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-exception
6//
7//===----------------------------------------------------------------------===//
8
9#ifndef LLD_ELF_RELOCATIONS_H
10#define LLD_ELF_RELOCATIONS_H
11
12#include "lld/Common/LLVM.h"
13#include "llvm/ADT/DenseMap.h"
14#include "llvm/ADT/STLExtras.h"
15#include <vector>
16
17namespace lld::elf {
18class Symbol;
19class InputSection;
20class InputSectionBase;
21class OutputSection;
22class SectionBase;
23
24// Represents a relocation type, such as R_X86_64_PC32 or R_ARM_THM_CALL.
25using RelType = uint32_t;
26using JumpModType = uint32_t;
27
28// List of target-independent relocation types. Relocations read
29// from files are converted to these types so that the main code
30// doesn't have to know about architecture-specific details.
31enum RelExpr {
32  R_ABS,
33  R_ADDEND,
34  R_DTPREL,
35  R_GOT,
36  R_GOT_OFF,
37  R_GOT_PC,
38  R_GOTONLY_PC,
39  R_GOTPLTONLY_PC,
40  R_GOTPLT,
41  R_GOTPLTREL,
42  R_GOTREL,
43  R_GOTPLT_GOTREL,
44  R_GOTPLT_PC,
45  R_NONE,
46  R_PC,
47  R_PLT,
48  R_PLT_PC,
49  R_PLT_GOTPLT,
50  R_PLT_GOTREL,
51  R_RELAX_HINT,
52  R_RELAX_GOT_PC,
53  R_RELAX_GOT_PC_NOPIC,
54  R_RELAX_TLS_GD_TO_IE,
55  R_RELAX_TLS_GD_TO_IE_ABS,
56  R_RELAX_TLS_GD_TO_IE_GOT_OFF,
57  R_RELAX_TLS_GD_TO_IE_GOTPLT,
58  R_RELAX_TLS_GD_TO_LE,
59  R_RELAX_TLS_GD_TO_LE_NEG,
60  R_RELAX_TLS_IE_TO_LE,
61  R_RELAX_TLS_LD_TO_LE,
62  R_RELAX_TLS_LD_TO_LE_ABS,
63  R_SIZE,
64  R_TPREL,
65  R_TPREL_NEG,
66  R_TLSDESC,
67  R_TLSDESC_CALL,
68  R_TLSDESC_PC,
69  R_TLSDESC_GOTPLT,
70  R_TLSGD_GOT,
71  R_TLSGD_GOTPLT,
72  R_TLSGD_PC,
73  R_TLSIE_HINT,
74  R_TLSLD_GOT,
75  R_TLSLD_GOTPLT,
76  R_TLSLD_GOT_OFF,
77  R_TLSLD_HINT,
78  R_TLSLD_PC,
79
80  // The following is abstract relocation types used for only one target.
81  //
82  // Even though RelExpr is intended to be a target-neutral representation
83  // of a relocation type, there are some relocations whose semantics are
84  // unique to a target. Such relocation are marked with R_<TARGET_NAME>.
85  R_AARCH64_GOT_PAGE_PC,
86  R_AARCH64_GOT_PAGE,
87  R_AARCH64_PAGE_PC,
88  R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC,
89  R_AARCH64_TLSDESC_PAGE,
90  R_ARM_PCA,
91  R_ARM_SBREL,
92  R_MIPS_GOTREL,
93  R_MIPS_GOT_GP,
94  R_MIPS_GOT_GP_PC,
95  R_MIPS_GOT_LOCAL_PAGE,
96  R_MIPS_GOT_OFF,
97  R_MIPS_GOT_OFF32,
98  R_MIPS_TLSGD,
99  R_MIPS_TLSLD,
100  R_PPC32_PLTREL,
101  R_PPC64_CALL,
102  R_PPC64_CALL_PLT,
103  R_PPC64_RELAX_TOC,
104  R_PPC64_TOCBASE,
105  R_PPC64_RELAX_GOT_PC,
106  R_RISCV_ADD,
107  R_RISCV_LEB128,
108  R_RISCV_PC_INDIRECT,
109  // Same as R_PC but with page-aligned semantics.
110  R_LOONGARCH_PAGE_PC,
111  // Same as R_PLT_PC but with page-aligned semantics.
112  R_LOONGARCH_PLT_PAGE_PC,
113  // In addition to having page-aligned semantics, LoongArch GOT relocs are
114  // also reused for TLS, making the semantics differ from other architectures.
115  R_LOONGARCH_GOT,
116  R_LOONGARCH_GOT_PAGE_PC,
117  R_LOONGARCH_TLSGD_PAGE_PC,
118};
119
120// Architecture-neutral representation of relocation.
121struct Relocation {
122  RelExpr expr;
123  RelType type;
124  uint64_t offset;
125  int64_t addend;
126  Symbol *sym;
127};
128
129// Manipulate jump instructions with these modifiers.  These are used to relax
130// jump instruction opcodes at basic block boundaries and are particularly
131// useful when basic block sections are enabled.
132struct JumpInstrMod {
133  uint64_t offset;
134  JumpModType original;
135  unsigned size;
136};
137
138// This function writes undefined symbol diagnostics to an internal buffer.
139// Call reportUndefinedSymbols() after calling scanRelocations() to emit
140// the diagnostics.
141template <class ELFT> void scanRelocations();
142void reportUndefinedSymbols();
143void postScanRelocations();
144void addGotEntry(Symbol &sym);
145
146void hexagonTLSSymbolUpdate(ArrayRef<OutputSection *> outputSections);
147bool hexagonNeedsTLSSymbol(ArrayRef<OutputSection *> outputSections);
148
149class ThunkSection;
150class Thunk;
151class InputSectionDescription;
152
153class ThunkCreator {
154public:
155  // Return true if Thunks have been added to OutputSections
156  bool createThunks(uint32_t pass, ArrayRef<OutputSection *> outputSections);
157
158private:
159  void mergeThunks(ArrayRef<OutputSection *> outputSections);
160
161  ThunkSection *getISDThunkSec(OutputSection *os, InputSection *isec,
162                               InputSectionDescription *isd,
163                               const Relocation &rel, uint64_t src);
164
165  ThunkSection *getISThunkSec(InputSection *isec);
166
167  void createInitialThunkSections(ArrayRef<OutputSection *> outputSections);
168
169  std::pair<Thunk *, bool> getThunk(InputSection *isec, Relocation &rel,
170                                    uint64_t src);
171
172  ThunkSection *addThunkSection(OutputSection *os, InputSectionDescription *,
173                                uint64_t off);
174
175  bool normalizeExistingThunk(Relocation &rel, uint64_t src);
176
177  // Record all the available Thunks for a (Symbol, addend) pair, where Symbol
178  // is represented as a (section, offset) pair. There may be multiple
179  // relocations sharing the same (section, offset + addend) pair. We may revert
180  // a relocation back to its original non-Thunk target, and restore the
181  // original addend, so we cannot fold offset + addend. A nested pair is used
182  // because DenseMapInfo is not specialized for std::tuple.
183  llvm::DenseMap<std::pair<std::pair<SectionBase *, uint64_t>, int64_t>,
184                 std::vector<Thunk *>>
185      thunkedSymbolsBySectionAndAddend;
186  llvm::DenseMap<std::pair<Symbol *, int64_t>, std::vector<Thunk *>>
187      thunkedSymbols;
188
189  // Find a Thunk from the Thunks symbol definition, we can use this to find
190  // the Thunk from a relocation to the Thunks symbol definition.
191  llvm::DenseMap<Symbol *, Thunk *> thunks;
192
193  // Track InputSections that have an inline ThunkSection placed in front
194  // an inline ThunkSection may have control fall through to the section below
195  // so we need to make sure that there is only one of them.
196  // The Mips LA25 Thunk is an example of an inline ThunkSection.
197  llvm::DenseMap<InputSection *, ThunkSection *> thunkedSections;
198
199  // The number of completed passes of createThunks this permits us
200  // to do one time initialization on Pass 0 and put a limit on the
201  // number of times it can be called to prevent infinite loops.
202  uint32_t pass = 0;
203};
204
205// Return a int64_t to make sure we get the sign extension out of the way as
206// early as possible.
207template <class ELFT>
208static inline int64_t getAddend(const typename ELFT::Rel &rel) {
209  return 0;
210}
211template <class ELFT>
212static inline int64_t getAddend(const typename ELFT::Rela &rel) {
213  return rel.r_addend;
214}
215
216template <typename RelTy>
217ArrayRef<RelTy> sortRels(ArrayRef<RelTy> rels, SmallVector<RelTy, 0> &storage) {
218  auto cmp = [](const RelTy &a, const RelTy &b) {
219    return a.r_offset < b.r_offset;
220  };
221  if (!llvm::is_sorted(rels, cmp)) {
222    storage.assign(rels.begin(), rels.end());
223    llvm::stable_sort(storage, cmp);
224    rels = storage;
225  }
226  return rels;
227}
228
229// Returns true if Expr refers a GOT entry. Note that this function returns
230// false for TLS variables even though they need GOT, because TLS variables uses
231// GOT differently than the regular variables.
232bool needsGot(RelExpr expr);
233} // namespace lld::elf
234
235#endif
236