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 "InputFiles.h"
11 #include "OutputSegment.h"
12 #include "Symbols.h"
13 #include "SyntheticSections.h"
14 #include "Target.h"
15 #include "UnwindInfoSection.h"
16 #include "Writer.h"
17 #include "lld/Common/Memory.h"
18 #include "llvm/Support/Endian.h"
19 #include "llvm/Support/xxhash.h"
20 
21 using namespace llvm;
22 using namespace llvm::MachO;
23 using namespace llvm::support;
24 using namespace lld;
25 using namespace lld::macho;
26 
27 std::vector<InputSection *> macho::inputSections;
28 
29 uint64_t InputSection::getFileSize() const {
30   return isZeroFill(flags) ? 0 : getSize();
31 }
32 
33 uint64_t InputSection::getVA(uint64_t off) const {
34   return parent->addr + getOffset(off);
35 }
36 
37 static uint64_t resolveSymbolVA(const Symbol *sym, uint8_t type) {
38   const RelocAttrs &relocAttrs = target->getRelocAttrs(type);
39   if (relocAttrs.hasAttr(RelocAttrBits::BRANCH))
40     return sym->resolveBranchVA();
41   else if (relocAttrs.hasAttr(RelocAttrBits::GOT))
42     return sym->resolveGotVA();
43   else if (relocAttrs.hasAttr(RelocAttrBits::TLV))
44     return sym->resolveTlvVA();
45   return sym->getVA();
46 }
47 
48 // ICF needs to hash any section that might potentially be duplicated so
49 // that it can match on content rather than identity.
50 bool ConcatInputSection::isHashableForICF(bool isText) const {
51   if (shouldOmitFromOutput())
52     return false;
53   switch (sectionType(flags)) {
54   case S_REGULAR:
55     if (isText)
56       return !hasPersonality;
57     // One might hope that we could hash __TEXT,__const subsections to fold
58     // references to duplicated values, but alas, many tests fail.
59     return false;
60   case S_CSTRING_LITERALS:
61   case S_4BYTE_LITERALS:
62   case S_8BYTE_LITERALS:
63   case S_16BYTE_LITERALS:
64   case S_LITERAL_POINTERS:
65     // FIXME(jezng): We should not have any ConcatInputSections of these types
66     // when running ICF.
67     return false;
68   case S_ZEROFILL:
69   case S_GB_ZEROFILL:
70   case S_NON_LAZY_SYMBOL_POINTERS:
71   case S_LAZY_SYMBOL_POINTERS:
72   case S_SYMBOL_STUBS:
73   case S_MOD_INIT_FUNC_POINTERS:
74   case S_MOD_TERM_FUNC_POINTERS:
75   case S_COALESCED:
76   case S_INTERPOSING:
77   case S_DTRACE_DOF:
78   case S_LAZY_DYLIB_SYMBOL_POINTERS:
79   case S_THREAD_LOCAL_REGULAR:
80   case S_THREAD_LOCAL_ZEROFILL:
81   case S_THREAD_LOCAL_VARIABLES:
82   case S_THREAD_LOCAL_VARIABLE_POINTERS:
83   case S_THREAD_LOCAL_INIT_FUNCTION_POINTERS:
84     return false;
85   default:
86     llvm_unreachable("Section type");
87   }
88 }
89 
90 void ConcatInputSection::hashForICF() {
91   assert(data.data()); // zeroFill section data has nullptr with non-zero size
92   assert(icfEqClass[0] == 0); // don't overwrite a unique ID!
93   // Turn-on the top bit to guarantee that valid hashes have no collisions
94   // with the small-integer unique IDs for ICF-ineligible sections
95   icfEqClass[0] = xxHash64(data) | (1ull << 63);
96 }
97 
98 void ConcatInputSection::foldIdentical(ConcatInputSection *copy) {
99   align = std::max(align, copy->align);
100   copy->live = false;
101   copy->wasCoalesced = true;
102   numRefs += copy->numRefs;
103   copy->numRefs = 0;
104   copy->replacement = this;
105 }
106 
107 void ConcatInputSection::writeTo(uint8_t *buf) {
108   assert(!shouldOmitFromOutput());
109 
110   if (getFileSize() == 0)
111     return;
112 
113   memcpy(buf, data.data(), data.size());
114 
115   for (size_t i = 0; i < relocs.size(); i++) {
116     const Reloc &r = relocs[i];
117     uint8_t *loc = buf + r.offset;
118     uint64_t referentVA = 0;
119     if (target->hasAttr(r.type, RelocAttrBits::SUBTRAHEND)) {
120       const Symbol *fromSym = r.referent.get<Symbol *>();
121       const Reloc &minuend = relocs[++i];
122       uint64_t minuendVA;
123       if (const Symbol *toSym = minuend.referent.dyn_cast<Symbol *>())
124         minuendVA = toSym->getVA() + minuend.addend;
125       else {
126         auto *referentIsec = minuend.referent.get<InputSection *>();
127         assert(!::shouldOmitFromOutput(referentIsec));
128         minuendVA = referentIsec->getVA(minuend.addend);
129       }
130       referentVA = minuendVA - fromSym->getVA();
131     } else if (auto *referentSym = r.referent.dyn_cast<Symbol *>()) {
132       if (target->hasAttr(r.type, RelocAttrBits::LOAD) &&
133           !referentSym->isInGot())
134         target->relaxGotLoad(loc, r.type);
135       referentVA = resolveSymbolVA(referentSym, r.type) + r.addend;
136 
137       if (isThreadLocalVariables(flags)) {
138         // References from thread-local variable sections are treated as offsets
139         // relative to the start of the thread-local data memory area, which
140         // is initialized via copying all the TLV data sections (which are all
141         // contiguous).
142         if (isa<Defined>(referentSym))
143           referentVA -= firstTLVDataSection->addr;
144       }
145     } else if (auto *referentIsec = r.referent.dyn_cast<InputSection *>()) {
146       assert(!::shouldOmitFromOutput(referentIsec));
147       referentVA = referentIsec->getVA(r.addend);
148     }
149     target->relocateOne(loc, r, referentVA, getVA() + r.offset);
150   }
151 }
152 
153 void CStringInputSection::splitIntoPieces() {
154   size_t off = 0;
155   StringRef s = toStringRef(data);
156   while (!s.empty()) {
157     size_t end = s.find(0);
158     if (end == StringRef::npos)
159       fatal(toString(this) + ": string is not null terminated");
160     size_t size = end + 1;
161     pieces.emplace_back(off, xxHash64(s.substr(0, size)));
162     s = s.substr(size);
163     off += size;
164   }
165 }
166 
167 StringPiece &CStringInputSection::getStringPiece(uint64_t off) {
168   if (off >= data.size())
169     fatal(toString(this) + ": offset is outside the section");
170 
171   auto it =
172       partition_point(pieces, [=](StringPiece p) { return p.inSecOff <= off; });
173   return it[-1];
174 }
175 
176 const StringPiece &CStringInputSection::getStringPiece(uint64_t off) const {
177   return const_cast<CStringInputSection *>(this)->getStringPiece(off);
178 }
179 
180 uint64_t CStringInputSection::getOffset(uint64_t off) const {
181   const StringPiece &piece = getStringPiece(off);
182   uint64_t addend = off - piece.inSecOff;
183   return piece.outSecOff + addend;
184 }
185 
186 WordLiteralInputSection::WordLiteralInputSection(StringRef segname,
187                                                  StringRef name,
188                                                  InputFile *file,
189                                                  ArrayRef<uint8_t> data,
190                                                  uint32_t align, uint32_t flags)
191     : InputSection(WordLiteralKind, segname, name, file, data, align, flags) {
192   switch (sectionType(flags)) {
193   case S_4BYTE_LITERALS:
194     power2LiteralSize = 2;
195     break;
196   case S_8BYTE_LITERALS:
197     power2LiteralSize = 3;
198     break;
199   case S_16BYTE_LITERALS:
200     power2LiteralSize = 4;
201     break;
202   default:
203     llvm_unreachable("invalid literal section type");
204   }
205 
206   live.resize(data.size() >> power2LiteralSize, !config->deadStrip);
207 }
208 
209 uint64_t WordLiteralInputSection::getOffset(uint64_t off) const {
210   auto *osec = cast<WordLiteralSection>(parent);
211   const uint8_t *buf = data.data();
212   switch (sectionType(flags)) {
213   case S_4BYTE_LITERALS:
214     return osec->getLiteral4Offset(buf + off);
215   case S_8BYTE_LITERALS:
216     return osec->getLiteral8Offset(buf + off);
217   case S_16BYTE_LITERALS:
218     return osec->getLiteral16Offset(buf + off);
219   default:
220     llvm_unreachable("invalid literal section type");
221   }
222 }
223 
224 bool macho::isCodeSection(const InputSection *isec) {
225   uint32_t type = sectionType(isec->flags);
226   if (type != S_REGULAR && type != S_COALESCED)
227     return false;
228 
229   uint32_t attr = isec->flags & SECTION_ATTRIBUTES_USR;
230   if (attr == S_ATTR_PURE_INSTRUCTIONS)
231     return true;
232 
233   if (isec->segname == segment_names::text)
234     return StringSwitch<bool>(isec->name)
235         .Cases(section_names::textCoalNt, section_names::staticInit, true)
236         .Default(false);
237 
238   return false;
239 }
240 
241 std::string lld::toString(const InputSection *isec) {
242   return (toString(isec->file) + ":(" + isec->name + ")").str();
243 }
244