersionMin: return MachO::LC_VERSION_MIN_WATCHOS; |
767 | } |
768 | llvm_unreachable("Invalid mc version min type" ); |
769 | } |
770 | |
771 | void MachObjectWriter::populateAddrSigSection(MCAssembler &Asm) { |
772 | MCSection *AddrSigSection = |
773 | Asm.getContext().getObjectFileInfo()->getAddrSigSection(); |
774 | unsigned Log2Size = is64Bit() ? 3 : 2; |
775 | for (const MCSymbol *S : getAddrsigSyms()) { |
776 | if (!S->isRegistered()) |
777 | continue; |
778 | MachO::any_relocation_info MRE; |
779 | MRE.r_word0 = 0; |
780 | MRE.r_word1 = (Log2Size << 25) | (MachO::GENERIC_RELOC_VANILLA << 28); |
781 | addRelocation(RelSymbol: S, Sec: AddrSigSection, MRE); |
782 | } |
783 | } |
784 | |
785 | uint64_t MachObjectWriter::writeObject(MCAssembler &Asm) { |
786 | uint64_t StartOffset = W.OS.tell(); |
787 | auto NumBytesWritten = [&] { return W.OS.tell() - StartOffset; }; |
788 | |
789 | populateAddrSigSection(Asm); |
790 | |
791 | // Compute symbol table information and bind symbol indices. |
792 | computeSymbolTable(Asm, LocalSymbolData, ExternalSymbolData, |
793 | UndefinedSymbolData); |
794 | |
795 | if (!CGProfile.empty()) { |
796 | MCSection *CGProfileSection = Asm.getContext().getMachOSection( |
797 | Segment: "__LLVM" , Section: "__cg_profile" , TypeAndAttributes: 0, K: SectionKind::getMetadata()); |
798 | auto &Frag = cast<MCDataFragment>(Val&: *CGProfileSection->begin()); |
799 | Frag.getContents().clear(); |
800 | raw_svector_ostream OS(Frag.getContents()); |
801 | for (const MCObjectWriter::CGProfileEntry &CGPE : CGProfile) { |
802 | uint32_t FromIndex = CGPE.From->getSymbol().getIndex(); |
803 | uint32_t ToIndex = CGPE.To->getSymbol().getIndex(); |
804 | support::endian::write(os&: OS, value: FromIndex, endian: W.Endian); |
805 | support::endian::write(os&: OS, value: ToIndex, endian: W.Endian); |
806 | support::endian::write(os&: OS, value: CGPE.Count, endian: W.Endian); |
807 | } |
808 | } |
809 | |
810 | unsigned NumSections = Asm.end() - Asm.begin(); |
811 | |
812 | // The section data starts after the header, the segment load command (and |
813 | // section headers) and the symbol table. |
814 | unsigned NumLoadCommands = 1; |
815 | uint64_t LoadCommandsSize = is64Bit() ? |
816 | sizeof(MachO::segment_command_64) + NumSections * sizeof(MachO::section_64): |
817 | sizeof(MachO::segment_command) + NumSections * sizeof(MachO::section); |
818 | |
819 | // Add the deployment target version info load command size, if used. |
820 | if (VersionInfo.Major != 0) { |
821 | ++NumLoadCommands; |
822 | if (VersionInfo.EmitBuildVersion) |
823 | LoadCommandsSize += sizeof(MachO::build_version_command); |
824 | else |
825 | LoadCommandsSize += sizeof(MachO::version_min_command); |
826 | } |
827 | |
828 | // Add the target variant version info load command size, if used. |
829 | if (TargetVariantVersionInfo.Major != 0) { |
830 | ++NumLoadCommands; |
831 | assert(TargetVariantVersionInfo.EmitBuildVersion && |
832 | "target variant should use build version" ); |
833 | LoadCommandsSize += sizeof(MachO::build_version_command); |
834 | } |
835 | |
836 | // Add the data-in-code load command size, if used. |
837 | unsigned NumDataRegions = DataRegions.size(); |
838 | if (NumDataRegions) { |
839 | ++NumLoadCommands; |
840 | LoadCommandsSize += sizeof(MachO::linkedit_data_command); |
841 | } |
842 | |
843 | // Add the loh load command size, if used. |
844 | uint64_t LOHRawSize = LOHContainer.getEmitSize(Asm, ObjWriter: *this); |
845 | uint64_t LOHSize = alignTo(Value: LOHRawSize, Align: is64Bit() ? 8 : 4); |
846 | if (LOHSize) { |
847 | ++NumLoadCommands; |
848 | LoadCommandsSize += sizeof(MachO::linkedit_data_command); |
849 | } |
850 | |
851 | // Add the symbol table load command sizes, if used. |
852 | unsigned NumSymbols = LocalSymbolData.size() + ExternalSymbolData.size() + |
853 | UndefinedSymbolData.size(); |
854 | if (NumSymbols) { |
855 | NumLoadCommands += 2; |
856 | LoadCommandsSize += (sizeof(MachO::symtab_command) + |
857 | sizeof(MachO::dysymtab_command)); |
858 | } |
859 | |
860 | // Add the linker option load commands sizes. |
861 | for (const auto &Option : LinkerOptions) { |
862 | ++NumLoadCommands; |
863 | LoadCommandsSize += ComputeLinkerOptionsLoadCommandSize(Options: Option, is64Bit: is64Bit()); |
864 | } |
865 | |
866 | // Compute the total size of the section data, as well as its file size and vm |
867 | // size. |
868 | uint64_t SectionDataStart = (is64Bit() ? sizeof(MachO::mach_header_64) : |
869 | sizeof(MachO::mach_header)) + LoadCommandsSize; |
870 | uint64_t SectionDataSize = 0; |
871 | uint64_t SectionDataFileSize = 0; |
872 | uint64_t VMSize = 0; |
873 | for (const MCSection &Sec : Asm) { |
874 | uint64_t Address = getSectionAddress(Sec: &Sec); |
875 | uint64_t Size = Asm.getSectionAddressSize(Sec); |
876 | uint64_t FileSize = Asm.getSectionFileSize(Sec); |
877 | FileSize += getPaddingSize(Asm, Sec: &Sec); |
878 | |
879 | VMSize = std::max(a: VMSize, b: Address + Size); |
880 | |
881 | if (Sec.isVirtualSection()) |
882 | continue; |
883 | |
884 | SectionDataSize = std::max(a: SectionDataSize, b: Address + Size); |
885 | SectionDataFileSize = std::max(a: SectionDataFileSize, b: Address + FileSize); |
886 | } |
887 | |
888 | // The section data is padded to pointer size bytes. |
889 | // |
890 | // FIXME: Is this machine dependent? |
891 | unsigned SectionDataPadding = |
892 | offsetToAlignment(Value: SectionDataFileSize, Alignment: is64Bit() ? Align(8) : Align(4)); |
893 | SectionDataFileSize += SectionDataPadding; |
894 | |
895 | // Write the prolog, starting with the header and load command... |
896 | writeHeader(Type: MachO::MH_OBJECT, NumLoadCommands, LoadCommandsSize, |
897 | SubsectionsViaSymbols); |
898 | uint32_t Prot = |
899 | MachO::VM_PROT_READ | MachO::VM_PROT_WRITE | MachO::VM_PROT_EXECUTE; |
900 | writeSegmentLoadCommand(Name: "" , NumSections, VMAddr: 0, VMSize, SectionDataStartOffset: SectionDataStart, |
901 | SectionDataSize, MaxProt: Prot, InitProt: Prot); |
902 | |
903 | // ... and then the section headers. |
904 | uint64_t RelocTableEnd = SectionDataStart + SectionDataFileSize; |
905 | for (const MCSection &Section : Asm) { |
906 | const auto &Sec = cast<MCSectionMachO>(Val: Section); |
907 | std::vector<RelAndSymbol> &Relocs = Relocations[&Sec]; |
908 | unsigned NumRelocs = Relocs.size(); |
909 | uint64_t SectionStart = SectionDataStart + getSectionAddress(Sec: &Sec); |
910 | unsigned Flags = Sec.getTypeAndAttributes(); |
911 | if (Sec.hasInstructions()) |
912 | Flags |= MachO::S_ATTR_SOME_INSTRUCTIONS; |
913 | if (!cast<MCSectionMachO>(Val: Sec).isVirtualSection() && |
914 | !isUInt<32>(x: SectionStart)) { |
915 | Asm.getContext().reportError( |
916 | L: SMLoc(), Msg: "cannot encode offset of section; object file too large" ); |
917 | return NumBytesWritten(); |
918 | } |
919 | if (NumRelocs && !isUInt<32>(x: RelocTableEnd)) { |
920 | Asm.getContext().reportError( |
921 | L: SMLoc(), |
922 | Msg: "cannot encode offset of relocations; object file too large" ); |
923 | return NumBytesWritten(); |
924 | } |
925 | writeSection(Asm, Sec, VMAddr: getSectionAddress(Sec: &Sec), FileOffset: SectionStart, Flags, |
926 | RelocationsStart: RelocTableEnd, NumRelocations: NumRelocs); |
927 | RelocTableEnd += NumRelocs * sizeof(MachO::any_relocation_info); |
928 | } |
929 | |
930 | // Write out the deployment target information, if it's available. |
931 | auto EmitDeploymentTargetVersion = |
932 | [&](const VersionInfoType &VersionInfo) { |
933 | auto EncodeVersion = [](VersionTuple V) -> uint32_t { |
934 | assert(!V.empty() && "empty version" ); |
935 | unsigned Update = V.getSubminor().value_or(u: 0); |
936 | unsigned Minor = V.getMinor().value_or(u: 0); |
937 | assert(Update < 256 && "unencodable update target version" ); |
938 | assert(Minor < 256 && "unencodable minor target version" ); |
939 | assert(V.getMajor() < 65536 && "unencodable major target version" ); |
940 | return Update | (Minor << 8) | (V.getMajor() << 16); |
941 | }; |
942 | uint32_t EncodedVersion = EncodeVersion(VersionTuple( |
943 | VersionInfo.Major, VersionInfo.Minor, VersionInfo.Update)); |
944 | uint32_t SDKVersion = !VersionInfo.SDKVersion.empty() |
945 | ? EncodeVersion(VersionInfo.SDKVersion) |
946 | : 0; |
947 | if (VersionInfo.EmitBuildVersion) { |
948 | // FIXME: Currently empty tools. Add clang version in the future. |
949 | W.write<uint32_t>(Val: MachO::LC_BUILD_VERSION); |
950 | W.write<uint32_t>(Val: sizeof(MachO::build_version_command)); |
951 | W.write<uint32_t>(Val: VersionInfo.TypeOrPlatform.Platform); |
952 | W.write<uint32_t>(Val: EncodedVersion); |
953 | W.write<uint32_t>(Val: SDKVersion); |
954 | W.write<uint32_t>(Val: 0); // Empty tools list. |
955 | } else { |
956 | MachO::LoadCommandType LCType = |
957 | getLCFromMCVM(Type: VersionInfo.TypeOrPlatform.Type); |
958 | W.write<uint32_t>(Val: LCType); |
959 | W.write<uint32_t>(Val: sizeof(MachO::version_min_command)); |
960 | W.write<uint32_t>(Val: EncodedVersion); |
961 | W.write<uint32_t>(Val: SDKVersion); |
962 | } |
963 | }; |
964 | if (VersionInfo.Major != 0) |
965 | EmitDeploymentTargetVersion(VersionInfo); |
966 | if (TargetVariantVersionInfo.Major != 0) |
967 | EmitDeploymentTargetVersion(TargetVariantVersionInfo); |
968 | |
969 | // Write the data-in-code load command, if used. |
970 | uint64_t DataInCodeTableEnd = RelocTableEnd + NumDataRegions * 8; |
971 | if (NumDataRegions) { |
972 | uint64_t DataRegionsOffset = RelocTableEnd; |
973 | uint64_t DataRegionsSize = NumDataRegions * 8; |
974 | writeLinkeditLoadCommand(Type: MachO::LC_DATA_IN_CODE, DataOffset: DataRegionsOffset, |
975 | DataSize: DataRegionsSize); |
976 | } |
977 | |
978 | // Write the loh load command, if used. |
979 | uint64_t LOHTableEnd = DataInCodeTableEnd + LOHSize; |
980 | if (LOHSize) |
981 | writeLinkeditLoadCommand(Type: MachO::LC_LINKER_OPTIMIZATION_HINT, |
982 | DataOffset: DataInCodeTableEnd, DataSize: LOHSize); |
983 | |
984 | // Write the symbol table load command, if used. |
985 | if (NumSymbols) { |
986 | unsigned FirstLocalSymbol = 0; |
987 | unsigned NumLocalSymbols = LocalSymbolData.size(); |
988 | unsigned FirstExternalSymbol = FirstLocalSymbol + NumLocalSymbols; |
989 | unsigned NumExternalSymbols = ExternalSymbolData.size(); |
990 | unsigned FirstUndefinedSymbol = FirstExternalSymbol + NumExternalSymbols; |
991 | unsigned NumUndefinedSymbols = UndefinedSymbolData.size(); |
992 | unsigned NumIndirectSymbols = IndirectSymbols.size(); |
993 | unsigned NumSymTabSymbols = |
994 | NumLocalSymbols + NumExternalSymbols + NumUndefinedSymbols; |
995 | uint64_t IndirectSymbolSize = NumIndirectSymbols * 4; |
996 | uint64_t IndirectSymbolOffset = 0; |
997 | |
998 | // If used, the indirect symbols are written after the section data. |
999 | if (NumIndirectSymbols) |
1000 | IndirectSymbolOffset = LOHTableEnd; |
1001 | |
1002 | // The symbol table is written after the indirect symbol data. |
1003 | uint64_t SymbolTableOffset = LOHTableEnd + IndirectSymbolSize; |
1004 | |
1005 | // The string table is written after symbol table. |
1006 | uint64_t StringTableOffset = |
1007 | SymbolTableOffset + NumSymTabSymbols * (is64Bit() ? |
1008 | sizeof(MachO::nlist_64) : |
1009 | sizeof(MachO::nlist)); |
1010 | writeSymtabLoadCommand(SymbolOffset: SymbolTableOffset, NumSymbols: NumSymTabSymbols, |
1011 | StringTableOffset, StringTableSize: StringTable.getSize()); |
1012 | |
1013 | writeDysymtabLoadCommand(FirstLocalSymbol, NumLocalSymbols, |
1014 | FirstExternalSymbol, NumExternalSymbols, |
1015 | FirstUndefinedSymbol, NumUndefinedSymbols, |
1016 | IndirectSymbolOffset, NumIndirectSymbols); |
1017 | } |
1018 | |
1019 | // Write the linker options load commands. |
1020 | for (const auto &Option : LinkerOptions) |
1021 | writeLinkerOptionsLoadCommand(Options: Option); |
1022 | |
1023 | // Write the actual section data. |
1024 | for (const MCSection &Sec : Asm) { |
1025 | Asm.writeSectionData(OS&: W.OS, Section: &Sec); |
1026 | |
1027 | uint64_t Pad = getPaddingSize(Asm, Sec: &Sec); |
1028 | W.OS.write_zeros(NumZeros: Pad); |
1029 | } |
1030 | |
1031 | // Write the extra padding. |
1032 | W.OS.write_zeros(NumZeros: SectionDataPadding); |
1033 | |
1034 | // Write the relocation entries. |
1035 | for (const MCSection &Sec : Asm) { |
1036 | // Write the section relocation entries, in reverse order to match 'as' |
1037 | // (approximately, the exact algorithm is more complicated than this). |
1038 | std::vector<RelAndSymbol> &Relocs = Relocations[&Sec]; |
1039 | for (const RelAndSymbol &Rel : llvm::reverse(C&: Relocs)) { |
1040 | W.write<uint32_t>(Val: Rel.MRE.r_word0); |
1041 | W.write<uint32_t>(Val: Rel.MRE.r_word1); |
1042 | } |
1043 | } |
1044 | |
1045 | // Write out the data-in-code region payload, if there is one. |
1046 | for (DataRegionData Data : DataRegions) { |
1047 | uint64_t Start = getSymbolAddress(S: *Data.Start, Asm); |
1048 | uint64_t End; |
1049 | if (Data.End) |
1050 | End = getSymbolAddress(S: *Data.End, Asm); |
1051 | else |
1052 | report_fatal_error(reason: "Data region not terminated" ); |
1053 | |
1054 | LLVM_DEBUG(dbgs() << "data in code region-- kind: " << Data.Kind |
1055 | << " start: " << Start << "(" << Data.Start->getName() |
1056 | << ")" << " end: " << End << "(" << Data.End->getName() |
1057 | << ")" << " size: " << End - Start << "\n" ); |
1058 | W.write<uint32_t>(Val: Start); |
1059 | W.write<uint16_t>(Val: End - Start); |
1060 | W.write<uint16_t>(Val: Data.Kind); |
1061 | } |
1062 | |
1063 | // Write out the loh commands, if there is one. |
1064 | if (LOHSize) { |
1065 | #ifndef NDEBUG |
1066 | unsigned Start = W.OS.tell(); |
1067 | #endif |
1068 | LOHContainer.emit(Asm, ObjWriter&: *this); |
1069 | // Pad to a multiple of the pointer size. |
1070 | W.OS.write_zeros( |
1071 | NumZeros: offsetToAlignment(Value: LOHRawSize, Alignment: is64Bit() ? Align(8) : Align(4))); |
1072 | assert(W.OS.tell() - Start == LOHSize); |
1073 | } |
1074 | |
1075 | // Write the symbol table data, if used. |
1076 | if (NumSymbols) { |
1077 | // Write the indirect symbol entries. |
1078 | for (auto &ISD : IndirectSymbols) { |
1079 | // Indirect symbols in the non-lazy symbol pointer section have some |
1080 | // special handling. |
1081 | const MCSectionMachO &Section = |
1082 | static_cast<const MCSectionMachO &>(*ISD.Section); |
1083 | if (Section.getType() == MachO::S_NON_LAZY_SYMBOL_POINTERS) { |
1084 | // If this symbol is defined and internal, mark it as such. |
1085 | if (ISD.Symbol->isDefined() && !ISD.Symbol->isExternal()) { |
1086 | uint32_t Flags = MachO::INDIRECT_SYMBOL_LOCAL; |
1087 | if (ISD.Symbol->isAbsolute()) |
1088 | Flags |= MachO::INDIRECT_SYMBOL_ABS; |
1089 | W.write<uint32_t>(Val: Flags); |
1090 | continue; |
1091 | } |
1092 | } |
1093 | |
1094 | W.write<uint32_t>(Val: ISD.Symbol->getIndex()); |
1095 | } |
1096 | |
1097 | // FIXME: Check that offsets match computed ones. |
1098 | |
1099 | // Write the symbol table entries. |
1100 | for (auto *SymbolData : |
1101 | {&LocalSymbolData, &ExternalSymbolData, &UndefinedSymbolData}) |
1102 | for (MachSymbolData &Entry : *SymbolData) |
1103 | writeNlist(MSD&: Entry, Asm); |
1104 | |
1105 | // Write the string table. |
1106 | StringTable.write(OS&: W.OS); |
1107 | } |
1108 | |
1109 | return NumBytesWritten(); |
1110 | } |
1111 | |
1112 | std::unique_ptr<MCObjectWriter> |
1113 | llvm::createMachObjectWriter(std::unique_ptr<MCMachObjectTargetWriter> MOTW, |
1114 | raw_pwrite_stream &OS, bool IsLittleEndian) { |
1115 | return std::make_unique<MachObjectWriter>(args: std::move(MOTW), args&: OS, |
1116 | args&: IsLittleEndian); |
1117 | } |
1118 | |