xref: /freebsd/contrib/llvm-project/lld/ELF/Target.h (revision 1db9f3b21e39176dd5b67cf8ac378633b172463e)
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 std::string toString(elf::RelType type);
24 
25 namespace elf {
26 class Defined;
27 class InputFile;
28 class Symbol;
29 
30 class TargetInfo {
31 public:
32   virtual uint32_t calcEFlags() const { return 0; }
33   virtual RelExpr getRelExpr(RelType type, const Symbol &s,
34                              const uint8_t *loc) const = 0;
35   virtual RelType getDynRel(RelType type) const { return 0; }
36   virtual void writeGotPltHeader(uint8_t *buf) const {}
37   virtual void writeGotHeader(uint8_t *buf) const {}
38   virtual void writeGotPlt(uint8_t *buf, const Symbol &s) const {};
39   virtual void writeIgotPlt(uint8_t *buf, const Symbol &s) const {}
40   virtual int64_t getImplicitAddend(const uint8_t *buf, RelType type) const;
41   virtual int getTlsGdRelaxSkip(RelType type) const { return 1; }
42 
43   // If lazy binding is supported, the first entry of the PLT has code
44   // to call the dynamic linker to resolve PLT entries the first time
45   // they are called. This function writes that code.
46   virtual void writePltHeader(uint8_t *buf) const {}
47 
48   virtual void writePlt(uint8_t *buf, const Symbol &sym,
49                         uint64_t pltEntryAddr) const {}
50   virtual void writeIplt(uint8_t *buf, const Symbol &sym,
51                          uint64_t pltEntryAddr) const {
52     // All but PPC32 and PPC64 use the same format for .plt and .iplt entries.
53     writePlt(buf, sym, pltEntryAddr);
54   }
55   virtual void writeIBTPlt(uint8_t *buf, size_t numEntries) const {}
56   virtual void addPltHeaderSymbols(InputSection &isec) const {}
57   virtual void addPltSymbols(InputSection &isec, uint64_t off) const {}
58 
59   // Returns true if a relocation only uses the low bits of a value such that
60   // all those bits are in the same page. For example, if the relocation
61   // only uses the low 12 bits in a system with 4k pages. If this is true, the
62   // bits will always have the same value at runtime and we don't have to emit
63   // a dynamic relocation.
64   virtual bool usesOnlyLowPageBits(RelType type) const;
65 
66   // Decide whether a Thunk is needed for the relocation from File
67   // targeting S.
68   virtual bool needsThunk(RelExpr expr, RelType relocType,
69                           const InputFile *file, uint64_t branchAddr,
70                           const Symbol &s, int64_t a) const;
71 
72   // On systems with range extensions we place collections of Thunks at
73   // regular spacings that enable the majority of branches reach the Thunks.
74   // a value of 0 means range extension thunks are not supported.
75   virtual uint32_t getThunkSectionSpacing() const { return 0; }
76 
77   // The function with a prologue starting at Loc was compiled with
78   // -fsplit-stack and it calls a function compiled without. Adjust the prologue
79   // to do the right thing. See https://gcc.gnu.org/wiki/SplitStacks.
80   // The symbols st_other flags are needed on PowerPC64 for determining the
81   // offset to the split-stack prologue.
82   virtual bool adjustPrologueForCrossSplitStack(uint8_t *loc, uint8_t *end,
83                                                 uint8_t stOther) const;
84 
85   // Return true if we can reach dst from src with RelType type.
86   virtual bool inBranchRange(RelType type, uint64_t src,
87                              uint64_t dst) const;
88 
89   virtual void relocate(uint8_t *loc, const Relocation &rel,
90                         uint64_t val) const = 0;
91   void relocateNoSym(uint8_t *loc, RelType type, uint64_t val) const {
92     relocate(loc, Relocation{R_NONE, type, 0, 0, nullptr}, val);
93   }
94   virtual void relocateAlloc(InputSectionBase &sec, uint8_t *buf) const;
95 
96   // Do a linker relaxation pass and return true if we changed something.
97   virtual bool relaxOnce(int pass) const { return false; }
98 
99   virtual void applyJumpInstrMod(uint8_t *loc, JumpModType type,
100                                  JumpModType val) const {}
101 
102   virtual ~TargetInfo();
103 
104   // This deletes a jump insn at the end of the section if it is a fall thru to
105   // the next section.  Further, if there is a conditional jump and a direct
106   // jump consecutively, it tries to flip the conditional jump to convert the
107   // direct jump into a fall thru and delete it.  Returns true if a jump
108   // instruction can be deleted.
109   virtual bool deleteFallThruJmpInsn(InputSection &is, InputFile *file,
110                                      InputSection *nextIS) const {
111     return false;
112   }
113 
114   unsigned defaultCommonPageSize = 4096;
115   unsigned defaultMaxPageSize = 4096;
116 
117   uint64_t getImageBase() const;
118 
119   // True if _GLOBAL_OFFSET_TABLE_ is relative to .got.plt, false if .got.
120   bool gotBaseSymInGotPlt = false;
121 
122   static constexpr RelType noneRel = 0;
123   RelType copyRel;
124   RelType gotRel;
125   RelType pltRel;
126   RelType relativeRel;
127   RelType iRelativeRel;
128   RelType symbolicRel;
129   RelType tlsDescRel;
130   RelType tlsGotRel;
131   RelType tlsModuleIndexRel;
132   RelType tlsOffsetRel;
133   unsigned gotEntrySize = config->wordsize;
134   unsigned pltEntrySize;
135   unsigned pltHeaderSize;
136   unsigned ipltEntrySize;
137 
138   // At least on x86_64 positions 1 and 2 are used by the first plt entry
139   // to support lazy loading.
140   unsigned gotPltHeaderEntriesNum = 3;
141 
142   // On PPC ELF V2 abi, the first entry in the .got is the .TOC.
143   unsigned gotHeaderEntriesNum = 0;
144 
145   // On PPC ELF V2 abi, the dynamic section needs DT_PPC64_OPT (DT_LOPROC + 3)
146   // to be set to 0x2 if there can be multiple TOC's. Although we do not emit
147   // multiple TOC's, there can be a mix of TOC and NOTOC addressing which
148   // is functionally equivalent.
149   int ppc64DynamicSectionOpt = 0;
150 
151   bool needsThunks = false;
152 
153   // A 4-byte field corresponding to one or more trap instructions, used to pad
154   // executable OutputSections.
155   std::array<uint8_t, 4> trapInstr;
156 
157   // Stores the NOP instructions of different sizes for the target and is used
158   // to pad sections that are relaxed.
159   std::optional<std::vector<std::vector<uint8_t>>> nopInstrs;
160 
161   // If a target needs to rewrite calls to __morestack to instead call
162   // __morestack_non_split when a split-stack enabled caller calls a
163   // non-split-stack callee this will return true. Otherwise returns false.
164   bool needsMoreStackNonSplit = true;
165 
166   virtual RelExpr adjustTlsExpr(RelType type, RelExpr expr) const;
167   virtual RelExpr adjustGotPcExpr(RelType type, int64_t addend,
168                                   const uint8_t *loc) const;
169 
170 protected:
171   // On FreeBSD x86_64 the first page cannot be mmaped.
172   // On Linux this is controlled by vm.mmap_min_addr. At least on some x86_64
173   // installs this is set to 65536, so the first 15 pages cannot be used.
174   // Given that, the smallest value that can be used in here is 0x10000.
175   uint64_t defaultImageBase = 0x10000;
176 };
177 
178 TargetInfo *getAArch64TargetInfo();
179 TargetInfo *getAMDGPUTargetInfo();
180 TargetInfo *getARMTargetInfo();
181 TargetInfo *getAVRTargetInfo();
182 TargetInfo *getHexagonTargetInfo();
183 TargetInfo *getLoongArchTargetInfo();
184 TargetInfo *getMSP430TargetInfo();
185 TargetInfo *getPPC64TargetInfo();
186 TargetInfo *getPPCTargetInfo();
187 TargetInfo *getRISCVTargetInfo();
188 TargetInfo *getSPARCV9TargetInfo();
189 TargetInfo *getX86TargetInfo();
190 TargetInfo *getX86_64TargetInfo();
191 template <class ELFT> TargetInfo *getMipsTargetInfo();
192 
193 struct ErrorPlace {
194   InputSectionBase *isec;
195   std::string loc;
196   std::string srcLoc;
197 };
198 
199 // Returns input section and corresponding source string for the given location.
200 ErrorPlace getErrorPlace(const uint8_t *loc);
201 
202 static inline std::string getErrorLocation(const uint8_t *loc) {
203   return getErrorPlace(loc).loc;
204 }
205 
206 void processArmCmseSymbols();
207 
208 void writePPC32GlinkSection(uint8_t *buf, size_t numEntries);
209 
210 unsigned getPPCDFormOp(unsigned secondaryOp);
211 unsigned getPPCDSFormOp(unsigned secondaryOp);
212 
213 // In the PowerPC64 Elf V2 abi a function can have 2 entry points.  The first
214 // is a global entry point (GEP) which typically is used to initialize the TOC
215 // pointer in general purpose register 2.  The second is a local entry
216 // point (LEP) which bypasses the TOC pointer initialization code. The
217 // offset between GEP and LEP is encoded in a function's st_other flags.
218 // This function will return the offset (in bytes) from the global entry-point
219 // to the local entry-point.
220 unsigned getPPC64GlobalEntryToLocalEntryOffset(uint8_t stOther);
221 
222 // Write a prefixed instruction, which is a 4-byte prefix followed by a 4-byte
223 // instruction (regardless of endianness). Therefore, the prefix is always in
224 // lower memory than the instruction.
225 void writePrefixedInstruction(uint8_t *loc, uint64_t insn);
226 
227 void addPPC64SaveRestore();
228 uint64_t getPPC64TocBase();
229 uint64_t getAArch64Page(uint64_t expr);
230 template <typename ELFT> void writeARMCmseImportLib();
231 uint64_t getLoongArchPageDelta(uint64_t dest, uint64_t pc);
232 void riscvFinalizeRelax(int passes);
233 void mergeRISCVAttributesSections();
234 void addArmInputSectionMappingSymbols();
235 void addArmSyntheticSectionMappingSymbol(Defined *);
236 void sortArmMappingSymbols();
237 void convertArmInstructionstoBE8(InputSection *sec, uint8_t *buf);
238 void createTaggedSymbols(const SmallVector<ELFFileBase *, 0> &files);
239 
240 LLVM_LIBRARY_VISIBILITY extern const TargetInfo *target;
241 TargetInfo *getTarget();
242 
243 template <class ELFT> bool isMipsPIC(const Defined *sym);
244 
245 void reportRangeError(uint8_t *loc, const Relocation &rel, const Twine &v,
246                       int64_t min, uint64_t max);
247 void reportRangeError(uint8_t *loc, int64_t v, int n, const Symbol &sym,
248                       const Twine &msg);
249 
250 // Make sure that V can be represented as an N bit signed integer.
251 inline void checkInt(uint8_t *loc, int64_t v, int n, const Relocation &rel) {
252   if (v != llvm::SignExtend64(v, n))
253     reportRangeError(loc, rel, Twine(v), llvm::minIntN(n), llvm::maxIntN(n));
254 }
255 
256 // Make sure that V can be represented as an N bit unsigned integer.
257 inline void checkUInt(uint8_t *loc, uint64_t v, int n, const Relocation &rel) {
258   if ((v >> n) != 0)
259     reportRangeError(loc, rel, Twine(v), 0, llvm::maxUIntN(n));
260 }
261 
262 // Make sure that V can be represented as an N bit signed or unsigned integer.
263 inline void checkIntUInt(uint8_t *loc, uint64_t v, int n,
264                          const Relocation &rel) {
265   // For the error message we should cast V to a signed integer so that error
266   // messages show a small negative value rather than an extremely large one
267   if (v != (uint64_t)llvm::SignExtend64(v, n) && (v >> n) != 0)
268     reportRangeError(loc, rel, Twine((int64_t)v), llvm::minIntN(n),
269                      llvm::maxUIntN(n));
270 }
271 
272 inline void checkAlignment(uint8_t *loc, uint64_t v, int n,
273                            const Relocation &rel) {
274   if ((v & (n - 1)) != 0)
275     error(getErrorLocation(loc) + "improper alignment for relocation " +
276           lld::toString(rel.type) + ": 0x" + llvm::utohexstr(v) +
277           " is not aligned to " + Twine(n) + " bytes");
278 }
279 
280 // Endianness-aware read/write.
281 inline uint16_t read16(const void *p) {
282   return llvm::support::endian::read16(p, config->endianness);
283 }
284 
285 inline uint32_t read32(const void *p) {
286   return llvm::support::endian::read32(p, config->endianness);
287 }
288 
289 inline uint64_t read64(const void *p) {
290   return llvm::support::endian::read64(p, config->endianness);
291 }
292 
293 inline void write16(void *p, uint16_t v) {
294   llvm::support::endian::write16(p, v, config->endianness);
295 }
296 
297 inline void write32(void *p, uint32_t v) {
298   llvm::support::endian::write32(p, v, config->endianness);
299 }
300 
301 inline void write64(void *p, uint64_t v) {
302   llvm::support::endian::write64(p, v, config->endianness);
303 }
304 
305 // Overwrite a ULEB128 value and keep the original length.
306 inline uint64_t overwriteULEB128(uint8_t *bufLoc, uint64_t val) {
307   while (*bufLoc & 0x80) {
308     *bufLoc++ = 0x80 | (val & 0x7f);
309     val >>= 7;
310   }
311   *bufLoc = val;
312   return val;
313 }
314 } // namespace elf
315 } // namespace lld
316 
317 #ifdef __clang__
318 #pragma clang diagnostic ignored "-Wgnu-zero-variadic-macro-arguments"
319 #endif
320 #define invokeELFT(f, ...)                                                     \
321   switch (config->ekind) {                                                     \
322   case lld::elf::ELF32LEKind:                                                  \
323     f<llvm::object::ELF32LE>(__VA_ARGS__);                                     \
324     break;                                                                     \
325   case lld::elf::ELF32BEKind:                                                  \
326     f<llvm::object::ELF32BE>(__VA_ARGS__);                                     \
327     break;                                                                     \
328   case lld::elf::ELF64LEKind:                                                  \
329     f<llvm::object::ELF64LE>(__VA_ARGS__);                                     \
330     break;                                                                     \
331   case lld::elf::ELF64BEKind:                                                  \
332     f<llvm::object::ELF64BE>(__VA_ARGS__);                                     \
333     break;                                                                     \
334   default:                                                                     \
335     llvm_unreachable("unknown config->ekind");                                 \
336   }
337 
338 #endif
339