1//===- InputSection.cpp ---------------------------------------------------===//
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 "InputSection.h"
10#include "ConcatOutputSection.h"
11#include "Config.h"
12#include "InputFiles.h"
13#include "OutputSegment.h"
14#include "Sections.h"
15#include "Symbols.h"
16#include "SyntheticSections.h"
17#include "Target.h"
18#include "Writer.h"
19
20#include "lld/Common/ErrorHandler.h"
21#include "lld/Common/Memory.h"
22#include "llvm/Support/xxhash.h"
23
24using namespace llvm;
25using namespace llvm::MachO;
26using namespace llvm::support;
27using namespace lld;
28using namespace lld::macho;
29
30// Verify ConcatInputSection's size on 64-bit builds. The size of std::vector
31// can differ based on STL debug levels (e.g. iterator debugging on MSVC's STL),
32// so account for that.
33static_assert(sizeof(void *) != 8 ||
34 sizeof(ConcatInputSection) == sizeof(std::vector<Reloc>) + 88,
35 "Try to minimize ConcatInputSection's size, we create many "
36 "instances of it");
37
38std::vector<ConcatInputSection *> macho::inputSections;
39int macho::inputSectionsOrder = 0;
40
41// Call this function to add a new InputSection and have it routed to the
42// appropriate container. Depending on its type and current config, it will
43// either be added to 'inputSections' vector or to a synthetic section.
44void lld::macho::addInputSection(InputSection *inputSection) {
45 if (auto *isec = dyn_cast<ConcatInputSection>(Val: inputSection)) {
46 if (isec->isCoalescedWeak())
47 return;
48 if (config->emitRelativeMethodLists &&
49 ObjCMethListSection::isMethodList(isec)) {
50 if (in.objcMethList->inputOrder == UnspecifiedInputOrder)
51 in.objcMethList->inputOrder = inputSectionsOrder++;
52 in.objcMethList->addInput(isec);
53 isec->parent = in.objcMethList;
54 return;
55 }
56 if (config->emitInitOffsets &&
57 sectionType(flags: isec->getFlags()) == S_MOD_INIT_FUNC_POINTERS) {
58 in.initOffsets->addInput(isec);
59 return;
60 }
61 isec->outSecOff = inputSectionsOrder++;
62 auto *osec = ConcatOutputSection::getOrCreateForInput(isec);
63 isec->parent = osec;
64 inputSections.push_back(x: isec);
65 } else if (auto *isec = dyn_cast<CStringInputSection>(Val: inputSection)) {
66 if (isec->getName() == section_names::objcMethname) {
67 if (in.objcMethnameSection->inputOrder == UnspecifiedInputOrder)
68 in.objcMethnameSection->inputOrder = inputSectionsOrder++;
69 in.objcMethnameSection->addInput(isec);
70 } else {
71 if (in.cStringSection->inputOrder == UnspecifiedInputOrder)
72 in.cStringSection->inputOrder = inputSectionsOrder++;
73 in.cStringSection->addInput(isec);
74 }
75 } else if (auto *isec = dyn_cast<WordLiteralInputSection>(Val: inputSection)) {
76 if (in.wordLiteralSection->inputOrder == UnspecifiedInputOrder)
77 in.wordLiteralSection->inputOrder = inputSectionsOrder++;
78 in.wordLiteralSection->addInput(isec);
79 } else {
80 llvm_unreachable("unexpected input section kind");
81 }
82
83 assert(inputSectionsOrder <= UnspecifiedInputOrder);
84}
85
86uint64_t InputSection::getFileSize() const {
87 return isZeroFill(flags: getFlags()) ? 0 : getSize();
88}
89
90uint64_t InputSection::getVA(uint64_t off) const {
91 return parent->addr + getOffset(off);
92}
93
94static uint64_t resolveSymbolVA(const Symbol *sym, uint8_t type) {
95 const RelocAttrs &relocAttrs = target->getRelocAttrs(type);
96 if (relocAttrs.hasAttr(b: RelocAttrBits::BRANCH))
97 return sym->resolveBranchVA();
98 if (relocAttrs.hasAttr(b: RelocAttrBits::GOT))
99 return sym->resolveGotVA();
100 if (relocAttrs.hasAttr(b: RelocAttrBits::TLV))
101 return sym->resolveTlvVA();
102 return sym->getVA();
103}
104
105const Defined *InputSection::getContainingSymbol(uint64_t off) const {
106 auto *nextSym = llvm::upper_bound(
107 Range: symbols, Value&: off, C: [](uint64_t a, const Defined *b) { return a < b->value; });
108 if (nextSym == symbols.begin())
109 return nullptr;
110 return *std::prev(x: nextSym);
111}
112
113std::string InputSection::getLocation(uint64_t off) const {
114 // First, try to find a symbol that's near the offset. Use it as a reference
115 // point.
116 if (auto *sym = getContainingSymbol(off))
117 return (toString(file: getFile()) + ":(symbol " + toString(*sym) + "+0x" +
118 Twine::utohexstr(Val: off - sym->value) + ")")
119 .str();
120
121 // If that fails, use the section itself as a reference point.
122 for (const Subsection &subsec : section.subsections) {
123 if (subsec.isec == this) {
124 off += subsec.offset;
125 break;
126 }
127 }
128
129 return (toString(file: getFile()) + ":(" + getName() + "+0x" +
130 Twine::utohexstr(Val: off) + ")")
131 .str();
132}
133
134std::string InputSection::getSourceLocation(uint64_t off) const {
135 auto *obj = dyn_cast_or_null<ObjFile>(Val: getFile());
136 if (!obj)
137 return {};
138
139 DWARFCache *dwarf = obj->getDwarf();
140 if (!dwarf)
141 return std::string();
142
143 for (const Subsection &subsec : section.subsections) {
144 if (subsec.isec == this) {
145 off += subsec.offset;
146 break;
147 }
148 }
149
150 auto createMsg = [&](StringRef path, unsigned line) {
151 std::string filename = sys::path::filename(path).str();
152 std::string lineStr = (":" + Twine(line)).str();
153 if (filename == path)
154 return filename + lineStr;
155 return (filename + lineStr + " (" + path + lineStr + ")").str();
156 };
157
158 // First, look up a function for a given offset.
159 if (std::optional<DILineInfo> li = dwarf->getDILineInfo(
160 offset: section.addr + off, sectionIndex: object::SectionedAddress::UndefSection))
161 return createMsg(li->FileName, li->Line);
162
163 // If it failed, look up again as a variable.
164 if (const Defined *sym = getContainingSymbol(off)) {
165 // Symbols are generally prefixed with an underscore, which is not included
166 // in the debug information.
167 StringRef symName = sym->getName();
168 symName.consume_front(Prefix: "_");
169
170 if (std::optional<std::pair<std::string, unsigned>> fileLine =
171 dwarf->getVariableLoc(name: symName))
172 return createMsg(fileLine->first, fileLine->second);
173 }
174
175 // Try to get the source file's name from the DWARF information.
176 if (obj->compileUnit)
177 return obj->sourceFile();
178
179 return {};
180}
181
182const Reloc *InputSection::getRelocAt(uint32_t off) const {
183 auto it = llvm::find_if(
184 Range: relocs, P: [=](const macho::Reloc &r) { return r.offset == off; });
185 if (it == relocs.end())
186 return nullptr;
187 return &*it;
188}
189
190void ConcatInputSection::foldIdentical(ConcatInputSection *copy,
191 Symbol::ICFFoldKind foldKind) {
192 align = std::max(a: align, b: copy->align);
193 copy->live = false;
194 copy->wasCoalesced = true;
195 copy->replacement = this;
196 for (auto &copySym : copy->symbols)
197 copySym->identicalCodeFoldingKind = foldKind;
198
199 symbols.insert(I: symbols.end(), From: copy->symbols.begin(), To: copy->symbols.end());
200 copy->symbols.clear();
201
202 // Remove duplicate compact unwind info for symbols at the same address.
203 if (symbols.empty())
204 return;
205 for (auto it = symbols.begin() + 1; it != symbols.end(); ++it) {
206 assert((*it)->value == 0);
207 (*it)->originalUnwindEntry = nullptr;
208 }
209}
210
211void ConcatInputSection::writeTo(uint8_t *buf) {
212 assert(!shouldOmitFromOutput());
213
214 if (getFileSize() == 0)
215 return;
216
217 memcpy(dest: buf, src: data.data(), n: data.size());
218
219 for (size_t i = 0; i < relocs.size(); i++) {
220 const Reloc &r = relocs[i];
221 uint8_t *loc = buf + r.offset;
222 uint64_t referentVA = 0;
223
224 const bool needsFixup = config->emitChainedFixups &&
225 target->hasAttr(type: r.type, bit: RelocAttrBits::UNSIGNED);
226 if (target->hasAttr(type: r.type, bit: RelocAttrBits::SUBTRAHEND)) {
227 const Symbol *fromSym = cast<Symbol *>(Val: r.referent);
228 const Reloc &minuend = relocs[++i];
229 uint64_t minuendVA;
230 if (const Symbol *toSym = minuend.referent.dyn_cast<Symbol *>())
231 minuendVA = toSym->getVA() + minuend.addend;
232 else {
233 auto *referentIsec = cast<InputSection *>(Val: minuend.referent);
234 assert(!::shouldOmitFromOutput(referentIsec));
235 minuendVA = referentIsec->getVA(off: minuend.addend);
236 }
237 referentVA = minuendVA - fromSym->getVA();
238 } else if (auto *referentSym = r.referent.dyn_cast<Symbol *>()) {
239 if (target->hasAttr(type: r.type, bit: RelocAttrBits::LOAD) &&
240 !referentSym->isInGot())
241 target->relaxGotLoad(loc, type: r.type);
242 // For dtrace symbols, do not handle them as normal undefined symbols
243 if (referentSym->getName().starts_with(Prefix: "___dtrace_")) {
244 // Change dtrace call site to pre-defined instructions
245 target->handleDtraceReloc(sym: referentSym, r, loc);
246 continue;
247 }
248 referentVA = resolveSymbolVA(sym: referentSym, type: r.type) + r.addend;
249
250 if (isThreadLocalVariables(flags: getFlags()) && isa<Defined>(Val: referentSym)) {
251 // References from thread-local variable sections are treated as offsets
252 // relative to the start of the thread-local data memory area, which
253 // is initialized via copying all the TLV data sections (which are all
254 // contiguous).
255 referentVA -= firstTLVDataSection->addr;
256 } else if (needsFixup) {
257 writeChainedFixup(buf: loc, sym: referentSym, addend: r.addend);
258 continue;
259 }
260 } else if (auto *referentIsec = r.referent.dyn_cast<InputSection *>()) {
261 assert(!::shouldOmitFromOutput(referentIsec));
262 referentVA = referentIsec->getVA(off: r.addend);
263
264 if (needsFixup) {
265 writeChainedRebase(buf: loc, targetVA: referentVA);
266 continue;
267 }
268 }
269 target->relocateOne(loc, r, va: referentVA, relocVA: getVA() + r.offset);
270 }
271}
272
273ConcatInputSection *macho::makeSyntheticInputSection(StringRef segName,
274 StringRef sectName,
275 uint32_t flags,
276 ArrayRef<uint8_t> data,
277 uint32_t align) {
278 Section &section =
279 *make<Section>(/*file=*/args: nullptr, args&: segName, args&: sectName, args&: flags, /*addr=*/args: 0);
280 auto isec = make<ConcatInputSection>(args&: section, args&: data, args&: align);
281 // Since this is an explicitly created 'fake' input section,
282 // it should not be dead stripped.
283 isec->live = true;
284 section.subsections.push_back(x: {.offset: 0, .isec: isec});
285 return isec;
286}
287
288void CStringInputSection::splitIntoPieces() {
289 size_t off = 0;
290 StringRef s = toStringRef(Input: data);
291 while (!s.empty()) {
292 size_t end = s.find(C: 0);
293 if (end == StringRef::npos)
294 fatal(msg: getLocation(off) + ": string is not null terminated");
295 uint32_t hash = deduplicateLiterals ? xxh3_64bits(data: s.take_front(N: end)) : 0;
296 pieces.emplace_back(args&: off, args&: hash);
297 size_t size = end + 1; // include null terminator
298 s = s.substr(Start: size);
299 off += size;
300 }
301}
302
303StringPiece &CStringInputSection::getStringPiece(uint64_t off) {
304 if (off >= data.size())
305 fatal(msg: toString(this) + ": offset is outside the section");
306
307 auto it =
308 partition_point(Range&: pieces, P: [=](StringPiece p) { return p.inSecOff <= off; });
309 return it[-1];
310}
311
312const StringPiece &CStringInputSection::getStringPiece(uint64_t off) const {
313 return const_cast<CStringInputSection *>(this)->getStringPiece(off);
314}
315
316size_t CStringInputSection::getStringPieceIndex(uint64_t off) const {
317 if (off >= data.size())
318 fatal(msg: toString(this) + ": offset is outside the section");
319
320 auto it =
321 partition_point(Range: pieces, P: [=](StringPiece p) { return p.inSecOff <= off; });
322 return std::distance(first: pieces.begin(), last: it) - 1;
323}
324
325uint64_t CStringInputSection::getOffset(uint64_t off) const {
326 const StringPiece &piece = getStringPiece(off);
327 uint64_t addend = off - piece.inSecOff;
328 return piece.outSecOff + addend;
329}
330
331WordLiteralInputSection::WordLiteralInputSection(const Section &section,
332 ArrayRef<uint8_t> data,
333 uint32_t align)
334 : InputSection(WordLiteralKind, section, data, align) {
335 switch (sectionType(flags: getFlags())) {
336 case S_4BYTE_LITERALS:
337 power2LiteralSize = 2;
338 break;
339 case S_8BYTE_LITERALS:
340 power2LiteralSize = 3;
341 break;
342 case S_16BYTE_LITERALS:
343 power2LiteralSize = 4;
344 break;
345 default:
346 llvm_unreachable("invalid literal section type");
347 }
348
349 live.resize(N: data.size() >> power2LiteralSize, t: !config->deadStrip);
350}
351
352uint64_t WordLiteralInputSection::getOffset(uint64_t off) const {
353 auto *osec = cast<WordLiteralSection>(Val: parent);
354 const uintptr_t buf = reinterpret_cast<uintptr_t>(data.data());
355 switch (sectionType(flags: getFlags())) {
356 case S_4BYTE_LITERALS:
357 return osec->getLiteral4Offset(buf: buf + (off & ~3LLU)) | (off & 3);
358 case S_8BYTE_LITERALS:
359 return osec->getLiteral8Offset(buf: buf + (off & ~7LLU)) | (off & 7);
360 case S_16BYTE_LITERALS:
361 return osec->getLiteral16Offset(buf: buf + (off & ~15LLU)) | (off & 15);
362 default:
363 llvm_unreachable("invalid literal section type");
364 }
365}
366
367bool macho::isCodeSection(const InputSection *isec) {
368 return sections::isCodeSection(name: isec->getName(), segName: isec->getSegName(),
369 flags: isec->getFlags());
370}
371
372bool macho::isCfStringSection(const InputSection *isec) {
373 return isec->getName() == section_names::cfString &&
374 isec->getSegName() == segment_names::data;
375}
376
377bool macho::isClassRefsSection(const InputSection *isec) {
378 return isec->getName() == section_names::objcClassRefs &&
379 isec->getSegName() == segment_names::data;
380}
381
382bool macho::isSelRefsSection(const InputSection *isec) {
383 return isec->getName() == section_names::objcSelrefs &&
384 isec->getSegName() == segment_names::data;
385}
386
387bool macho::isEhFrameSection(const InputSection *isec) {
388 return isec->getName() == section_names::ehFrame &&
389 isec->getSegName() == segment_names::text;
390}
391
392bool macho::isGccExceptTabSection(const InputSection *isec) {
393 return isec->getName() == section_names::gccExceptTab &&
394 isec->getSegName() == segment_names::text;
395}
396
397std::string lld::toString(const InputSection *isec) {
398 return (toString(file: isec->getFile()) + ":(" + isec->getName() + ")").str();
399}
400