| 1 | //===- DLL.cpp ------------------------------------------------------------===// |
| 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 | // This file defines various types of chunks for the DLL import or export |
| 10 | // descriptor tables. They are inherently Windows-specific. |
| 11 | // You need to read Microsoft PE/COFF spec to understand details |
| 12 | // about the data structures. |
| 13 | // |
| 14 | // If you are not particularly interested in linking against Windows |
| 15 | // DLL, you can skip this file, and you should still be able to |
| 16 | // understand the rest of the linker. |
| 17 | // |
| 18 | //===----------------------------------------------------------------------===// |
| 19 | |
| 20 | #include "DLL.h" |
| 21 | #include "COFFLinkerContext.h" |
| 22 | #include "Chunks.h" |
| 23 | #include "SymbolTable.h" |
| 24 | #include "llvm/ADT/STLExtras.h" |
| 25 | #include "llvm/Object/COFF.h" |
| 26 | #include "llvm/Support/Endian.h" |
| 27 | #include "llvm/Support/Path.h" |
| 28 | |
| 29 | using namespace llvm; |
| 30 | using namespace llvm::object; |
| 31 | using namespace llvm::support::endian; |
| 32 | using namespace llvm::COFF; |
| 33 | |
| 34 | namespace lld::coff { |
| 35 | namespace { |
| 36 | |
| 37 | // Import table |
| 38 | |
| 39 | // A chunk for the import descriptor table. |
| 40 | class HintNameChunk : public NonSectionChunk { |
| 41 | public: |
| 42 | HintNameChunk(StringRef n, uint16_t h) : name(n), hint(h) {} |
| 43 | |
| 44 | size_t getSize() const override { |
| 45 | // Starts with 2 byte Hint field, followed by a null-terminated string, |
| 46 | // ends with 0 or 1 byte padding. |
| 47 | return alignTo(Value: name.size() + 3, Align: 2); |
| 48 | } |
| 49 | |
| 50 | void writeTo(uint8_t *buf) const override { |
| 51 | memset(s: buf, c: 0, n: getSize()); |
| 52 | write16le(P: buf, V: hint); |
| 53 | memcpy(dest: buf + 2, src: name.data(), n: name.size()); |
| 54 | } |
| 55 | |
| 56 | private: |
| 57 | StringRef name; |
| 58 | uint16_t hint; |
| 59 | }; |
| 60 | |
| 61 | // A chunk for the import descriptor table. |
| 62 | class LookupChunk : public NonSectionChunk { |
| 63 | public: |
| 64 | explicit LookupChunk(COFFLinkerContext &ctx, Chunk *c) |
| 65 | : hintName(c), ctx(ctx) { |
| 66 | setAlignment(ctx.config.wordsize); |
| 67 | } |
| 68 | size_t getSize() const override { return ctx.config.wordsize; } |
| 69 | |
| 70 | void writeTo(uint8_t *buf) const override { |
| 71 | if (ctx.config.is64()) |
| 72 | write64le(P: buf, V: hintName->getRVA()); |
| 73 | else |
| 74 | write32le(P: buf, V: hintName->getRVA()); |
| 75 | } |
| 76 | |
| 77 | Chunk *hintName; |
| 78 | |
| 79 | private: |
| 80 | COFFLinkerContext &ctx; |
| 81 | }; |
| 82 | |
| 83 | // A chunk for the import descriptor table. |
| 84 | // This chunk represent import-by-ordinal symbols. |
| 85 | // See Microsoft PE/COFF spec 7.1. Import Header for details. |
| 86 | class OrdinalOnlyChunk : public NonSectionChunk { |
| 87 | public: |
| 88 | explicit OrdinalOnlyChunk(COFFLinkerContext &c, uint16_t v) |
| 89 | : ordinal(v), ctx(c) { |
| 90 | setAlignment(ctx.config.wordsize); |
| 91 | } |
| 92 | size_t getSize() const override { return ctx.config.wordsize; } |
| 93 | |
| 94 | void writeTo(uint8_t *buf) const override { |
| 95 | // An import-by-ordinal slot has MSB 1 to indicate that |
| 96 | // this is import-by-ordinal (and not import-by-name). |
| 97 | if (ctx.config.is64()) { |
| 98 | write64le(P: buf, V: (1ULL << 63) | ordinal); |
| 99 | } else { |
| 100 | write32le(P: buf, V: (1ULL << 31) | ordinal); |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | uint16_t ordinal; |
| 105 | |
| 106 | private: |
| 107 | COFFLinkerContext &ctx; |
| 108 | }; |
| 109 | |
| 110 | // A chunk for the import descriptor table. |
| 111 | class ImportDirectoryChunk : public NonSectionChunk { |
| 112 | public: |
| 113 | explicit ImportDirectoryChunk(Chunk *n) : dllName(n) { setAlignment(4); } |
| 114 | size_t getSize() const override { return sizeof(ImportDirectoryTableEntry); } |
| 115 | |
| 116 | void writeTo(uint8_t *buf) const override { |
| 117 | memset(s: buf, c: 0, n: getSize()); |
| 118 | |
| 119 | auto *e = (coff_import_directory_table_entry *)(buf); |
| 120 | e->ImportLookupTableRVA = lookupTab->getRVA(); |
| 121 | e->NameRVA = dllName->getRVA(); |
| 122 | e->ImportAddressTableRVA = addressTab->getRVA(); |
| 123 | } |
| 124 | |
| 125 | Chunk *dllName; |
| 126 | Chunk *lookupTab; |
| 127 | Chunk *addressTab; |
| 128 | }; |
| 129 | |
| 130 | // A chunk representing null terminator in the import table. |
| 131 | // Contents of this chunk is always null bytes. |
| 132 | class NullChunk : public NonSectionChunk { |
| 133 | public: |
| 134 | explicit NullChunk(size_t n, uint32_t align) : size(n) { |
| 135 | setAlignment(align); |
| 136 | } |
| 137 | explicit NullChunk(COFFLinkerContext &ctx) |
| 138 | : NullChunk(ctx.config.wordsize, ctx.config.wordsize) {} |
| 139 | explicit NullChunk(COFFLinkerContext &ctx, size_t n) |
| 140 | : NullChunk(n, ctx.config.wordsize) {} |
| 141 | size_t getSize() const override { return size; } |
| 142 | |
| 143 | void writeTo(uint8_t *buf) const override { |
| 144 | memset(s: buf, c: 0, n: size); |
| 145 | } |
| 146 | |
| 147 | private: |
| 148 | size_t size; |
| 149 | }; |
| 150 | |
| 151 | // A chunk for ARM64EC auxiliary IAT. |
| 152 | class AuxImportChunk : public NonSectionChunk { |
| 153 | public: |
| 154 | explicit AuxImportChunk(ImportFile *file) : file(file) { |
| 155 | setAlignment(sizeof(uint64_t)); |
| 156 | } |
| 157 | size_t getSize() const override { return sizeof(uint64_t); } |
| 158 | |
| 159 | void writeTo(uint8_t *buf) const override { |
| 160 | uint64_t impchkVA = 0; |
| 161 | if (file->impchkThunk) |
| 162 | impchkVA = |
| 163 | file->impchkThunk->getRVA() + file->symtab.ctx.config.imageBase; |
| 164 | write64le(P: buf, V: impchkVA); |
| 165 | } |
| 166 | |
| 167 | void getBaserels(std::vector<Baserel> *res) override { |
| 168 | if (file->impchkThunk) |
| 169 | res->emplace_back(args&: rva, args&: file->symtab.machine); |
| 170 | } |
| 171 | |
| 172 | private: |
| 173 | ImportFile *file; |
| 174 | }; |
| 175 | |
| 176 | static std::vector<std::vector<DefinedImportData *>> |
| 177 | binImports(COFFLinkerContext &ctx, |
| 178 | const std::vector<DefinedImportData *> &imports) { |
| 179 | // Group DLL-imported symbols by DLL name because that's how |
| 180 | // symbols are laid out in the import descriptor table. |
| 181 | auto less = [&ctx](const std::string &a, const std::string &b) { |
| 182 | return ctx.config.dllOrder[a] < ctx.config.dllOrder[b]; |
| 183 | }; |
| 184 | std::map<std::string, std::vector<DefinedImportData *>, decltype(less)> m( |
| 185 | less); |
| 186 | for (DefinedImportData *sym : imports) |
| 187 | m[sym->getDLLName().lower()].push_back(x: sym); |
| 188 | |
| 189 | std::vector<std::vector<DefinedImportData *>> v; |
| 190 | for (auto &kv : m) { |
| 191 | // Sort symbols by name for each group. |
| 192 | std::vector<DefinedImportData *> &syms = kv.second; |
| 193 | llvm::sort(C&: syms, Comp: [](DefinedImportData *a, DefinedImportData *b) { |
| 194 | auto getBaseName = [](DefinedImportData *sym) { |
| 195 | StringRef name = sym->getName(); |
| 196 | name.consume_front(Prefix: "__imp_" ); |
| 197 | // Skip aux_ part of ARM64EC function symbol name. |
| 198 | if (sym->file->impchkThunk) |
| 199 | name.consume_front(Prefix: "aux_" ); |
| 200 | return name; |
| 201 | }; |
| 202 | return getBaseName(a) < getBaseName(b); |
| 203 | }); |
| 204 | v.push_back(x: std::move(syms)); |
| 205 | } |
| 206 | return v; |
| 207 | } |
| 208 | |
| 209 | // See Microsoft PE/COFF spec 4.3 for details. |
| 210 | |
| 211 | // A chunk for the delay import descriptor table etnry. |
| 212 | class DelayDirectoryChunk : public NonSectionChunk { |
| 213 | public: |
| 214 | explicit DelayDirectoryChunk(Chunk *n) : dllName(n) { setAlignment(4); } |
| 215 | |
| 216 | size_t getSize() const override { |
| 217 | return sizeof(delay_import_directory_table_entry); |
| 218 | } |
| 219 | |
| 220 | void writeTo(uint8_t *buf) const override { |
| 221 | memset(s: buf, c: 0, n: getSize()); |
| 222 | |
| 223 | auto *e = (delay_import_directory_table_entry *)(buf); |
| 224 | e->Attributes = 1; |
| 225 | e->Name = dllName->getRVA(); |
| 226 | e->ModuleHandle = moduleHandle->getRVA(); |
| 227 | e->DelayImportAddressTable = addressTab->getRVA(); |
| 228 | e->DelayImportNameTable = nameTab->getRVA(); |
| 229 | } |
| 230 | |
| 231 | Chunk *dllName; |
| 232 | Chunk *moduleHandle; |
| 233 | Chunk *addressTab; |
| 234 | Chunk *nameTab; |
| 235 | }; |
| 236 | |
| 237 | // Initial contents for delay-loaded functions. |
| 238 | // This code calls __delayLoadHelper2 function to resolve a symbol |
| 239 | // which then overwrites its jump table slot with the result |
| 240 | // for subsequent function calls. |
| 241 | static const uint8_t thunkX64[] = { |
| 242 | 0x48, 0x8D, 0x05, 0, 0, 0, 0, // lea rax, [__imp_<FUNCNAME>] |
| 243 | 0xE9, 0, 0, 0, 0, // jmp __tailMerge_<lib> |
| 244 | }; |
| 245 | |
| 246 | static const uint8_t tailMergeX64[] = { |
| 247 | 0x48, 0x89, 0x4C, 0x24, 0x08, // mov qword ptr [rsp+8], rcx |
| 248 | 0x48, 0x89, 0x54, 0x24, 0x10, // mov qword ptr [rsp+10h], rdx |
| 249 | 0x4C, 0x89, 0x44, 0x24, 0x18, // mov qword ptr [rsp+18h], r8 |
| 250 | 0x4C, 0x89, 0x4C, 0x24, 0x20, // mov qword ptr [rsp+20h], r9 |
| 251 | 0x48, 0x83, 0xEC, 0x68, // sub rsp, 68h |
| 252 | 0x66, 0x0F, 0x7F, 0x44, 0x24, 0x20, // movdqa xmmword ptr [rsp+20h], xmm0 |
| 253 | 0x66, 0x0F, 0x7F, 0x4C, 0x24, 0x30, // movdqa xmmword ptr [rsp+30h], xmm1 |
| 254 | 0x66, 0x0F, 0x7F, 0x54, 0x24, 0x40, // movdqa xmmword ptr [rsp+40h], xmm2 |
| 255 | 0x66, 0x0F, 0x7F, 0x5C, 0x24, 0x50, // movdqa xmmword ptr [rsp+50h], xmm3 |
| 256 | 0x48, 0x8B, 0xD0, // mov rdx, rax |
| 257 | 0x48, 0x8D, 0x0D, 0, 0, 0, 0, // lea rcx, [___DELAY_IMPORT_...] |
| 258 | 0xE8, 0, 0, 0, 0, // call __delayLoadHelper2 |
| 259 | 0x66, 0x0F, 0x6F, 0x44, 0x24, 0x20, // movdqa xmm0, xmmword ptr [rsp+20h] |
| 260 | 0x66, 0x0F, 0x6F, 0x4C, 0x24, 0x30, // movdqa xmm1, xmmword ptr [rsp+30h] |
| 261 | 0x66, 0x0F, 0x6F, 0x54, 0x24, 0x40, // movdqa xmm2, xmmword ptr [rsp+40h] |
| 262 | 0x66, 0x0F, 0x6F, 0x5C, 0x24, 0x50, // movdqa xmm3, xmmword ptr [rsp+50h] |
| 263 | 0x48, 0x8B, 0x4C, 0x24, 0x70, // mov rcx, qword ptr [rsp+70h] |
| 264 | 0x48, 0x8B, 0x54, 0x24, 0x78, // mov rdx, qword ptr [rsp+78h] |
| 265 | 0x4C, 0x8B, 0x84, 0x24, 0x80, 0, 0, 0, // mov r8, qword ptr [rsp+80h] |
| 266 | 0x4C, 0x8B, 0x8C, 0x24, 0x88, 0, 0, 0, // mov r9, qword ptr [rsp+88h] |
| 267 | 0x48, 0x83, 0xC4, 0x68, // add rsp, 68h |
| 268 | 0xFF, 0xE0, // jmp rax |
| 269 | }; |
| 270 | |
| 271 | static const uint8_t tailMergeUnwindInfoX64[] = { |
| 272 | 0x01, // Version=1, Flags=UNW_FLAG_NHANDLER |
| 273 | 0x18, // Size of prolog |
| 274 | 0x01, // Count of unwind codes |
| 275 | 0x00, // No frame register |
| 276 | 0x18, 0xC2, // Offset 0x18: UWOP_ALLOC_SMALL(0x68) |
| 277 | 0x00, 0x00 // Padding to align on 32-bits |
| 278 | }; |
| 279 | |
| 280 | static const uint8_t thunkX86[] = { |
| 281 | 0xB8, 0, 0, 0, 0, // mov eax, offset ___imp__<FUNCNAME> |
| 282 | 0xE9, 0, 0, 0, 0, // jmp __tailMerge_<lib> |
| 283 | }; |
| 284 | |
| 285 | static const uint8_t tailMergeX86[] = { |
| 286 | 0x51, // push ecx |
| 287 | 0x52, // push edx |
| 288 | 0x50, // push eax |
| 289 | 0x68, 0, 0, 0, 0, // push offset ___DELAY_IMPORT_DESCRIPTOR_<DLLNAME>_dll |
| 290 | 0xE8, 0, 0, 0, 0, // call ___delayLoadHelper2@8 |
| 291 | 0x5A, // pop edx |
| 292 | 0x59, // pop ecx |
| 293 | 0xFF, 0xE0, // jmp eax |
| 294 | }; |
| 295 | |
| 296 | static const uint8_t thunkARM[] = { |
| 297 | 0x40, 0xf2, 0x00, 0x0c, // mov.w ip, #0 __imp_<FUNCNAME> |
| 298 | 0xc0, 0xf2, 0x00, 0x0c, // mov.t ip, #0 __imp_<FUNCNAME> |
| 299 | 0x00, 0xf0, 0x00, 0xb8, // b.w __tailMerge_<lib> |
| 300 | }; |
| 301 | |
| 302 | static const uint8_t tailMergeARM[] = { |
| 303 | 0x2d, 0xe9, 0x0f, 0x48, // push.w {r0, r1, r2, r3, r11, lr} |
| 304 | 0x0d, 0xf2, 0x10, 0x0b, // addw r11, sp, #16 |
| 305 | 0x2d, 0xed, 0x10, 0x0b, // vpush {d0, d1, d2, d3, d4, d5, d6, d7} |
| 306 | 0x61, 0x46, // mov r1, ip |
| 307 | 0x40, 0xf2, 0x00, 0x00, // mov.w r0, #0 DELAY_IMPORT_DESCRIPTOR |
| 308 | 0xc0, 0xf2, 0x00, 0x00, // mov.t r0, #0 DELAY_IMPORT_DESCRIPTOR |
| 309 | 0x00, 0xf0, 0x00, 0xd0, // bl #0 __delayLoadHelper2 |
| 310 | 0x84, 0x46, // mov ip, r0 |
| 311 | 0xbd, 0xec, 0x10, 0x0b, // vpop {d0, d1, d2, d3, d4, d5, d6, d7} |
| 312 | 0xbd, 0xe8, 0x0f, 0x48, // pop.w {r0, r1, r2, r3, r11, lr} |
| 313 | 0x60, 0x47, // bx ip |
| 314 | }; |
| 315 | |
| 316 | static const uint8_t thunkARM64[] = { |
| 317 | 0x11, 0x00, 0x00, 0x90, // adrp x17, #0 __imp_<FUNCNAME> |
| 318 | 0x31, 0x02, 0x00, 0x91, // add x17, x17, #0 :lo12:__imp_<FUNCNAME> |
| 319 | 0x00, 0x00, 0x00, 0x14, // b __tailMerge_<lib> |
| 320 | }; |
| 321 | |
| 322 | static const uint8_t tailMergeARM64[] = { |
| 323 | 0xfd, 0x7b, 0xb2, 0xa9, // stp x29, x30, [sp, #-224]! |
| 324 | 0xfd, 0x03, 0x00, 0x91, // mov x29, sp |
| 325 | 0xe0, 0x07, 0x01, 0xa9, // stp x0, x1, [sp, #16] |
| 326 | 0xe2, 0x0f, 0x02, 0xa9, // stp x2, x3, [sp, #32] |
| 327 | 0xe4, 0x17, 0x03, 0xa9, // stp x4, x5, [sp, #48] |
| 328 | 0xe6, 0x1f, 0x04, 0xa9, // stp x6, x7, [sp, #64] |
| 329 | 0xe8, 0x2b, 0x00, 0xf9, // str x8, [sp, #80] |
| 330 | 0xe0, 0x07, 0x03, 0xad, // stp q0, q1, [sp, #96] |
| 331 | 0xe2, 0x0f, 0x04, 0xad, // stp q2, q3, [sp, #128] |
| 332 | 0xe4, 0x17, 0x05, 0xad, // stp q4, q5, [sp, #160] |
| 333 | 0xe6, 0x1f, 0x06, 0xad, // stp q6, q7, [sp, #192] |
| 334 | 0xe1, 0x03, 0x11, 0xaa, // mov x1, x17 |
| 335 | 0x00, 0x00, 0x00, 0x90, // adrp x0, #0 DELAY_IMPORT_DESCRIPTOR |
| 336 | 0x00, 0x00, 0x00, 0x91, // add x0, x0, #0 :lo12:DELAY_IMPORT_DESCRIPTOR |
| 337 | 0x02, 0x00, 0x00, 0x90, // adrp x2, #0 __delayLoadHelper2 |
| 338 | 0x42, 0x00, 0x00, 0x91, // add x2, x2, #0 :lo12:__delayLoadHelper2 |
| 339 | 0x40, 0x00, 0x3f, 0xd6, // blr x2 |
| 340 | 0xf0, 0x03, 0x00, 0xaa, // mov x16, x0 |
| 341 | 0xe6, 0x1f, 0x46, 0xad, // ldp q6, q7, [sp, #192] |
| 342 | 0xe4, 0x17, 0x45, 0xad, // ldp q4, q5, [sp, #160] |
| 343 | 0xe2, 0x0f, 0x44, 0xad, // ldp q2, q3, [sp, #128] |
| 344 | 0xe0, 0x07, 0x43, 0xad, // ldp q0, q1, [sp, #96] |
| 345 | 0xe8, 0x2b, 0x40, 0xf9, // ldr x8, [sp, #80] |
| 346 | 0xe6, 0x1f, 0x44, 0xa9, // ldp x6, x7, [sp, #64] |
| 347 | 0xe4, 0x17, 0x43, 0xa9, // ldp x4, x5, [sp, #48] |
| 348 | 0xe2, 0x0f, 0x42, 0xa9, // ldp x2, x3, [sp, #32] |
| 349 | 0xe0, 0x07, 0x41, 0xa9, // ldp x0, x1, [sp, #16] |
| 350 | 0xfd, 0x7b, 0xce, 0xa8, // ldp x29, x30, [sp], #224 |
| 351 | 0x00, 0x02, 0x1f, 0xd6, // br x16 |
| 352 | }; |
| 353 | |
| 354 | // A chunk for the delay import thunk. |
| 355 | class ThunkChunkX64 : public NonSectionCodeChunk { |
| 356 | public: |
| 357 | ThunkChunkX64(Defined *i, Chunk *tm) : imp(i), tailMerge(tm) {} |
| 358 | |
| 359 | size_t getSize() const override { return sizeof(thunkX64); } |
| 360 | MachineTypes getMachine() const override { return AMD64; } |
| 361 | |
| 362 | void writeTo(uint8_t *buf) const override { |
| 363 | memcpy(dest: buf, src: thunkX64, n: sizeof(thunkX64)); |
| 364 | write32le(P: buf + 3, V: imp->getRVA() - rva - 7); |
| 365 | write32le(P: buf + 8, V: tailMerge->getRVA() - rva - 12); |
| 366 | } |
| 367 | |
| 368 | Defined *imp = nullptr; |
| 369 | Chunk *tailMerge = nullptr; |
| 370 | }; |
| 371 | |
| 372 | class TailMergeChunkX64 : public NonSectionCodeChunk { |
| 373 | public: |
| 374 | TailMergeChunkX64(Chunk *d, Defined *h) : desc(d), helper(h) {} |
| 375 | |
| 376 | size_t getSize() const override { return sizeof(tailMergeX64); } |
| 377 | MachineTypes getMachine() const override { return AMD64; } |
| 378 | |
| 379 | void writeTo(uint8_t *buf) const override { |
| 380 | memcpy(dest: buf, src: tailMergeX64, n: sizeof(tailMergeX64)); |
| 381 | write32le(P: buf + 54, V: desc->getRVA() - rva - 58); |
| 382 | write32le(P: buf + 59, V: helper->getRVA() - rva - 63); |
| 383 | } |
| 384 | |
| 385 | Chunk *desc = nullptr; |
| 386 | Defined *helper = nullptr; |
| 387 | }; |
| 388 | |
| 389 | class TailMergePDataChunkX64 : public NonSectionChunk { |
| 390 | public: |
| 391 | TailMergePDataChunkX64(Chunk *tm, Chunk *unwind) : tm(tm), unwind(unwind) { |
| 392 | // See |
| 393 | // https://learn.microsoft.com/en-us/cpp/build/exception-handling-x64#struct-runtime_function |
| 394 | setAlignment(4); |
| 395 | } |
| 396 | |
| 397 | size_t getSize() const override { return 3 * sizeof(uint32_t); } |
| 398 | MachineTypes getMachine() const override { return AMD64; } |
| 399 | |
| 400 | void writeTo(uint8_t *buf) const override { |
| 401 | write32le(P: buf + 0, V: tm->getRVA()); // TailMergeChunk start RVA |
| 402 | write32le(P: buf + 4, V: tm->getRVA() + tm->getSize()); // TailMergeChunk stop RVA |
| 403 | write32le(P: buf + 8, V: unwind->getRVA()); // UnwindInfo RVA |
| 404 | } |
| 405 | |
| 406 | Chunk *tm = nullptr; |
| 407 | Chunk *unwind = nullptr; |
| 408 | }; |
| 409 | |
| 410 | class TailMergeUnwindInfoX64 : public NonSectionChunk { |
| 411 | public: |
| 412 | TailMergeUnwindInfoX64() { |
| 413 | // See |
| 414 | // https://learn.microsoft.com/en-us/cpp/build/exception-handling-x64#struct-unwind_info |
| 415 | setAlignment(4); |
| 416 | } |
| 417 | |
| 418 | size_t getSize() const override { return sizeof(tailMergeUnwindInfoX64); } |
| 419 | MachineTypes getMachine() const override { return AMD64; } |
| 420 | |
| 421 | void writeTo(uint8_t *buf) const override { |
| 422 | memcpy(dest: buf, src: tailMergeUnwindInfoX64, n: sizeof(tailMergeUnwindInfoX64)); |
| 423 | } |
| 424 | }; |
| 425 | |
| 426 | class ThunkChunkX86 : public NonSectionCodeChunk { |
| 427 | public: |
| 428 | ThunkChunkX86(COFFLinkerContext &ctx, Defined *i, Chunk *tm) |
| 429 | : imp(i), tailMerge(tm), ctx(ctx) {} |
| 430 | |
| 431 | size_t getSize() const override { return sizeof(thunkX86); } |
| 432 | MachineTypes getMachine() const override { return I386; } |
| 433 | |
| 434 | void writeTo(uint8_t *buf) const override { |
| 435 | memcpy(dest: buf, src: thunkX86, n: sizeof(thunkX86)); |
| 436 | write32le(P: buf + 1, V: imp->getRVA() + ctx.config.imageBase); |
| 437 | write32le(P: buf + 6, V: tailMerge->getRVA() - rva - 10); |
| 438 | } |
| 439 | |
| 440 | void getBaserels(std::vector<Baserel> *res) override { |
| 441 | res->emplace_back(args: rva + 1, args: ctx.config.machine); |
| 442 | } |
| 443 | |
| 444 | Defined *imp = nullptr; |
| 445 | Chunk *tailMerge = nullptr; |
| 446 | |
| 447 | private: |
| 448 | const COFFLinkerContext &ctx; |
| 449 | }; |
| 450 | |
| 451 | class TailMergeChunkX86 : public NonSectionCodeChunk { |
| 452 | public: |
| 453 | TailMergeChunkX86(COFFLinkerContext &ctx, Chunk *d, Defined *h) |
| 454 | : desc(d), helper(h), ctx(ctx) {} |
| 455 | |
| 456 | size_t getSize() const override { return sizeof(tailMergeX86); } |
| 457 | MachineTypes getMachine() const override { return I386; } |
| 458 | |
| 459 | void writeTo(uint8_t *buf) const override { |
| 460 | memcpy(dest: buf, src: tailMergeX86, n: sizeof(tailMergeX86)); |
| 461 | write32le(P: buf + 4, V: desc->getRVA() + ctx.config.imageBase); |
| 462 | write32le(P: buf + 9, V: helper->getRVA() - rva - 13); |
| 463 | } |
| 464 | |
| 465 | void getBaserels(std::vector<Baserel> *res) override { |
| 466 | res->emplace_back(args: rva + 4, args: ctx.config.machine); |
| 467 | } |
| 468 | |
| 469 | Chunk *desc = nullptr; |
| 470 | Defined *helper = nullptr; |
| 471 | |
| 472 | private: |
| 473 | const COFFLinkerContext &ctx; |
| 474 | }; |
| 475 | |
| 476 | class ThunkChunkARM : public NonSectionCodeChunk { |
| 477 | public: |
| 478 | ThunkChunkARM(COFFLinkerContext &ctx, Defined *i, Chunk *tm) |
| 479 | : imp(i), tailMerge(tm), ctx(ctx) { |
| 480 | setAlignment(2); |
| 481 | } |
| 482 | |
| 483 | size_t getSize() const override { return sizeof(thunkARM); } |
| 484 | MachineTypes getMachine() const override { return ARMNT; } |
| 485 | |
| 486 | void writeTo(uint8_t *buf) const override { |
| 487 | memcpy(dest: buf, src: thunkARM, n: sizeof(thunkARM)); |
| 488 | applyMOV32T(off: buf + 0, v: imp->getRVA() + ctx.config.imageBase); |
| 489 | applyBranch24T(off: buf + 8, v: tailMerge->getRVA() - rva - 12); |
| 490 | } |
| 491 | |
| 492 | void getBaserels(std::vector<Baserel> *res) override { |
| 493 | res->emplace_back(args: rva + 0, args: IMAGE_REL_BASED_ARM_MOV32T); |
| 494 | } |
| 495 | |
| 496 | Defined *imp = nullptr; |
| 497 | Chunk *tailMerge = nullptr; |
| 498 | |
| 499 | private: |
| 500 | const COFFLinkerContext &ctx; |
| 501 | }; |
| 502 | |
| 503 | class TailMergeChunkARM : public NonSectionCodeChunk { |
| 504 | public: |
| 505 | TailMergeChunkARM(COFFLinkerContext &ctx, Chunk *d, Defined *h) |
| 506 | : desc(d), helper(h), ctx(ctx) { |
| 507 | setAlignment(2); |
| 508 | } |
| 509 | |
| 510 | size_t getSize() const override { return sizeof(tailMergeARM); } |
| 511 | MachineTypes getMachine() const override { return ARMNT; } |
| 512 | |
| 513 | void writeTo(uint8_t *buf) const override { |
| 514 | memcpy(dest: buf, src: tailMergeARM, n: sizeof(tailMergeARM)); |
| 515 | applyMOV32T(off: buf + 14, v: desc->getRVA() + ctx.config.imageBase); |
| 516 | applyBranch24T(off: buf + 22, v: helper->getRVA() - rva - 26); |
| 517 | } |
| 518 | |
| 519 | void getBaserels(std::vector<Baserel> *res) override { |
| 520 | res->emplace_back(args: rva + 14, args: IMAGE_REL_BASED_ARM_MOV32T); |
| 521 | } |
| 522 | |
| 523 | Chunk *desc = nullptr; |
| 524 | Defined *helper = nullptr; |
| 525 | |
| 526 | private: |
| 527 | const COFFLinkerContext &ctx; |
| 528 | }; |
| 529 | |
| 530 | class ThunkChunkARM64 : public NonSectionCodeChunk { |
| 531 | public: |
| 532 | ThunkChunkARM64(Defined *i, Chunk *tm) : imp(i), tailMerge(tm) { |
| 533 | setAlignment(4); |
| 534 | } |
| 535 | |
| 536 | size_t getSize() const override { return sizeof(thunkARM64); } |
| 537 | MachineTypes getMachine() const override { return ARM64; } |
| 538 | |
| 539 | void writeTo(uint8_t *buf) const override { |
| 540 | memcpy(dest: buf, src: thunkARM64, n: sizeof(thunkARM64)); |
| 541 | applyArm64Addr(off: buf + 0, s: imp->getRVA(), p: rva + 0, shift: 12); |
| 542 | applyArm64Imm(off: buf + 4, imm: imp->getRVA() & 0xfff, rangeLimit: 0); |
| 543 | applyArm64Branch26(off: buf + 8, v: tailMerge->getRVA() - rva - 8); |
| 544 | } |
| 545 | |
| 546 | Defined *imp = nullptr; |
| 547 | Chunk *tailMerge = nullptr; |
| 548 | }; |
| 549 | |
| 550 | class TailMergeChunkARM64 : public NonSectionCodeChunk { |
| 551 | public: |
| 552 | TailMergeChunkARM64(Chunk *d, Defined *h) : desc(d), helper(h) { |
| 553 | setAlignment(4); |
| 554 | } |
| 555 | |
| 556 | size_t getSize() const override { return sizeof(tailMergeARM64); } |
| 557 | MachineTypes getMachine() const override { return ARM64; } |
| 558 | |
| 559 | void writeTo(uint8_t *buf) const override { |
| 560 | memcpy(dest: buf, src: tailMergeARM64, n: sizeof(tailMergeARM64)); |
| 561 | applyArm64Addr(off: buf + 48, s: desc->getRVA(), p: rva + 48, shift: 12); |
| 562 | applyArm64Imm(off: buf + 52, imm: desc->getRVA() & 0xfff, rangeLimit: 0); |
| 563 | if (helper) { |
| 564 | applyArm64Addr(off: buf + 56, s: helper->getRVA(), p: rva + 56, shift: 12); |
| 565 | applyArm64Imm(off: buf + 60, imm: helper->getRVA() & 0xfff, rangeLimit: 0); |
| 566 | } |
| 567 | } |
| 568 | |
| 569 | Chunk *desc = nullptr; |
| 570 | Defined *helper = nullptr; |
| 571 | }; |
| 572 | |
| 573 | // A chunk for the import descriptor table. |
| 574 | class DelayAddressChunk : public NonSectionChunk { |
| 575 | public: |
| 576 | explicit DelayAddressChunk(COFFLinkerContext &ctx, Chunk *c) |
| 577 | : thunk(c), ctx(ctx) { |
| 578 | setAlignment(ctx.config.wordsize); |
| 579 | } |
| 580 | size_t getSize() const override { return ctx.config.wordsize; } |
| 581 | |
| 582 | void writeTo(uint8_t *buf) const override { |
| 583 | if (ctx.config.is64()) { |
| 584 | write64le(P: buf, V: thunk->getRVA() + ctx.config.imageBase); |
| 585 | } else { |
| 586 | uint32_t bit = 0; |
| 587 | // Pointer to thumb code must have the LSB set, so adjust it. |
| 588 | if (ctx.config.machine == ARMNT) |
| 589 | bit = 1; |
| 590 | write32le(P: buf, V: (thunk->getRVA() + ctx.config.imageBase) | bit); |
| 591 | } |
| 592 | } |
| 593 | |
| 594 | void getBaserels(std::vector<Baserel> *res) override { |
| 595 | res->emplace_back(args&: rva, args: ctx.config.machine); |
| 596 | } |
| 597 | |
| 598 | Chunk *thunk; |
| 599 | |
| 600 | private: |
| 601 | const COFFLinkerContext &ctx; |
| 602 | }; |
| 603 | |
| 604 | // Export table |
| 605 | // Read Microsoft PE/COFF spec 5.3 for details. |
| 606 | |
| 607 | // A chunk for the export descriptor table. |
| 608 | class ExportDirectoryChunk : public NonSectionChunk { |
| 609 | public: |
| 610 | ExportDirectoryChunk(int baseOrdinal, int maxOrdinal, int nameTabSize, |
| 611 | Chunk *d, Chunk *a, Chunk *n, Chunk *o) |
| 612 | : baseOrdinal(baseOrdinal), maxOrdinal(maxOrdinal), |
| 613 | nameTabSize(nameTabSize), dllName(d), addressTab(a), nameTab(n), |
| 614 | ordinalTab(o) {} |
| 615 | |
| 616 | size_t getSize() const override { |
| 617 | return sizeof(export_directory_table_entry); |
| 618 | } |
| 619 | |
| 620 | void writeTo(uint8_t *buf) const override { |
| 621 | memset(s: buf, c: 0, n: getSize()); |
| 622 | |
| 623 | auto *e = (export_directory_table_entry *)(buf); |
| 624 | e->NameRVA = dllName->getRVA(); |
| 625 | e->OrdinalBase = baseOrdinal; |
| 626 | e->AddressTableEntries = (maxOrdinal - baseOrdinal) + 1; |
| 627 | e->NumberOfNamePointers = nameTabSize; |
| 628 | e->ExportAddressTableRVA = addressTab->getRVA(); |
| 629 | e->NamePointerRVA = nameTab->getRVA(); |
| 630 | e->OrdinalTableRVA = ordinalTab->getRVA(); |
| 631 | } |
| 632 | |
| 633 | uint16_t baseOrdinal; |
| 634 | uint16_t maxOrdinal; |
| 635 | uint16_t nameTabSize; |
| 636 | Chunk *dllName; |
| 637 | Chunk *addressTab; |
| 638 | Chunk *nameTab; |
| 639 | Chunk *ordinalTab; |
| 640 | }; |
| 641 | |
| 642 | class AddressTableChunk : public NonSectionChunk { |
| 643 | public: |
| 644 | explicit AddressTableChunk(SymbolTable &symtab, size_t baseOrdinal, |
| 645 | size_t maxOrdinal) |
| 646 | : baseOrdinal(baseOrdinal), size((maxOrdinal - baseOrdinal) + 1), |
| 647 | symtab(symtab) {} |
| 648 | size_t getSize() const override { return size * 4; } |
| 649 | |
| 650 | void writeTo(uint8_t *buf) const override { |
| 651 | memset(s: buf, c: 0, n: getSize()); |
| 652 | |
| 653 | for (const Export &e : symtab.exports) { |
| 654 | assert(e.ordinal >= baseOrdinal && "Export symbol has invalid ordinal" ); |
| 655 | // Subtract the OrdinalBase to get the index. |
| 656 | uint8_t *p = buf + (e.ordinal - baseOrdinal) * 4; |
| 657 | uint32_t bit = 0; |
| 658 | // Pointer to thumb code must have the LSB set, so adjust it. |
| 659 | if (symtab.machine == ARMNT && !e.data) |
| 660 | bit = 1; |
| 661 | if (e.forwardChunk) { |
| 662 | write32le(P: p, V: e.forwardChunk->getRVA() | bit); |
| 663 | } else { |
| 664 | assert(cast<Defined>(e.sym)->getRVA() != 0 && |
| 665 | "Exported symbol unmapped" ); |
| 666 | write32le(P: p, V: cast<Defined>(Val: e.sym)->getRVA() | bit); |
| 667 | } |
| 668 | } |
| 669 | } |
| 670 | |
| 671 | private: |
| 672 | size_t baseOrdinal; |
| 673 | size_t size; |
| 674 | const SymbolTable &symtab; |
| 675 | }; |
| 676 | |
| 677 | class NamePointersChunk : public NonSectionChunk { |
| 678 | public: |
| 679 | explicit NamePointersChunk(std::vector<Chunk *> &v) : chunks(v) {} |
| 680 | size_t getSize() const override { return chunks.size() * 4; } |
| 681 | |
| 682 | void writeTo(uint8_t *buf) const override { |
| 683 | for (Chunk *c : chunks) { |
| 684 | write32le(P: buf, V: c->getRVA()); |
| 685 | buf += 4; |
| 686 | } |
| 687 | } |
| 688 | |
| 689 | private: |
| 690 | std::vector<Chunk *> chunks; |
| 691 | }; |
| 692 | |
| 693 | class ExportOrdinalChunk : public NonSectionChunk { |
| 694 | public: |
| 695 | explicit ExportOrdinalChunk(const SymbolTable &symtab, size_t baseOrdinal, |
| 696 | size_t tableSize) |
| 697 | : baseOrdinal(baseOrdinal), size(tableSize), symtab(symtab) {} |
| 698 | size_t getSize() const override { return size * 2; } |
| 699 | |
| 700 | void writeTo(uint8_t *buf) const override { |
| 701 | for (const Export &e : symtab.exports) { |
| 702 | if (e.noname) |
| 703 | continue; |
| 704 | assert(e.ordinal >= baseOrdinal && "Export symbol has invalid ordinal" ); |
| 705 | // This table stores unbiased indices, so subtract OrdinalBase. |
| 706 | write16le(P: buf, V: e.ordinal - baseOrdinal); |
| 707 | buf += 2; |
| 708 | } |
| 709 | } |
| 710 | |
| 711 | private: |
| 712 | size_t baseOrdinal; |
| 713 | size_t size; |
| 714 | const SymbolTable &symtab; |
| 715 | }; |
| 716 | |
| 717 | } // anonymous namespace |
| 718 | |
| 719 | void IdataContents::create(COFFLinkerContext &ctx) { |
| 720 | std::vector<std::vector<DefinedImportData *>> v = binImports(ctx, imports); |
| 721 | |
| 722 | // In hybrid images, EC and native code are usually very similar, |
| 723 | // resulting in a highly similar set of imported symbols. Consequently, |
| 724 | // their import tables can be shared, with ARM64X relocations handling any |
| 725 | // differences. Identify matching import files used by EC and native code, and |
| 726 | // merge them into a single hybrid import entry. |
| 727 | if (ctx.hybridSymtab) { |
| 728 | for (std::vector<DefinedImportData *> &syms : v) { |
| 729 | std::vector<DefinedImportData *> hybridSyms; |
| 730 | ImportFile *prev = nullptr; |
| 731 | for (DefinedImportData *sym : syms) { |
| 732 | ImportFile *file = sym->file; |
| 733 | // At this stage, symbols are sorted by base name, ensuring that |
| 734 | // compatible import files, if present, are adjacent. Check if the |
| 735 | // current symbol's file imports the same symbol as the previously added |
| 736 | // one (if any and if it was not already merged). Additionally, verify |
| 737 | // that one of them is native while the other is EC. In rare cases, |
| 738 | // separate matching import entries may exist within the same namespace, |
| 739 | // which cannot be merged. |
| 740 | if (!prev || file->isEC() == prev->isEC() || |
| 741 | !file->isSameImport(other: prev)) { |
| 742 | // We can't merge the import file, just add it to hybridSyms |
| 743 | // and set prev to its file so that we can try to match the next |
| 744 | // symbol. |
| 745 | hybridSyms.push_back(x: sym); |
| 746 | prev = file; |
| 747 | continue; |
| 748 | } |
| 749 | |
| 750 | // A matching symbol may appear in syms in any order. The native variant |
| 751 | // exposes a subset of EC symbols and chunks, so always use the EC |
| 752 | // variant as the hybrid import file. If the native file was already |
| 753 | // added, replace it with the EC symbol in hybridSyms. Otherwise, the EC |
| 754 | // variant is already pushed, so we can simply merge it. |
| 755 | if (file->isEC()) { |
| 756 | hybridSyms.pop_back(); |
| 757 | hybridSyms.push_back(x: sym); |
| 758 | } |
| 759 | |
| 760 | // Merge import files by storing their hybrid form in the corresponding |
| 761 | // file class. |
| 762 | prev->hybridFile = file; |
| 763 | file->hybridFile = prev; |
| 764 | prev = nullptr; // A hybrid import file cannot be merged again. |
| 765 | } |
| 766 | |
| 767 | // Sort symbols by type: native-only files first, followed by merged |
| 768 | // hybrid files, and then EC-only files. |
| 769 | llvm::stable_sort(Range&: hybridSyms, |
| 770 | C: [](DefinedImportData *a, DefinedImportData *b) { |
| 771 | if (a->file->hybridFile) |
| 772 | return !b->file->hybridFile && b->file->isEC(); |
| 773 | return !a->file->isEC() && b->file->isEC(); |
| 774 | }); |
| 775 | syms = std::move(hybridSyms); |
| 776 | } |
| 777 | } |
| 778 | |
| 779 | // Create .idata contents for each DLL. |
| 780 | for (std::vector<DefinedImportData *> &syms : v) { |
| 781 | // Create lookup and address tables. If they have external names, |
| 782 | // we need to create hintName chunks to store the names. |
| 783 | // If they don't (if they are import-by-ordinals), we store only |
| 784 | // ordinal values to the table. |
| 785 | size_t base = lookups.size(); |
| 786 | Chunk *lookupsTerminator = nullptr, *addressesTerminator = nullptr; |
| 787 | uint32_t nativeOnly = 0; |
| 788 | for (DefinedImportData *s : syms) { |
| 789 | uint16_t ord = s->getOrdinal(); |
| 790 | HintNameChunk *hintChunk = nullptr; |
| 791 | Chunk *lookupsChunk, *addressesChunk; |
| 792 | |
| 793 | if (s->getExternalName().empty()) { |
| 794 | lookupsChunk = make<OrdinalOnlyChunk>(args&: ctx, args&: ord); |
| 795 | addressesChunk = make<OrdinalOnlyChunk>(args&: ctx, args&: ord); |
| 796 | } else { |
| 797 | hintChunk = make<HintNameChunk>(args: s->getExternalName(), args&: ord); |
| 798 | lookupsChunk = make<LookupChunk>(args&: ctx, args&: hintChunk); |
| 799 | addressesChunk = make<LookupChunk>(args&: ctx, args&: hintChunk); |
| 800 | hints.push_back(x: hintChunk); |
| 801 | } |
| 802 | |
| 803 | // Detect the first EC-only import in the hybrid IAT. Emit null chunk |
| 804 | // as a terminator for the native view, and add an ARM64X relocation to |
| 805 | // replace it with the correct import for the EC view. |
| 806 | // |
| 807 | // Additionally, for MSVC compatibility, store the lookup and address |
| 808 | // chunks and append them at the end of EC-only imports, where a null |
| 809 | // terminator chunk would typically be placed. Since they appear after |
| 810 | // the native terminator, they will be ignored in the native view. |
| 811 | // In the EC view, they should act as terminators, so emit ZEROFILL |
| 812 | // relocations overriding them. |
| 813 | if (ctx.config.machine == ARM64X && !lookupsTerminator && |
| 814 | s->file->isEC() && !s->file->hybridFile) { |
| 815 | lookupsTerminator = lookupsChunk; |
| 816 | addressesTerminator = addressesChunk; |
| 817 | lookupsChunk = make<NullChunk>(args&: ctx); |
| 818 | addressesChunk = make<NullChunk>(args&: ctx); |
| 819 | |
| 820 | Arm64XRelocVal relocVal = hintChunk; |
| 821 | if (!hintChunk) |
| 822 | relocVal = (1ULL << 63) | ord; |
| 823 | ctx.dynamicRelocs->add(type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_VALUE, |
| 824 | size: sizeof(uint64_t), offset: lookupsChunk, value: relocVal); |
| 825 | ctx.dynamicRelocs->add(type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_VALUE, |
| 826 | size: sizeof(uint64_t), offset: addressesChunk, value: relocVal); |
| 827 | ctx.dynamicRelocs->add(type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_ZEROFILL, |
| 828 | size: sizeof(uint64_t), offset: lookupsTerminator); |
| 829 | ctx.dynamicRelocs->add(type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_ZEROFILL, |
| 830 | size: sizeof(uint64_t), offset: addressesTerminator); |
| 831 | } |
| 832 | |
| 833 | lookups.push_back(x: lookupsChunk); |
| 834 | addresses.push_back(x: addressesChunk); |
| 835 | |
| 836 | if (s->file->isEC()) { |
| 837 | auto chunk = make<AuxImportChunk>(args&: s->file); |
| 838 | auxIat.push_back(x: chunk); |
| 839 | s->file->impECSym->setLocation(chunk); |
| 840 | |
| 841 | chunk = make<AuxImportChunk>(args&: s->file); |
| 842 | auxIatCopy.push_back(x: chunk); |
| 843 | s->file->auxImpCopySym->setLocation(chunk); |
| 844 | } else if (ctx.hybridSymtab) { |
| 845 | // Fill the auxiliary IAT with null chunks for native-only imports. |
| 846 | auxIat.push_back(x: make<NullChunk>(args&: ctx)); |
| 847 | auxIatCopy.push_back(x: make<NullChunk>(args&: ctx)); |
| 848 | ++nativeOnly; |
| 849 | } |
| 850 | } |
| 851 | // Terminate with null values. |
| 852 | lookups.push_back(x: lookupsTerminator ? lookupsTerminator |
| 853 | : make<NullChunk>(args&: ctx)); |
| 854 | addresses.push_back(x: addressesTerminator ? addressesTerminator |
| 855 | : make<NullChunk>(args&: ctx)); |
| 856 | if (ctx.symtab.isEC()) { |
| 857 | auxIat.push_back(x: make<NullChunk>(args&: ctx)); |
| 858 | auxIatCopy.push_back(x: make<NullChunk>(args&: ctx)); |
| 859 | } |
| 860 | |
| 861 | for (int i = 0, e = syms.size(); i < e; ++i) { |
| 862 | syms[i]->setLocation(addresses[base + i]); |
| 863 | if (syms[i]->file->hybridFile) |
| 864 | syms[i]->file->hybridFile->impSym->setLocation(addresses[base + i]); |
| 865 | } |
| 866 | |
| 867 | // Create the import table header. |
| 868 | dllNames.push_back(x: make<StringChunk>(args: syms[0]->getDLLName())); |
| 869 | auto *dir = make<ImportDirectoryChunk>(args&: dllNames.back()); |
| 870 | |
| 871 | if (ctx.hybridSymtab && nativeOnly) { |
| 872 | if (ctx.config.machine != ARM64X) |
| 873 | // On pure ARM64EC targets, skip native-only imports in the import |
| 874 | // directory. |
| 875 | base += nativeOnly; |
| 876 | else if (nativeOnly) { |
| 877 | // If native-only imports exist, they will appear as a prefix to all |
| 878 | // imports. Emit ARM64X relocations to skip them in the EC view. |
| 879 | ctx.dynamicRelocs->add( |
| 880 | type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_DELTA, size: 0, |
| 881 | offset: Arm64XRelocVal( |
| 882 | dir, offsetof(ImportDirectoryTableEntry, ImportLookupTableRVA)), |
| 883 | value: nativeOnly * sizeof(uint64_t)); |
| 884 | ctx.dynamicRelocs->add( |
| 885 | type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_DELTA, size: 0, |
| 886 | offset: Arm64XRelocVal(dir, offsetof(ImportDirectoryTableEntry, |
| 887 | ImportAddressTableRVA)), |
| 888 | value: nativeOnly * sizeof(uint64_t)); |
| 889 | } |
| 890 | } |
| 891 | |
| 892 | dir->lookupTab = lookups[base]; |
| 893 | dir->addressTab = addresses[base]; |
| 894 | dirs.push_back(x: dir); |
| 895 | } |
| 896 | // Add null terminator. |
| 897 | dirs.push_back(x: make<NullChunk>(args: sizeof(ImportDirectoryTableEntry), args: 4)); |
| 898 | } |
| 899 | |
| 900 | std::vector<Chunk *> DelayLoadContents::getChunks() { |
| 901 | std::vector<Chunk *> v; |
| 902 | v.insert(position: v.end(), first: dirs.begin(), last: dirs.end()); |
| 903 | v.insert(position: v.end(), first: names.begin(), last: names.end()); |
| 904 | v.insert(position: v.end(), first: hintNames.begin(), last: hintNames.end()); |
| 905 | v.insert(position: v.end(), first: dllNames.begin(), last: dllNames.end()); |
| 906 | return v; |
| 907 | } |
| 908 | |
| 909 | std::vector<Chunk *> DelayLoadContents::getDataChunks() { |
| 910 | std::vector<Chunk *> v; |
| 911 | v.insert(position: v.end(), first: moduleHandles.begin(), last: moduleHandles.end()); |
| 912 | v.insert(position: v.end(), first: addresses.begin(), last: addresses.end()); |
| 913 | return v; |
| 914 | } |
| 915 | |
| 916 | uint64_t DelayLoadContents::getDirSize() { |
| 917 | return dirs.size() * sizeof(delay_import_directory_table_entry); |
| 918 | } |
| 919 | |
| 920 | void DelayLoadContents::create() { |
| 921 | std::vector<std::vector<DefinedImportData *>> v = binImports(ctx, imports); |
| 922 | |
| 923 | // Create .didat contents for each DLL. |
| 924 | for (std::vector<DefinedImportData *> &syms : v) { |
| 925 | // Create the delay import table header. |
| 926 | dllNames.push_back(x: make<StringChunk>(args: syms[0]->getDLLName())); |
| 927 | auto *dir = make<DelayDirectoryChunk>(args&: dllNames.back()); |
| 928 | |
| 929 | size_t base = addresses.size(); |
| 930 | ctx.forEachSymtab(f: [&](SymbolTable &symtab) { |
| 931 | if (symtab.isEC()) { |
| 932 | if (ctx.config.machine == ARM64X) { |
| 933 | // For hybrid images, emit null-terminated native import entries |
| 934 | // followed by null-terminated EC entries. If a view is missing |
| 935 | // imports for a given module, only terminators are emitted. Emit |
| 936 | // ARM64X relocations to skip native entries in the EC view. |
| 937 | ctx.dynamicRelocs->add( |
| 938 | type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_DELTA, size: 0, |
| 939 | offset: Arm64XRelocVal(dir, offsetof(delay_import_directory_table_entry, |
| 940 | DelayImportAddressTable)), |
| 941 | value: (addresses.size() - base) * sizeof(uint64_t)); |
| 942 | ctx.dynamicRelocs->add( |
| 943 | type: IMAGE_DVRT_ARM64X_FIXUP_TYPE_DELTA, size: 0, |
| 944 | offset: Arm64XRelocVal(dir, offsetof(delay_import_directory_table_entry, |
| 945 | DelayImportNameTable)), |
| 946 | value: (addresses.size() - base) * sizeof(uint64_t)); |
| 947 | } else { |
| 948 | base = addresses.size(); |
| 949 | } |
| 950 | } |
| 951 | |
| 952 | Chunk *tm = nullptr; |
| 953 | |
| 954 | for (DefinedImportData *s : syms) { |
| 955 | // Process only the symbols belonging to the current symtab. |
| 956 | if (symtab.isEC() != s->file->isEC()) |
| 957 | continue; |
| 958 | |
| 959 | if (!tm) { |
| 960 | tm = newTailMergeChunk(symtab, dir); |
| 961 | Chunk *pdataChunk = newTailMergePDataChunk(symtab, tm); |
| 962 | if (pdataChunk) |
| 963 | pdata.push_back(x: pdataChunk); |
| 964 | } |
| 965 | |
| 966 | Chunk *t = newThunkChunk(s, tailMerge: tm); |
| 967 | auto *a = make<DelayAddressChunk>(args&: ctx, args&: t); |
| 968 | addresses.push_back(x: a); |
| 969 | s->setLocation(a); |
| 970 | thunks.push_back(x: t); |
| 971 | StringRef extName = s->getExternalName(); |
| 972 | if (extName.empty()) { |
| 973 | names.push_back(x: make<OrdinalOnlyChunk>(args&: ctx, args: s->getOrdinal())); |
| 974 | } else { |
| 975 | auto *c = make<HintNameChunk>(args&: extName, args: 0); |
| 976 | names.push_back(x: make<LookupChunk>(args&: ctx, args&: c)); |
| 977 | hintNames.push_back(x: c); |
| 978 | // Add a synthetic symbol for this load thunk, using the |
| 979 | // "__imp___load" prefix, in case this thunk needs to be added to the |
| 980 | // list of valid call targets for Control Flow Guard. |
| 981 | StringRef symName = saver().save(S: "__imp___load_" + extName); |
| 982 | s->loadThunkSym = |
| 983 | cast<DefinedSynthetic>(Val: symtab.addSynthetic(n: symName, c: t)); |
| 984 | } |
| 985 | |
| 986 | if (symtab.isEC()) { |
| 987 | auto chunk = make<AuxImportChunk>(args&: s->file); |
| 988 | auxIat.push_back(x: chunk); |
| 989 | s->file->impECSym->setLocation(chunk); |
| 990 | |
| 991 | chunk = make<AuxImportChunk>(args&: s->file); |
| 992 | auxIatCopy.push_back(x: chunk); |
| 993 | s->file->auxImpCopySym->setLocation(chunk); |
| 994 | } else if (ctx.config.machine == ARM64X) { |
| 995 | // Fill the auxiliary IAT with null chunks for native imports. |
| 996 | auxIat.push_back(x: make<NullChunk>(args&: ctx)); |
| 997 | auxIatCopy.push_back(x: make<NullChunk>(args&: ctx)); |
| 998 | } |
| 999 | } |
| 1000 | |
| 1001 | if (tm) { |
| 1002 | thunks.push_back(x: tm); |
| 1003 | StringRef tmName = |
| 1004 | saver().save(S: "__tailMerge_" + syms[0]->getDLLName().lower()); |
| 1005 | symtab.addSynthetic(n: tmName, c: tm); |
| 1006 | } |
| 1007 | |
| 1008 | // Skip terminators on pure ARM64EC target if there are no native imports. |
| 1009 | if (!tm && !symtab.isEC() && ctx.config.machine != ARM64X) |
| 1010 | return; |
| 1011 | |
| 1012 | // Terminate with null values. |
| 1013 | addresses.push_back(x: make<NullChunk>(args&: ctx, args: 8)); |
| 1014 | names.push_back(x: make<NullChunk>(args&: ctx, args: 8)); |
| 1015 | if (ctx.symtab.isEC()) { |
| 1016 | auxIat.push_back(x: make<NullChunk>(args&: ctx, args: 8)); |
| 1017 | auxIatCopy.push_back(x: make<NullChunk>(args&: ctx, args: 8)); |
| 1018 | } |
| 1019 | }); |
| 1020 | |
| 1021 | auto *mh = make<NullChunk>(args: 8, args: 8); |
| 1022 | moduleHandles.push_back(x: mh); |
| 1023 | |
| 1024 | // Fill the delay import table header fields. |
| 1025 | dir->moduleHandle = mh; |
| 1026 | dir->addressTab = addresses[base]; |
| 1027 | dir->nameTab = names[base]; |
| 1028 | dirs.push_back(x: dir); |
| 1029 | } |
| 1030 | |
| 1031 | ctx.forEachSymtab(f: [&](SymbolTable &symtab) { |
| 1032 | if (symtab.tailMergeUnwindInfoChunk) |
| 1033 | unwindinfo.push_back(x: symtab.tailMergeUnwindInfoChunk); |
| 1034 | }); |
| 1035 | // Add null terminator. |
| 1036 | dirs.push_back( |
| 1037 | x: make<NullChunk>(args: sizeof(delay_import_directory_table_entry), args: 4)); |
| 1038 | } |
| 1039 | |
| 1040 | Chunk *DelayLoadContents::newTailMergeChunk(SymbolTable &symtab, Chunk *dir) { |
| 1041 | auto helper = cast_or_null<Defined>(Val: symtab.delayLoadHelper); |
| 1042 | switch (symtab.machine) { |
| 1043 | case AMD64: |
| 1044 | case ARM64EC: |
| 1045 | return make<TailMergeChunkX64>(args&: dir, args&: helper); |
| 1046 | case I386: |
| 1047 | return make<TailMergeChunkX86>(args&: ctx, args&: dir, args&: helper); |
| 1048 | case ARMNT: |
| 1049 | return make<TailMergeChunkARM>(args&: ctx, args&: dir, args&: helper); |
| 1050 | case ARM64: |
| 1051 | return make<TailMergeChunkARM64>(args&: dir, args&: helper); |
| 1052 | default: |
| 1053 | llvm_unreachable("unsupported machine type" ); |
| 1054 | } |
| 1055 | } |
| 1056 | |
| 1057 | Chunk *DelayLoadContents::newTailMergePDataChunk(SymbolTable &symtab, |
| 1058 | Chunk *tm) { |
| 1059 | switch (symtab.machine) { |
| 1060 | case AMD64: |
| 1061 | case ARM64EC: |
| 1062 | if (!symtab.tailMergeUnwindInfoChunk) |
| 1063 | symtab.tailMergeUnwindInfoChunk = make<TailMergeUnwindInfoX64>(); |
| 1064 | return make<TailMergePDataChunkX64>(args&: tm, args&: symtab.tailMergeUnwindInfoChunk); |
| 1065 | // FIXME: Add support for other architectures. |
| 1066 | default: |
| 1067 | return nullptr; // Just don't generate unwind info. |
| 1068 | } |
| 1069 | } |
| 1070 | |
| 1071 | Chunk *DelayLoadContents::newThunkChunk(DefinedImportData *s, |
| 1072 | Chunk *tailMerge) { |
| 1073 | switch (s->file->getMachineType()) { |
| 1074 | case AMD64: |
| 1075 | case ARM64EC: |
| 1076 | return make<ThunkChunkX64>(args&: s, args&: tailMerge); |
| 1077 | case I386: |
| 1078 | return make<ThunkChunkX86>(args&: ctx, args&: s, args&: tailMerge); |
| 1079 | case ARMNT: |
| 1080 | return make<ThunkChunkARM>(args&: ctx, args&: s, args&: tailMerge); |
| 1081 | case ARM64: |
| 1082 | return make<ThunkChunkARM64>(args&: s, args&: tailMerge); |
| 1083 | default: |
| 1084 | llvm_unreachable("unsupported machine type" ); |
| 1085 | } |
| 1086 | } |
| 1087 | |
| 1088 | void createEdataChunks(SymbolTable &symtab, std::vector<Chunk *> &chunks) { |
| 1089 | unsigned baseOrdinal = 1 << 16, maxOrdinal = 0; |
| 1090 | for (Export &e : symtab.exports) { |
| 1091 | baseOrdinal = std::min(a: baseOrdinal, b: (unsigned)e.ordinal); |
| 1092 | maxOrdinal = std::max(a: maxOrdinal, b: (unsigned)e.ordinal); |
| 1093 | } |
| 1094 | // Ordinals must start at 1 as suggested in: |
| 1095 | // https://learn.microsoft.com/en-us/cpp/build/reference/export-exports-a-function?view=msvc-170 |
| 1096 | assert(baseOrdinal >= 1); |
| 1097 | |
| 1098 | auto *dllName = |
| 1099 | make<StringChunk>(args: sys::path::filename(path: symtab.ctx.config.outputFile)); |
| 1100 | auto *addressTab = make<AddressTableChunk>(args&: symtab, args&: baseOrdinal, args&: maxOrdinal); |
| 1101 | std::vector<Chunk *> names; |
| 1102 | for (Export &e : symtab.exports) |
| 1103 | if (!e.noname) |
| 1104 | names.push_back(x: make<StringChunk>(args&: e.exportName)); |
| 1105 | |
| 1106 | std::vector<Chunk *> forwards; |
| 1107 | for (Export &e : symtab.exports) { |
| 1108 | if (e.forwardTo.empty()) |
| 1109 | continue; |
| 1110 | e.forwardChunk = make<StringChunk>(args&: e.forwardTo); |
| 1111 | forwards.push_back(x: e.forwardChunk); |
| 1112 | } |
| 1113 | |
| 1114 | auto *nameTab = make<NamePointersChunk>(args&: names); |
| 1115 | auto *ordinalTab = |
| 1116 | make<ExportOrdinalChunk>(args&: symtab, args&: baseOrdinal, args: names.size()); |
| 1117 | auto *dir = |
| 1118 | make<ExportDirectoryChunk>(args&: baseOrdinal, args&: maxOrdinal, args: names.size(), args&: dllName, |
| 1119 | args&: addressTab, args&: nameTab, args&: ordinalTab); |
| 1120 | chunks.push_back(x: dir); |
| 1121 | chunks.push_back(x: dllName); |
| 1122 | chunks.push_back(x: addressTab); |
| 1123 | chunks.push_back(x: nameTab); |
| 1124 | chunks.push_back(x: ordinalTab); |
| 1125 | chunks.insert(position: chunks.end(), first: names.begin(), last: names.end()); |
| 1126 | chunks.insert(position: chunks.end(), first: forwards.begin(), last: forwards.end()); |
| 1127 | } |
| 1128 | |
| 1129 | } // namespace lld::coff |
| 1130 | |