385 lines · c
1//===- Relocations.h -------------------------------------------*- C++ -*-===//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-exception6//7//===----------------------------------------------------------------------===//8 9#ifndef LLD_ELF_RELOCATIONS_H10#define LLD_ELF_RELOCATIONS_H11 12#include "lld/Common/LLVM.h"13#include "llvm/ADT/DenseMap.h"14#include "llvm/ADT/STLExtras.h"15#include "llvm/Object/ELFTypes.h"16#include <vector>17 18namespace lld::elf {19struct Ctx;20struct ELFSyncStream;21class Defined;22class Undefined;23class Symbol;24class InputSection;25class InputSectionBase;26class OutputSection;27class RelocationBaseSection;28class SectionBase;29 30// Represents a relocation type, such as R_X86_64_PC32 or R_ARM_THM_CALL.31struct RelType {32 uint32_t v = 0;33 /*implicit*/ constexpr RelType(uint32_t v = 0) : v(v) {}34 /*implicit*/ operator uint32_t() const { return v; }35};36 37using JumpModType = uint32_t;38 39// List of target-independent relocation types. Relocations read40// from files are converted to these types so that the main code41// doesn't have to know about architecture-specific details.42enum RelExpr {43 R_ABS,44 R_ADDEND,45 R_DTPREL,46 R_GOT,47 R_GOT_OFF,48 R_GOT_PC,49 R_GOTONLY_PC,50 R_GOTPLTONLY_PC,51 R_GOTPLT,52 R_GOTPLTREL,53 R_GOTREL,54 R_GOTPLT_GOTREL,55 R_GOTPLT_PC,56 R_NONE,57 R_PC,58 R_PLT,59 R_PLT_PC,60 R_PLT_GOTPLT,61 R_PLT_GOTREL,62 R_RELAX_HINT,63 R_RELAX_GOT_PC,64 R_RELAX_GOT_PC_NOPIC,65 R_RELAX_TLS_GD_TO_IE,66 R_RELAX_TLS_GD_TO_IE_ABS,67 R_RELAX_TLS_GD_TO_IE_GOT_OFF,68 R_RELAX_TLS_GD_TO_IE_GOTPLT,69 R_RELAX_TLS_GD_TO_LE,70 R_RELAX_TLS_GD_TO_LE_NEG,71 R_RELAX_TLS_IE_TO_LE,72 R_RELAX_TLS_LD_TO_LE,73 R_RELAX_TLS_LD_TO_LE_ABS,74 R_SIZE,75 R_TPREL,76 R_TPREL_NEG,77 R_TLSDESC,78 R_TLSDESC_CALL,79 R_TLSDESC_PC,80 R_TLSDESC_GOTPLT,81 R_TLSGD_GOT,82 R_TLSGD_GOTPLT,83 R_TLSGD_PC,84 R_TLSIE_HINT,85 R_TLSLD_GOT,86 R_TLSLD_GOTPLT,87 R_TLSLD_GOT_OFF,88 R_TLSLD_HINT,89 R_TLSLD_PC,90 91 // The following is abstract relocation types used for only one target.92 //93 // Even though RelExpr is intended to be a target-neutral representation94 // of a relocation type, there are some relocations whose semantics are95 // unique to a target. Such relocation are marked with RE_<TARGET_NAME>.96 RE_AARCH64_GOT_PAGE_PC,97 RE_AARCH64_AUTH_GOT_PAGE_PC,98 RE_AARCH64_GOT_PAGE,99 RE_AARCH64_AUTH_GOT,100 RE_AARCH64_AUTH_GOT_PC,101 RE_AARCH64_PAGE_PC,102 RE_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC,103 RE_AARCH64_TLSDESC_PAGE,104 RE_AARCH64_AUTH_TLSDESC_PAGE,105 RE_AARCH64_AUTH_TLSDESC,106 RE_AARCH64_AUTH,107 RE_ARM_PCA,108 RE_ARM_SBREL,109 RE_MIPS_GOTREL,110 RE_MIPS_GOT_GP,111 RE_MIPS_GOT_GP_PC,112 RE_MIPS_GOT_LOCAL_PAGE,113 RE_MIPS_GOT_OFF,114 RE_MIPS_GOT_OFF32,115 RE_MIPS_OSEC_LOCAL_PAGE,116 RE_MIPS_TLSGD,117 RE_MIPS_TLSLD,118 RE_PPC32_PLTREL,119 RE_PPC64_CALL,120 RE_PPC64_CALL_PLT,121 RE_PPC64_RELAX_TOC,122 RE_PPC64_TOCBASE,123 RE_PPC64_RELAX_GOT_PC,124 RE_RISCV_ADD,125 RE_RISCV_LEB128,126 RE_RISCV_PC_INDIRECT,127 // Same as R_PC but with page-aligned semantics.128 RE_LOONGARCH_PAGE_PC,129 // Same as R_PLT_PC but with page-aligned semantics.130 RE_LOONGARCH_PLT_PAGE_PC,131 // In addition to having page-aligned semantics, LoongArch GOT relocs are132 // also reused for TLS, making the semantics differ from other architectures.133 RE_LOONGARCH_GOT,134 RE_LOONGARCH_GOT_PAGE_PC,135 RE_LOONGARCH_TLSGD_PAGE_PC,136 RE_LOONGARCH_TLSDESC_PAGE_PC,137 RE_LOONGARCH_RELAX_TLS_GD_TO_IE_PAGE_PC,138};139 140// Architecture-neutral representation of relocation.141struct Relocation {142 RelExpr expr;143 RelType type;144 uint64_t offset;145 int64_t addend;146 Symbol *sym;147};148 149// Manipulate jump instructions with these modifiers. These are used to relax150// jump instruction opcodes at basic block boundaries and are particularly151// useful when basic block sections are enabled.152struct JumpInstrMod {153 uint64_t offset;154 JumpModType original;155 unsigned size;156};157 158void printLocation(ELFSyncStream &s, InputSectionBase &sec, const Symbol &sym,159 uint64_t off);160 161// This function writes undefined symbol diagnostics to an internal buffer.162// Call reportUndefinedSymbols() after calling scanRelocations() to emit163// the diagnostics.164template <class ELFT> void scanRelocations(Ctx &ctx);165template <class ELFT> void checkNoCrossRefs(Ctx &ctx);166void reportUndefinedSymbols(Ctx &);167bool maybeReportUndefined(Ctx &, Undefined &sym, InputSectionBase &sec,168 uint64_t offset);169void postScanRelocations(Ctx &ctx);170void addGotEntry(Ctx &ctx, Symbol &sym);171 172void hexagonTLSSymbolUpdate(Ctx &ctx);173bool hexagonNeedsTLSSymbol(ArrayRef<OutputSection *> outputSections);174 175bool isAbsolute(const Symbol &sym);176 177class ThunkSection;178class Thunk;179class InputSectionDescription;180 181class ThunkCreator {182public:183 // Thunk may be incomplete. Avoid inline ctor/dtor.184 ThunkCreator(Ctx &ctx);185 ~ThunkCreator();186 // Return true if Thunks have been added to OutputSections187 bool createThunks(uint32_t pass, ArrayRef<OutputSection *> outputSections);188 189private:190 void mergeThunks(ArrayRef<OutputSection *> outputSections);191 192 ThunkSection *getISDThunkSec(OutputSection *os, InputSection *isec,193 InputSectionDescription *isd,194 const Relocation &rel, uint64_t src);195 196 ThunkSection *getISThunkSec(InputSection *isec);197 198 void createInitialThunkSections(ArrayRef<OutputSection *> outputSections);199 200 std::pair<Thunk *, bool> getThunk(InputSection *isec, Relocation &rel,201 uint64_t src);202 203 std::pair<Thunk *, bool> getSyntheticLandingPad(Defined &d, int64_t a);204 205 ThunkSection *addThunkSection(OutputSection *os, InputSectionDescription *,206 uint64_t off);207 208 bool normalizeExistingThunk(Relocation &rel, uint64_t src);209 210 bool addSyntheticLandingPads();211 212 Ctx &ctx;213 214 // Record all the available Thunks for a (Symbol, addend) pair, where Symbol215 // is represented as a (section, offset) pair. There may be multiple216 // relocations sharing the same (section, offset + addend) pair. We may revert217 // a relocation back to its original non-Thunk target, and restore the218 // original addend, so we cannot fold offset + addend. A nested pair is used219 // because DenseMapInfo is not specialized for std::tuple.220 llvm::DenseMap<std::pair<std::pair<SectionBase *, uint64_t>, int64_t>,221 SmallVector<std::unique_ptr<Thunk>, 0>>222 thunkedSymbolsBySectionAndAddend;223 llvm::DenseMap<std::pair<Symbol *, int64_t>,224 SmallVector<std::unique_ptr<Thunk>, 0>>225 thunkedSymbols;226 227 // Find a Thunk from the Thunks symbol definition, we can use this to find228 // the Thunk from a relocation to the Thunks symbol definition.229 llvm::DenseMap<Symbol *, Thunk *> thunks;230 231 // Track InputSections that have an inline ThunkSection placed in front232 // an inline ThunkSection may have control fall through to the section below233 // so we need to make sure that there is only one of them.234 // The Mips LA25 Thunk is an example of an inline ThunkSection, as is235 // the AArch64BTLandingPadThunk.236 llvm::DenseMap<InputSection *, ThunkSection *> thunkedSections;237 238 // Record landing pads, generated for a section + offset destination.239 // Landling pads are alternative entry points for destinations that need240 // to be reached via thunks that use indirect branches. A destination241 // needs at most one landing pad as that can be reused by all callers.242 llvm::DenseMap<std::pair<std::pair<SectionBase *, uint64_t>, int64_t>,243 std::unique_ptr<Thunk>>244 landingPadsBySectionAndAddend;245 246 // All the nonLandingPad thunks that have been created, in order of creation.247 std::vector<Thunk *> allThunks;248 249 // The number of completed passes of createThunks this permits us250 // to do one time initialization on Pass 0 and put a limit on the251 // number of times it can be called to prevent infinite loops.252 uint32_t pass = 0;253};254 255// Decode LEB128 without error checking. Only used by performance critical code256// like RelocsCrel.257inline uint64_t readLEB128(const uint8_t *&p, uint64_t leb) {258 uint64_t acc = 0, shift = 0, byte;259 do {260 byte = *p++;261 acc |= (byte - 128 * (byte >= leb)) << shift;262 shift += 7;263 } while (byte >= 128);264 return acc;265}266inline uint64_t readULEB128(const uint8_t *&p) { return readLEB128(p, 128); }267inline int64_t readSLEB128(const uint8_t *&p) { return readLEB128(p, 64); }268 269// This class implements a CREL iterator that does not allocate extra memory.270template <bool is64> struct RelocsCrel {271 using uint = std::conditional_t<is64, uint64_t, uint32_t>;272 struct const_iterator {273 using iterator_category = std::forward_iterator_tag;274 using value_type = llvm::object::Elf_Crel_Impl<is64>;275 using difference_type = ptrdiff_t;276 using pointer = value_type *;277 using reference = const value_type &;278 uint32_t count;279 uint8_t flagBits, shift;280 const uint8_t *p;281 llvm::object::Elf_Crel_Impl<is64> crel{};282 const_iterator(size_t hdr, const uint8_t *p)283 : count(hdr / 8), flagBits(hdr & 4 ? 3 : 2), shift(hdr % 4), p(p) {284 if (count)285 step();286 }287 void step() {288 // See object::decodeCrel.289 const uint8_t b = *p++;290 crel.r_offset += b >> flagBits << shift;291 if (b >= 0x80)292 crel.r_offset +=293 ((readULEB128(p) << (7 - flagBits)) - (0x80 >> flagBits)) << shift;294 if (b & 1)295 crel.r_symidx += readSLEB128(p);296 if (b & 2)297 crel.r_type += readSLEB128(p);298 if (b & 4 && flagBits == 3)299 crel.r_addend += static_cast<uint>(readSLEB128(p));300 }301 llvm::object::Elf_Crel_Impl<is64> operator*() const { return crel; };302 const llvm::object::Elf_Crel_Impl<is64> *operator->() const {303 return &crel;304 }305 // For llvm::enumerate.306 bool operator==(const const_iterator &r) const { return count == r.count; }307 bool operator!=(const const_iterator &r) const { return count != r.count; }308 const_iterator &operator++() {309 if (--count)310 step();311 return *this;312 }313 // For RelocationScanner::scanOne.314 void operator+=(size_t n) {315 for (; n; --n)316 operator++();317 }318 };319 320 size_t hdr = 0;321 const uint8_t *p = nullptr;322 323 constexpr RelocsCrel() = default;324 RelocsCrel(const uint8_t *p) : hdr(readULEB128(p)) { this->p = p; }325 size_t size() const { return hdr / 8; }326 const_iterator begin() const { return {hdr, p}; }327 const_iterator end() const { return {0, nullptr}; }328};329 330template <class RelTy> struct Relocs : ArrayRef<RelTy> {331 Relocs() = default;332 Relocs(ArrayRef<RelTy> a) : ArrayRef<RelTy>(a) {}333};334 335template <bool is64>336struct Relocs<llvm::object::Elf_Crel_Impl<is64>> : RelocsCrel<is64> {337 using RelocsCrel<is64>::RelocsCrel;338};339 340// Return a int64_t to make sure we get the sign extension out of the way as341// early as possible.342template <class ELFT>343static inline int64_t getAddend(const typename ELFT::Rel &rel) {344 return 0;345}346template <class ELFT>347static inline int64_t getAddend(const typename ELFT::Rela &rel) {348 return rel.r_addend;349}350template <class ELFT>351static inline int64_t getAddend(const typename ELFT::Crel &rel) {352 return rel.r_addend;353}354 355template <typename RelTy>356inline Relocs<RelTy> sortRels(Relocs<RelTy> rels,357 SmallVector<RelTy, 0> &storage) {358 auto cmp = [](const RelTy &a, const RelTy &b) {359 return a.r_offset < b.r_offset;360 };361 if (!llvm::is_sorted(rels, cmp)) {362 storage.assign(rels.begin(), rels.end());363 llvm::stable_sort(storage, cmp);364 rels = Relocs<RelTy>(storage);365 }366 return rels;367}368 369template <bool is64>370inline Relocs<llvm::object::Elf_Crel_Impl<is64>>371sortRels(Relocs<llvm::object::Elf_Crel_Impl<is64>> rels,372 SmallVector<llvm::object::Elf_Crel_Impl<is64>, 0> &storage) {373 return {};374}375 376RelocationBaseSection &getIRelativeSection(Ctx &ctx);377 378// Returns true if Expr refers a GOT entry. Note that this function returns379// false for TLS variables even though they need GOT, because TLS variables uses380// GOT differently than the regular variables.381bool needsGot(RelExpr expr);382} // namespace lld::elf383 384#endif385