xref: /freebsd/contrib/llvm-project/lld/ELF/Relocations.h (revision 924226fba12cc9a228c73b956e1b7fa24c60b055)
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 <map>
16 #include <vector>
17 
18 namespace lld {
19 namespace elf {
20 class Symbol;
21 class InputSection;
22 class InputSectionBase;
23 class OutputSection;
24 class SectionBase;
25 
26 // Represents a relocation type, such as R_X86_64_PC32 or R_ARM_THM_CALL.
27 using RelType = uint32_t;
28 using JumpModType = uint32_t;
29 
30 // List of target-independent relocation types. Relocations read
31 // from files are converted to these types so that the main code
32 // doesn't have to know about architecture-specific details.
33 enum RelExpr {
34   R_ABS,
35   R_ADDEND,
36   R_DTPREL,
37   R_GOT,
38   R_GOT_OFF,
39   R_GOT_PC,
40   R_GOTONLY_PC,
41   R_GOTPLTONLY_PC,
42   R_GOTPLT,
43   R_GOTPLTREL,
44   R_GOTREL,
45   R_NONE,
46   R_PC,
47   R_PLT,
48   R_PLT_PC,
49   R_PLT_GOTPLT,
50   R_RELAX_GOT_PC,
51   R_RELAX_GOT_PC_NOPIC,
52   R_RELAX_TLS_GD_TO_IE,
53   R_RELAX_TLS_GD_TO_IE_ABS,
54   R_RELAX_TLS_GD_TO_IE_GOT_OFF,
55   R_RELAX_TLS_GD_TO_IE_GOTPLT,
56   R_RELAX_TLS_GD_TO_LE,
57   R_RELAX_TLS_GD_TO_LE_NEG,
58   R_RELAX_TLS_IE_TO_LE,
59   R_RELAX_TLS_LD_TO_LE,
60   R_RELAX_TLS_LD_TO_LE_ABS,
61   R_SIZE,
62   R_TPREL,
63   R_TPREL_NEG,
64   R_TLSDESC,
65   R_TLSDESC_CALL,
66   R_TLSDESC_PC,
67   R_TLSDESC_GOTPLT,
68   R_TLSGD_GOT,
69   R_TLSGD_GOTPLT,
70   R_TLSGD_PC,
71   R_TLSIE_HINT,
72   R_TLSLD_GOT,
73   R_TLSLD_GOTPLT,
74   R_TLSLD_GOT_OFF,
75   R_TLSLD_HINT,
76   R_TLSLD_PC,
77 
78   // The following is abstract relocation types used for only one target.
79   //
80   // Even though RelExpr is intended to be a target-neutral representation
81   // of a relocation type, there are some relocations whose semantics are
82   // unique to a target. Such relocation are marked with R_<TARGET_NAME>.
83   R_AARCH64_GOT_PAGE_PC,
84   R_AARCH64_GOT_PAGE,
85   R_AARCH64_PAGE_PC,
86   R_AARCH64_RELAX_TLS_GD_TO_IE_PAGE_PC,
87   R_AARCH64_TLSDESC_PAGE,
88   R_ARM_PCA,
89   R_ARM_SBREL,
90   R_MIPS_GOTREL,
91   R_MIPS_GOT_GP,
92   R_MIPS_GOT_GP_PC,
93   R_MIPS_GOT_LOCAL_PAGE,
94   R_MIPS_GOT_OFF,
95   R_MIPS_GOT_OFF32,
96   R_MIPS_TLSGD,
97   R_MIPS_TLSLD,
98   R_PPC32_PLTREL,
99   R_PPC64_CALL,
100   R_PPC64_CALL_PLT,
101   R_PPC64_RELAX_TOC,
102   R_PPC64_TOCBASE,
103   R_PPC64_RELAX_GOT_PC,
104   R_RISCV_ADD,
105   R_RISCV_PC_INDIRECT,
106 };
107 
108 // Architecture-neutral representation of relocation.
109 struct Relocation {
110   RelExpr expr;
111   RelType type;
112   uint64_t offset;
113   int64_t addend;
114   Symbol *sym;
115 };
116 
117 // Manipulate jump instructions with these modifiers.  These are used to relax
118 // jump instruction opcodes at basic block boundaries and are particularly
119 // useful when basic block sections are enabled.
120 struct JumpInstrMod {
121   uint64_t offset;
122   JumpModType original;
123   unsigned size;
124 };
125 
126 // This function writes undefined symbol diagnostics to an internal buffer.
127 // Call reportUndefinedSymbols() after calling scanRelocations() to emit
128 // the diagnostics.
129 template <class ELFT> void scanRelocations(InputSectionBase &);
130 void postScanRelocations();
131 
132 template <class ELFT> void reportUndefinedSymbols();
133 
134 void hexagonTLSSymbolUpdate(ArrayRef<OutputSection *> outputSections);
135 bool hexagonNeedsTLSSymbol(ArrayRef<OutputSection *> outputSections);
136 
137 class ThunkSection;
138 class Thunk;
139 class InputSectionDescription;
140 
141 class ThunkCreator {
142 public:
143   // Return true if Thunks have been added to OutputSections
144   bool createThunks(ArrayRef<OutputSection *> outputSections);
145 
146   // The number of completed passes of createThunks this permits us
147   // to do one time initialization on Pass 0 and put a limit on the
148   // number of times it can be called to prevent infinite loops.
149   uint32_t pass = 0;
150 
151 private:
152   void mergeThunks(ArrayRef<OutputSection *> outputSections);
153 
154   ThunkSection *getISDThunkSec(OutputSection *os, InputSection *isec,
155                                InputSectionDescription *isd,
156                                const Relocation &rel, uint64_t src);
157 
158   ThunkSection *getISThunkSec(InputSection *isec);
159 
160   void createInitialThunkSections(ArrayRef<OutputSection *> outputSections);
161 
162   std::pair<Thunk *, bool> getThunk(InputSection *isec, Relocation &rel,
163                                     uint64_t src);
164 
165   ThunkSection *addThunkSection(OutputSection *os, InputSectionDescription *,
166                                 uint64_t off);
167 
168   bool normalizeExistingThunk(Relocation &rel, uint64_t src);
169 
170   // Record all the available Thunks for a (Symbol, addend) pair, where Symbol
171   // is represented as a (section, offset) pair. There may be multiple
172   // relocations sharing the same (section, offset + addend) pair. We may revert
173   // a relocation back to its original non-Thunk target, and restore the
174   // original addend, so we cannot fold offset + addend. A nested pair is used
175   // because DenseMapInfo is not specialized for std::tuple.
176   llvm::DenseMap<std::pair<std::pair<SectionBase *, uint64_t>, int64_t>,
177                  std::vector<Thunk *>>
178       thunkedSymbolsBySectionAndAddend;
179   llvm::DenseMap<std::pair<Symbol *, int64_t>, std::vector<Thunk *>>
180       thunkedSymbols;
181 
182   // Find a Thunk from the Thunks symbol definition, we can use this to find
183   // the Thunk from a relocation to the Thunks symbol definition.
184   llvm::DenseMap<Symbol *, Thunk *> thunks;
185 
186   // Track InputSections that have an inline ThunkSection placed in front
187   // an inline ThunkSection may have control fall through to the section below
188   // so we need to make sure that there is only one of them.
189   // The Mips LA25 Thunk is an example of an inline ThunkSection.
190   llvm::DenseMap<InputSection *, ThunkSection *> thunkedSections;
191 };
192 
193 // Return a int64_t to make sure we get the sign extension out of the way as
194 // early as possible.
195 template <class ELFT>
196 static inline int64_t getAddend(const typename ELFT::Rel &rel) {
197   return 0;
198 }
199 template <class ELFT>
200 static inline int64_t getAddend(const typename ELFT::Rela &rel) {
201   return rel.r_addend;
202 }
203 
204 template <typename RelTy>
205 ArrayRef<RelTy> sortRels(ArrayRef<RelTy> rels, SmallVector<RelTy, 0> &storage) {
206   auto cmp = [](const RelTy &a, const RelTy &b) {
207     return a.r_offset < b.r_offset;
208   };
209   if (!llvm::is_sorted(rels, cmp)) {
210     storage.assign(rels.begin(), rels.end());
211     llvm::stable_sort(storage, cmp);
212     rels = storage;
213   }
214   return rels;
215 }
216 } // namespace elf
217 } // namespace lld
218 
219 #endif
220