1 | //===- Target.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_TARGET_H |
10 | #define LLD_ELF_TARGET_H |
11 | |
12 | #include "Config.h" |
13 | #include "InputSection.h" |
14 | #include "lld/Common/ErrorHandler.h" |
15 | #include "llvm/ADT/StringExtras.h" |
16 | #include "llvm/Object/ELF.h" |
17 | #include "llvm/Object/ELFTypes.h" |
18 | #include "llvm/Support/Compiler.h" |
19 | #include "llvm/Support/MathExtras.h" |
20 | #include <array> |
21 | |
22 | namespace lld { |
23 | namespace elf { |
24 | class Defined; |
25 | class InputFile; |
26 | class Symbol; |
27 | |
28 | std::string toStr(Ctx &, RelType type); |
29 | |
30 | class TargetInfo { |
31 | public: |
32 | TargetInfo(Ctx &ctx) : ctx(ctx) {} |
33 | virtual uint32_t calcEFlags() const { return 0; } |
34 | virtual RelExpr getRelExpr(RelType type, const Symbol &s, |
35 | const uint8_t *loc) const = 0; |
36 | virtual RelType getDynRel(RelType type) const { return 0; } |
37 | virtual void (uint8_t *buf) const {} |
38 | virtual void (uint8_t *buf) const {} |
39 | virtual void writeGotPlt(uint8_t *buf, const Symbol &s) const {}; |
40 | virtual void writeIgotPlt(uint8_t *buf, const Symbol &s) const {} |
41 | virtual int64_t getImplicitAddend(const uint8_t *buf, RelType type) const; |
42 | virtual int getTlsGdRelaxSkip(RelType type) const { return 1; } |
43 | |
44 | // If lazy binding is supported, the first entry of the PLT has code |
45 | // to call the dynamic linker to resolve PLT entries the first time |
46 | // they are called. This function writes that code. |
47 | virtual void (uint8_t *buf) const {} |
48 | |
49 | virtual void writePlt(uint8_t *buf, const Symbol &sym, |
50 | uint64_t pltEntryAddr) const {} |
51 | virtual void writeIplt(uint8_t *buf, const Symbol &sym, |
52 | uint64_t pltEntryAddr) const { |
53 | // All but PPC32 and PPC64 use the same format for .plt and .iplt entries. |
54 | writePlt(buf, sym, pltEntryAddr); |
55 | } |
56 | virtual void writeIBTPlt(uint8_t *buf, size_t numEntries) const {} |
57 | virtual void (InputSection &isec) const {} |
58 | virtual void addPltSymbols(InputSection &isec, uint64_t off) const {} |
59 | |
60 | // Returns true if a relocation only uses the low bits of a value such that |
61 | // all those bits are in the same page. For example, if the relocation |
62 | // only uses the low 12 bits in a system with 4k pages. If this is true, the |
63 | // bits will always have the same value at runtime and we don't have to emit |
64 | // a dynamic relocation. |
65 | virtual bool usesOnlyLowPageBits(RelType type) const; |
66 | |
67 | // Decide whether a Thunk is needed for the relocation from File |
68 | // targeting S. |
69 | virtual bool needsThunk(RelExpr expr, RelType relocType, |
70 | const InputFile *file, uint64_t branchAddr, |
71 | const Symbol &s, int64_t a) const; |
72 | |
73 | // On systems with range extensions we place collections of Thunks at |
74 | // regular spacings that enable the majority of branches reach the Thunks. |
75 | // a value of 0 means range extension thunks are not supported. |
76 | virtual uint32_t getThunkSectionSpacing() const { return 0; } |
77 | |
78 | // The function with a prologue starting at Loc was compiled with |
79 | // -fsplit-stack and it calls a function compiled without. Adjust the prologue |
80 | // to do the right thing. See https://gcc.gnu.org/wiki/SplitStacks. |
81 | // The symbols st_other flags are needed on PowerPC64 for determining the |
82 | // offset to the split-stack prologue. |
83 | virtual bool adjustPrologueForCrossSplitStack(uint8_t *loc, uint8_t *end, |
84 | uint8_t stOther) const; |
85 | |
86 | // Return true if we can reach dst from src with RelType type. |
87 | virtual bool inBranchRange(RelType type, uint64_t src, |
88 | uint64_t dst) const; |
89 | |
90 | virtual void relocate(uint8_t *loc, const Relocation &rel, |
91 | uint64_t val) const = 0; |
92 | void relocateNoSym(uint8_t *loc, RelType type, uint64_t val) const { |
93 | relocate(loc, rel: Relocation{.expr: R_NONE, .type: type, .offset: 0, .addend: 0, .sym: nullptr}, val); |
94 | } |
95 | virtual void relocateAlloc(InputSectionBase &sec, uint8_t *buf) const; |
96 | |
97 | // Do a linker relaxation pass and return true if we changed something. |
98 | virtual bool relaxOnce(int pass) const { return false; } |
99 | // Do finalize relaxation after collecting relaxation infos. |
100 | virtual void finalizeRelax(int passes) const {} |
101 | |
102 | virtual void applyJumpInstrMod(uint8_t *loc, JumpModType type, |
103 | JumpModType val) const {} |
104 | virtual void applyBranchToBranchOpt() const {} |
105 | |
106 | virtual ~TargetInfo(); |
107 | |
108 | // This deletes a jump insn at the end of the section if it is a fall thru to |
109 | // the next section. Further, if there is a conditional jump and a direct |
110 | // jump consecutively, it tries to flip the conditional jump to convert the |
111 | // direct jump into a fall thru and delete it. Returns true if a jump |
112 | // instruction can be deleted. |
113 | virtual bool deleteFallThruJmpInsn(InputSection &is, InputFile *file, |
114 | InputSection *nextIS) const { |
115 | return false; |
116 | } |
117 | |
118 | Ctx &ctx; |
119 | unsigned defaultCommonPageSize = 4096; |
120 | unsigned defaultMaxPageSize = 4096; |
121 | |
122 | uint64_t getImageBase() const; |
123 | |
124 | // True if _GLOBAL_OFFSET_TABLE_ is relative to .got.plt, false if .got. |
125 | bool gotBaseSymInGotPlt = false; |
126 | |
127 | static constexpr RelType noneRel = 0; |
128 | RelType copyRel = 0; |
129 | RelType gotRel = 0; |
130 | RelType pltRel = 0; |
131 | RelType relativeRel = 0; |
132 | RelType iRelativeRel = 0; |
133 | RelType symbolicRel = 0; |
134 | RelType tlsDescRel = 0; |
135 | RelType tlsGotRel = 0; |
136 | RelType tlsModuleIndexRel = 0; |
137 | RelType tlsOffsetRel = 0; |
138 | unsigned gotEntrySize = ctx.arg.wordsize; |
139 | unsigned pltEntrySize = 0; |
140 | unsigned = 0; |
141 | unsigned ipltEntrySize = 0; |
142 | |
143 | // At least on x86_64 positions 1 and 2 are used by the first plt entry |
144 | // to support lazy loading. |
145 | unsigned = 3; |
146 | |
147 | // On PPC ELF V2 abi, the first entry in the .got is the .TOC. |
148 | unsigned = 0; |
149 | |
150 | // On PPC ELF V2 abi, the dynamic section needs DT_PPC64_OPT (DT_LOPROC + 3) |
151 | // to be set to 0x2 if there can be multiple TOC's. Although we do not emit |
152 | // multiple TOC's, there can be a mix of TOC and NOTOC addressing which |
153 | // is functionally equivalent. |
154 | int ppc64DynamicSectionOpt = 0; |
155 | |
156 | bool needsThunks = false; |
157 | |
158 | // A 4-byte field corresponding to one or more trap instructions, used to pad |
159 | // executable OutputSections. |
160 | std::array<uint8_t, 4> trapInstr = {}; |
161 | |
162 | // Stores the NOP instructions of different sizes for the target and is used |
163 | // to pad sections that are relaxed. |
164 | std::optional<std::vector<std::vector<uint8_t>>> nopInstrs; |
165 | |
166 | // If a target needs to rewrite calls to __morestack to instead call |
167 | // __morestack_non_split when a split-stack enabled caller calls a |
168 | // non-split-stack callee this will return true. Otherwise returns false. |
169 | bool needsMoreStackNonSplit = true; |
170 | |
171 | virtual RelExpr adjustTlsExpr(RelType type, RelExpr expr) const; |
172 | virtual RelExpr adjustGotPcExpr(RelType type, int64_t addend, |
173 | const uint8_t *loc) const; |
174 | |
175 | protected: |
176 | // On FreeBSD x86_64 the first page cannot be mmaped. |
177 | // On Linux this is controlled by vm.mmap_min_addr. At least on some x86_64 |
178 | // installs this is set to 65536, so the first 15 pages cannot be used. |
179 | // Given that, the smallest value that can be used in here is 0x10000. |
180 | uint64_t defaultImageBase = 0x10000; |
181 | }; |
182 | |
183 | void setAArch64TargetInfo(Ctx &); |
184 | void setAMDGPUTargetInfo(Ctx &); |
185 | void setARMTargetInfo(Ctx &); |
186 | void setAVRTargetInfo(Ctx &); |
187 | void setHexagonTargetInfo(Ctx &); |
188 | void setLoongArchTargetInfo(Ctx &); |
189 | void setMSP430TargetInfo(Ctx &); |
190 | void setMipsTargetInfo(Ctx &); |
191 | void setPPC64TargetInfo(Ctx &); |
192 | void setPPCTargetInfo(Ctx &); |
193 | void setRISCVTargetInfo(Ctx &); |
194 | void setSPARCV9TargetInfo(Ctx &); |
195 | void setSystemZTargetInfo(Ctx &); |
196 | void setX86TargetInfo(Ctx &); |
197 | void setX86_64TargetInfo(Ctx &); |
198 | |
199 | struct ErrorPlace { |
200 | InputSectionBase *isec; |
201 | std::string loc; |
202 | std::string srcLoc; |
203 | }; |
204 | |
205 | // Returns input section and corresponding source string for the given location. |
206 | ErrorPlace getErrorPlace(Ctx &ctx, const uint8_t *loc); |
207 | |
208 | static inline std::string getErrorLoc(Ctx &ctx, const uint8_t *loc) { |
209 | return getErrorPlace(ctx, loc).loc; |
210 | } |
211 | |
212 | void processArmCmseSymbols(Ctx &); |
213 | |
214 | template <class ELFT> uint32_t calcMipsEFlags(Ctx &); |
215 | uint8_t getMipsFpAbiFlag(Ctx &, InputFile *file, uint8_t oldFlag, |
216 | uint8_t newFlag); |
217 | bool isMipsN32Abi(Ctx &, const InputFile &f); |
218 | bool isMicroMips(Ctx &); |
219 | bool isMipsR6(Ctx &); |
220 | |
221 | void writePPC32GlinkSection(Ctx &, uint8_t *buf, size_t numEntries); |
222 | |
223 | unsigned getPPCDFormOp(unsigned secondaryOp); |
224 | unsigned getPPCDSFormOp(unsigned secondaryOp); |
225 | |
226 | // In the PowerPC64 Elf V2 abi a function can have 2 entry points. The first |
227 | // is a global entry point (GEP) which typically is used to initialize the TOC |
228 | // pointer in general purpose register 2. The second is a local entry |
229 | // point (LEP) which bypasses the TOC pointer initialization code. The |
230 | // offset between GEP and LEP is encoded in a function's st_other flags. |
231 | // This function will return the offset (in bytes) from the global entry-point |
232 | // to the local entry-point. |
233 | unsigned getPPC64GlobalEntryToLocalEntryOffset(Ctx &, uint8_t stOther); |
234 | |
235 | // Write a prefixed instruction, which is a 4-byte prefix followed by a 4-byte |
236 | // instruction (regardless of endianness). Therefore, the prefix is always in |
237 | // lower memory than the instruction. |
238 | void writePrefixedInst(Ctx &, uint8_t *loc, uint64_t insn); |
239 | |
240 | void addPPC64SaveRestore(Ctx &); |
241 | uint64_t getPPC64TocBase(Ctx &ctx); |
242 | uint64_t getAArch64Page(uint64_t expr); |
243 | bool isAArch64BTILandingPad(Ctx &, Symbol &s, int64_t a); |
244 | template <typename ELFT> void writeARMCmseImportLib(Ctx &); |
245 | uint64_t getLoongArchPageDelta(uint64_t dest, uint64_t pc, RelType type); |
246 | void riscvFinalizeRelax(int passes); |
247 | void mergeRISCVAttributesSections(Ctx &); |
248 | void addArmInputSectionMappingSymbols(Ctx &); |
249 | void addArmSyntheticSectionMappingSymbol(Defined *); |
250 | void sortArmMappingSymbols(Ctx &); |
251 | void convertArmInstructionstoBE8(Ctx &, InputSection *sec, uint8_t *buf); |
252 | void createTaggedSymbols(Ctx &); |
253 | void initSymbolAnchors(Ctx &); |
254 | |
255 | void setTarget(Ctx &); |
256 | |
257 | template <class ELFT> bool isMipsPIC(const Defined *sym); |
258 | |
259 | const ELFSyncStream &operator<<(const ELFSyncStream &, RelType); |
260 | |
261 | void reportRangeError(Ctx &, uint8_t *loc, const Relocation &rel, |
262 | const Twine &v, int64_t min, uint64_t max); |
263 | void reportRangeError(Ctx &ctx, uint8_t *loc, int64_t v, int n, |
264 | const Symbol &sym, const Twine &msg); |
265 | |
266 | // Make sure that V can be represented as an N bit signed integer. |
267 | inline void checkInt(Ctx &ctx, uint8_t *loc, int64_t v, int n, |
268 | const Relocation &rel) { |
269 | if (v != llvm::SignExtend64(X: v, B: n)) |
270 | reportRangeError(ctx, loc, rel, v: Twine(v), min: llvm::minIntN(N: n), |
271 | max: llvm::maxIntN(N: n)); |
272 | } |
273 | |
274 | // Make sure that V can be represented as an N bit unsigned integer. |
275 | inline void checkUInt(Ctx &ctx, uint8_t *loc, uint64_t v, int n, |
276 | const Relocation &rel) { |
277 | if ((v >> n) != 0) |
278 | reportRangeError(ctx, loc, rel, v: Twine(v), min: 0, max: llvm::maxUIntN(N: n)); |
279 | } |
280 | |
281 | // Make sure that V can be represented as an N bit signed or unsigned integer. |
282 | inline void checkIntUInt(Ctx &ctx, uint8_t *loc, uint64_t v, int n, |
283 | const Relocation &rel) { |
284 | // For the error message we should cast V to a signed integer so that error |
285 | // messages show a small negative value rather than an extremely large one |
286 | if (v != (uint64_t)llvm::SignExtend64(X: v, B: n) && (v >> n) != 0) |
287 | reportRangeError(ctx, loc, rel, v: Twine((int64_t)v), min: llvm::minIntN(N: n), |
288 | max: llvm::maxUIntN(N: n)); |
289 | } |
290 | |
291 | inline void checkAlignment(Ctx &ctx, uint8_t *loc, uint64_t v, int n, |
292 | const Relocation &rel) { |
293 | if ((v & (n - 1)) != 0) |
294 | Err(ctx) << getErrorLoc(ctx, loc) << "improper alignment for relocation " |
295 | << rel.type << ": 0x" << llvm::utohexstr(X: v) |
296 | << " is not aligned to " << n << " bytes" ; |
297 | } |
298 | |
299 | // Endianness-aware read/write. |
300 | inline uint16_t read16(Ctx &ctx, const void *p) { |
301 | return llvm::support::endian::read16(P: p, E: ctx.arg.endianness); |
302 | } |
303 | |
304 | inline uint32_t read32(Ctx &ctx, const void *p) { |
305 | return llvm::support::endian::read32(P: p, E: ctx.arg.endianness); |
306 | } |
307 | |
308 | inline uint64_t read64(Ctx &ctx, const void *p) { |
309 | return llvm::support::endian::read64(P: p, E: ctx.arg.endianness); |
310 | } |
311 | |
312 | inline void write16(Ctx &ctx, void *p, uint16_t v) { |
313 | llvm::support::endian::write16(P: p, V: v, E: ctx.arg.endianness); |
314 | } |
315 | |
316 | inline void write32(Ctx &ctx, void *p, uint32_t v) { |
317 | llvm::support::endian::write32(P: p, V: v, E: ctx.arg.endianness); |
318 | } |
319 | |
320 | inline void write64(Ctx &ctx, void *p, uint64_t v) { |
321 | llvm::support::endian::write64(P: p, V: v, E: ctx.arg.endianness); |
322 | } |
323 | |
324 | // Overwrite a ULEB128 value and keep the original length. |
325 | inline uint64_t overwriteULEB128(uint8_t *bufLoc, uint64_t val) { |
326 | while (*bufLoc & 0x80) { |
327 | *bufLoc++ = 0x80 | (val & 0x7f); |
328 | val >>= 7; |
329 | } |
330 | *bufLoc = val; |
331 | return val; |
332 | } |
333 | } // namespace elf |
334 | } // namespace lld |
335 | |
336 | #ifdef __clang__ |
337 | #pragma clang diagnostic ignored "-Wgnu-zero-variadic-macro-arguments" |
338 | #endif |
339 | #define invokeELFT(f, ...) \ |
340 | switch (ctx.arg.ekind) { \ |
341 | case lld::elf::ELF32LEKind: \ |
342 | f<llvm::object::ELF32LE>(__VA_ARGS__); \ |
343 | break; \ |
344 | case lld::elf::ELF32BEKind: \ |
345 | f<llvm::object::ELF32BE>(__VA_ARGS__); \ |
346 | break; \ |
347 | case lld::elf::ELF64LEKind: \ |
348 | f<llvm::object::ELF64LE>(__VA_ARGS__); \ |
349 | break; \ |
350 | case lld::elf::ELF64BEKind: \ |
351 | f<llvm::object::ELF64BE>(__VA_ARGS__); \ |
352 | break; \ |
353 | default: \ |
354 | llvm_unreachable("unknown ctx.arg.ekind"); \ |
355 | } |
356 | |
357 | #endif |
358 | |