lld/COFF/DLL.cpp

   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(name.size() + 3, 2);
  48   }
  49
  50   void writeTo(uint8_t *buf) const override {
  51     memset(buf, 0, getSize());
  52     write16le(buf, hint);
  53     memcpy(buf + 2, name.data(), 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(buf, hintName->getRVA());
  73     else
  74       write32le(buf, 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(buf, (1ULL << 63) | ordinal);
  99     } else {
 100       write32le(buf, (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(buf, 0, 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     hasData = false;
 136     setAlignment(align);
 137   }
 138   explicit NullChunk(COFFLinkerContext &ctx)
 139       : NullChunk(ctx.config.wordsize, ctx.config.wordsize) {}
 140   explicit NullChunk(COFFLinkerContext &ctx, size_t n)
 141       : NullChunk(n, ctx.config.wordsize) {}
 142   size_t getSize() const override { return size; }
 143
 144   void writeTo(uint8_t *buf) const override {
 145     memset(buf, 0, size);
 146   }
 147
 148 private:
 149   size_t size;
 150 };
 151
 152 // A chunk for ARM64EC auxiliary IAT.
 153 class AuxImportChunk : public NonSectionChunk {
 154 public:
 155   explicit AuxImportChunk(ImportFile *file) : file(file) {
 156     setAlignment(sizeof(uint64_t));
 157   }
 158   size_t getSize() const override { return sizeof(uint64_t); }
 159
 160   void writeTo(uint8_t *buf) const override {
 161     uint64_t impchkVA = 0;
 162     if (file->impchkThunk)
 163       impchkVA = file->impchkThunk->getRVA() + file->ctx.config.imageBase;
 164     write64le(buf, impchkVA);
 165   }
 166
 167   void getBaserels(std::vector<Baserel> *res) override {
 168     if (file->impchkThunk)
 169       res->emplace_back(rva, file->ctx.config.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(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(syms, [](DefinedImportData *a, DefinedImportData *b) {
 194       auto getBaseName = [](DefinedImportData *sym) {
 195         StringRef name = sym->getName();
 196         name.consume_front("__imp_");
 197         // Skip aux_ part of ARM64EC function symbol name.
 198         if (sym->file->impchkThunk)
 199           name.consume_front("aux_");
 200         return name;
 201       };
 202       return getBaseName(a) < getBaseName(b);
 203     });
 204     v.push_back(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(buf, 0, 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     0x51,                               // push    rcx
 248     0x52,                               // push    rdx
 249     0x41, 0x50,                         // push    r8
 250     0x41, 0x51,                         // push    r9
 251     0x48, 0x83, 0xEC, 0x48,             // sub     rsp, 48h
 252     0x66, 0x0F, 0x7F, 0x04, 0x24,       // movdqa  xmmword ptr [rsp], xmm0
 253     0x66, 0x0F, 0x7F, 0x4C, 0x24, 0x10, // movdqa  xmmword ptr [rsp+10h], xmm1
 254     0x66, 0x0F, 0x7F, 0x54, 0x24, 0x20, // movdqa  xmmword ptr [rsp+20h], xmm2
 255     0x66, 0x0F, 0x7F, 0x5C, 0x24, 0x30, // movdqa  xmmword ptr [rsp+30h], 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, 0x04, 0x24,       // movdqa  xmm0, xmmword ptr [rsp]
 260     0x66, 0x0F, 0x6F, 0x4C, 0x24, 0x10, // movdqa  xmm1, xmmword ptr [rsp+10h]
 261     0x66, 0x0F, 0x6F, 0x54, 0x24, 0x20, // movdqa  xmm2, xmmword ptr [rsp+20h]
 262     0x66, 0x0F, 0x6F, 0x5C, 0x24, 0x30, // movdqa  xmm3, xmmword ptr [rsp+30h]
 263     0x48, 0x83, 0xC4, 0x48,             // add     rsp, 48h
 264     0x41, 0x59,                         // pop     r9
 265     0x41, 0x58,                         // pop     r8
 266     0x5A,                               // pop     rdx
 267     0x59,                               // pop     rcx
 268     0xFF, 0xE0,                         // jmp     rax
 269 };
 270
 271 static const uint8_t tailMergeUnwindInfoX64[] = {
 272     0x01,       // Version=1, Flags=UNW_FLAG_NHANDLER
 273     0x0a,       // Size of prolog
 274     0x05,       // Count of unwind codes
 275     0x00,       // No frame register
 276     0x0a, 0x82, // Offset 0xa: UWOP_ALLOC_SMALL(0x48)
 277     0x06, 0x02, // Offset 6: UWOP_ALLOC_SMALL(8)
 278     0x04, 0x02, // Offset 4: UWOP_ALLOC_SMALL(8)
 279     0x02, 0x02, // Offset 2: UWOP_ALLOC_SMALL(8)
 280     0x01, 0x02, // Offset 1: UWOP_ALLOC_SMALL(8)
 281     0x00, 0x00  // Padding to align on 32-bits
 282 };
 283
 284 static const uint8_t thunkX86[] = {
 285     0xB8, 0, 0, 0, 0,  // mov   eax, offset ___imp__<FUNCNAME>
 286     0xE9, 0, 0, 0, 0,  // jmp   __tailMerge_<lib>
 287 };
 288
 289 static const uint8_t tailMergeX86[] = {
 290     0x51,              // push  ecx
 291     0x52,              // push  edx
 292     0x50,              // push  eax
 293     0x68, 0, 0, 0, 0,  // push  offset ___DELAY_IMPORT_DESCRIPTOR_<DLLNAME>_dll
 294     0xE8, 0, 0, 0, 0,  // call  ___delayLoadHelper2@8
 295     0x5A,              // pop   edx
 296     0x59,              // pop   ecx
 297     0xFF, 0xE0,        // jmp   eax
 298 };
 299
 300 static const uint8_t thunkARM[] = {
 301     0x40, 0xf2, 0x00, 0x0c, // mov.w   ip, #0 __imp_<FUNCNAME>
 302     0xc0, 0xf2, 0x00, 0x0c, // mov.t   ip, #0 __imp_<FUNCNAME>
 303     0x00, 0xf0, 0x00, 0xb8, // b.w     __tailMerge_<lib>
 304 };
 305
 306 static const uint8_t tailMergeARM[] = {
 307     0x2d, 0xe9, 0x0f, 0x48, // push.w  {r0, r1, r2, r3, r11, lr}
 308     0x0d, 0xf2, 0x10, 0x0b, // addw    r11, sp, #16
 309     0x2d, 0xed, 0x10, 0x0b, // vpush   {d0, d1, d2, d3, d4, d5, d6, d7}
 310     0x61, 0x46,             // mov     r1, ip
 311     0x40, 0xf2, 0x00, 0x00, // mov.w   r0, #0 DELAY_IMPORT_DESCRIPTOR
 312     0xc0, 0xf2, 0x00, 0x00, // mov.t   r0, #0 DELAY_IMPORT_DESCRIPTOR
 313     0x00, 0xf0, 0x00, 0xd0, // bl      #0 __delayLoadHelper2
 314     0x84, 0x46,             // mov     ip, r0
 315     0xbd, 0xec, 0x10, 0x0b, // vpop    {d0, d1, d2, d3, d4, d5, d6, d7}
 316     0xbd, 0xe8, 0x0f, 0x48, // pop.w   {r0, r1, r2, r3, r11, lr}
 317     0x60, 0x47,             // bx      ip
 318 };
 319
 320 static const uint8_t thunkARM64[] = {
 321     0x11, 0x00, 0x00, 0x90, // adrp    x17, #0      __imp_<FUNCNAME>
 322     0x31, 0x02, 0x00, 0x91, // add     x17, x17, #0 :lo12:__imp_<FUNCNAME>
 323     0x00, 0x00, 0x00, 0x14, // b       __tailMerge_<lib>
 324 };
 325
 326 static const uint8_t tailMergeARM64[] = {
 327     0xfd, 0x7b, 0xb3, 0xa9, // stp     x29, x30, [sp, #-208]!
 328     0xfd, 0x03, 0x00, 0x91, // mov     x29, sp
 329     0xe0, 0x07, 0x01, 0xa9, // stp     x0, x1, [sp, #16]
 330     0xe2, 0x0f, 0x02, 0xa9, // stp     x2, x3, [sp, #32]
 331     0xe4, 0x17, 0x03, 0xa9, // stp     x4, x5, [sp, #48]
 332     0xe6, 0x1f, 0x04, 0xa9, // stp     x6, x7, [sp, #64]
 333     0xe0, 0x87, 0x02, 0xad, // stp     q0, q1, [sp, #80]
 334     0xe2, 0x8f, 0x03, 0xad, // stp     q2, q3, [sp, #112]
 335     0xe4, 0x97, 0x04, 0xad, // stp     q4, q5, [sp, #144]
 336     0xe6, 0x9f, 0x05, 0xad, // stp     q6, q7, [sp, #176]
 337     0xe1, 0x03, 0x11, 0xaa, // mov     x1, x17
 338     0x00, 0x00, 0x00, 0x90, // adrp    x0, #0     DELAY_IMPORT_DESCRIPTOR
 339     0x00, 0x00, 0x00, 0x91, // add     x0, x0, #0 :lo12:DELAY_IMPORT_DESCRIPTOR
 340     0x00, 0x00, 0x00, 0x94, // bl      #0 __delayLoadHelper2
 341     0xf0, 0x03, 0x00, 0xaa, // mov     x16, x0
 342     0xe6, 0x9f, 0x45, 0xad, // ldp     q6, q7, [sp, #176]
 343     0xe4, 0x97, 0x44, 0xad, // ldp     q4, q5, [sp, #144]
 344     0xe2, 0x8f, 0x43, 0xad, // ldp     q2, q3, [sp, #112]
 345     0xe0, 0x87, 0x42, 0xad, // ldp     q0, q1, [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, 0xcd, 0xa8, // ldp     x29, x30, [sp], #208
 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(buf, thunkX64, sizeof(thunkX64));
 364     write32le(buf + 3, imp->getRVA() - rva - 7);
 365     write32le(buf + 8, 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(buf, tailMergeX64, sizeof(tailMergeX64));
 381     write32le(buf + 39, desc->getRVA() - rva - 43);
 382     write32le(buf + 44, helper->getRVA() - rva - 48);
 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(buf + 0, tm->getRVA()); // TailMergeChunk start RVA
 402     write32le(buf + 4, tm->getRVA() + tm->getSize()); // TailMergeChunk stop RVA
 403     write32le(buf + 8, 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(buf, tailMergeUnwindInfoX64, 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(buf, thunkX86, sizeof(thunkX86));
 436     write32le(buf + 1, imp->getRVA() + ctx.config.imageBase);
 437     write32le(buf + 6, tailMerge->getRVA() - rva - 10);
 438   }
 439
 440   void getBaserels(std::vector<Baserel> *res) override {
 441     res->emplace_back(rva + 1, 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(buf, tailMergeX86, sizeof(tailMergeX86));
 461     write32le(buf + 4, desc->getRVA() + ctx.config.imageBase);
 462     write32le(buf + 9, helper->getRVA() - rva - 13);
 463   }
 464
 465   void getBaserels(std::vector<Baserel> *res) override {
 466     res->emplace_back(rva + 4, 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(buf, thunkARM, sizeof(thunkARM));
 488     applyMOV32T(buf + 0, imp->getRVA() + ctx.config.imageBase);
 489     applyBranch24T(buf + 8, tailMerge->getRVA() - rva - 12);
 490   }
 491
 492   void getBaserels(std::vector<Baserel> *res) override {
 493     res->emplace_back(rva + 0, 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(buf, tailMergeARM, sizeof(tailMergeARM));
 515     applyMOV32T(buf + 14, desc->getRVA() + ctx.config.imageBase);
 516     applyBranch24T(buf + 22, helper->getRVA() - rva - 26);
 517   }
 518
 519   void getBaserels(std::vector<Baserel> *res) override {
 520     res->emplace_back(rva + 14, 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(buf, thunkARM64, sizeof(thunkARM64));
 541     applyArm64Addr(buf + 0, imp->getRVA(), rva + 0, 12);
 542     applyArm64Imm(buf + 4, imp->getRVA() & 0xfff, 0);
 543     applyArm64Branch26(buf + 8, 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(buf, tailMergeARM64, sizeof(tailMergeARM64));
 561     applyArm64Addr(buf + 44, desc->getRVA(), rva + 44, 12);
 562     applyArm64Imm(buf + 48, desc->getRVA() & 0xfff, 0);
 563     applyArm64Branch26(buf + 52, helper->getRVA() - rva - 52);
 564   }
 565
 566   Chunk *desc = nullptr;
 567   Defined *helper = nullptr;
 568 };
 569
 570 // A chunk for the import descriptor table.
 571 class DelayAddressChunk : public NonSectionChunk {
 572 public:
 573   explicit DelayAddressChunk(COFFLinkerContext &ctx, Chunk *c)
 574       : thunk(c), ctx(ctx) {
 575     setAlignment(ctx.config.wordsize);
 576   }
 577   size_t getSize() const override { return ctx.config.wordsize; }
 578
 579   void writeTo(uint8_t *buf) const override {
 580     if (ctx.config.is64()) {
 581       write64le(buf, thunk->getRVA() + ctx.config.imageBase);
 582     } else {
 583       uint32_t bit = 0;
 584       // Pointer to thumb code must have the LSB set, so adjust it.
 585       if (ctx.config.machine == ARMNT)
 586         bit = 1;
 587       write32le(buf, (thunk->getRVA() + ctx.config.imageBase) | bit);
 588     }
 589   }
 590
 591   void getBaserels(std::vector<Baserel> *res) override {
 592     res->emplace_back(rva, ctx.config.machine);
 593   }
 594
 595   Chunk *thunk;
 596
 597 private:
 598   const COFFLinkerContext &ctx;
 599 };
 600
 601 // Export table
 602 // Read Microsoft PE/COFF spec 5.3 for details.
 603
 604 // A chunk for the export descriptor table.
 605 class ExportDirectoryChunk : public NonSectionChunk {
 606 public:
 607   ExportDirectoryChunk(int baseOrdinal, int maxOrdinal, int nameTabSize,
 608                        Chunk *d, Chunk *a, Chunk *n, Chunk *o)
 609       : baseOrdinal(baseOrdinal), maxOrdinal(maxOrdinal),
 610         nameTabSize(nameTabSize), dllName(d), addressTab(a), nameTab(n),
 611         ordinalTab(o) {}
 612
 613   size_t getSize() const override {
 614     return sizeof(export_directory_table_entry);
 615   }
 616
 617   void writeTo(uint8_t *buf) const override {
 618     memset(buf, 0, getSize());
 619
 620     auto *e = (export_directory_table_entry *)(buf);
 621     e->NameRVA = dllName->getRVA();
 622     e->OrdinalBase = baseOrdinal;
 623     e->AddressTableEntries = (maxOrdinal - baseOrdinal) + 1;
 624     e->NumberOfNamePointers = nameTabSize;
 625     e->ExportAddressTableRVA = addressTab->getRVA();
 626     e->NamePointerRVA = nameTab->getRVA();
 627     e->OrdinalTableRVA = ordinalTab->getRVA();
 628   }
 629
 630   uint16_t baseOrdinal;
 631   uint16_t maxOrdinal;
 632   uint16_t nameTabSize;
 633   Chunk *dllName;
 634   Chunk *addressTab;
 635   Chunk *nameTab;
 636   Chunk *ordinalTab;
 637 };
 638
 639 class AddressTableChunk : public NonSectionChunk {
 640 public:
 641   explicit AddressTableChunk(COFFLinkerContext &ctx, size_t baseOrdinal,
 642                              size_t maxOrdinal)
 643       : baseOrdinal(baseOrdinal), size((maxOrdinal - baseOrdinal) + 1),
 644         ctx(ctx) {}
 645   size_t getSize() const override { return size * 4; }
 646
 647   void writeTo(uint8_t *buf) const override {
 648     memset(buf, 0, getSize());
 649
 650     for (const Export &e : ctx.config.exports) {
 651       assert(e.ordinal >= baseOrdinal && "Export symbol has invalid ordinal");
 652       // Subtract the OrdinalBase to get the index.
 653       uint8_t *p = buf + (e.ordinal - baseOrdinal) * 4;
 654       uint32_t bit = 0;
 655       // Pointer to thumb code must have the LSB set, so adjust it.
 656       if (ctx.config.machine == ARMNT && !e.data)
 657         bit = 1;
 658       if (e.forwardChunk) {
 659         write32le(p, e.forwardChunk->getRVA() | bit);
 660       } else {
 661         assert(cast<Defined>(e.sym)->getRVA() != 0 &&
 662                "Exported symbol unmapped");
 663         write32le(p, cast<Defined>(e.sym)->getRVA() | bit);
 664       }
 665     }
 666   }
 667
 668 private:
 669   size_t baseOrdinal;
 670   size_t size;
 671   const COFFLinkerContext &ctx;
 672 };
 673
 674 class NamePointersChunk : public NonSectionChunk {
 675 public:
 676   explicit NamePointersChunk(std::vector<Chunk *> &v) : chunks(v) {}
 677   size_t getSize() const override { return chunks.size() * 4; }
 678
 679   void writeTo(uint8_t *buf) const override {
 680     for (Chunk *c : chunks) {
 681       write32le(buf, c->getRVA());
 682       buf += 4;
 683     }
 684   }
 685
 686 private:
 687   std::vector<Chunk *> chunks;
 688 };
 689
 690 class ExportOrdinalChunk : public NonSectionChunk {
 691 public:
 692   explicit ExportOrdinalChunk(const COFFLinkerContext &ctx, size_t baseOrdinal,
 693                               size_t tableSize)
 694       : baseOrdinal(baseOrdinal), size(tableSize), ctx(ctx) {}
 695   size_t getSize() const override { return size * 2; }
 696
 697   void writeTo(uint8_t *buf) const override {
 698     for (const Export &e : ctx.config.exports) {
 699       if (e.noname)
 700         continue;
 701       assert(e.ordinal >= baseOrdinal && "Export symbol has invalid ordinal");
 702       // This table stores unbiased indices, so subtract OrdinalBase.
 703       write16le(buf, e.ordinal - baseOrdinal);
 704       buf += 2;
 705     }
 706   }
 707
 708 private:
 709   size_t baseOrdinal;
 710   size_t size;
 711   const COFFLinkerContext &ctx;
 712 };
 713
 714 } // anonymous namespace
 715
 716 void IdataContents::create(COFFLinkerContext &ctx) {
 717   std::vector<std::vector<DefinedImportData *>> v = binImports(ctx, imports);
 718
 719   // Create .idata contents for each DLL.
 720   for (std::vector<DefinedImportData *> &syms : v) {
 721     // Create lookup and address tables. If they have external names,
 722     // we need to create hintName chunks to store the names.
 723     // If they don't (if they are import-by-ordinals), we store only
 724     // ordinal values to the table.
 725     size_t base = lookups.size();
 726     for (DefinedImportData *s : syms) {
 727       uint16_t ord = s->getOrdinal();
 728       if (s->getExternalName().empty()) {
 729         lookups.push_back(make<OrdinalOnlyChunk>(ctx, ord));
 730         addresses.push_back(make<OrdinalOnlyChunk>(ctx, ord));
 731       } else {
 732         auto *c = make<HintNameChunk>(s->getExternalName(), ord);
 733         lookups.push_back(make<LookupChunk>(ctx, c));
 734         addresses.push_back(make<LookupChunk>(ctx, c));
 735         hints.push_back(c);
 736       }
 737
 738       if (s->file->impECSym) {
 739         auto chunk = make<AuxImportChunk>(s->file);
 740         auxIat.push_back(chunk);
 741         s->file->impECSym->setLocation(chunk);
 742
 743         chunk = make<AuxImportChunk>(s->file);
 744         auxIatCopy.push_back(chunk);
 745         s->file->auxImpCopySym->setLocation(chunk);
 746       }
 747     }
 748     // Terminate with null values.
 749     lookups.push_back(make<NullChunk>(ctx));
 750     addresses.push_back(make<NullChunk>(ctx));
 751     if (ctx.config.machine == ARM64EC) {
 752       auxIat.push_back(make<NullChunk>(ctx));
 753       auxIatCopy.push_back(make<NullChunk>(ctx));
 754     }
 755
 756     for (int i = 0, e = syms.size(); i < e; ++i)
 757       syms[i]->setLocation(addresses[base + i]);
 758
 759     // Create the import table header.
 760     dllNames.push_back(make<StringChunk>(syms[0]->getDLLName()));
 761     auto *dir = make<ImportDirectoryChunk>(dllNames.back());
 762     dir->lookupTab = lookups[base];
 763     dir->addressTab = addresses[base];
 764     dirs.push_back(dir);
 765   }
 766   // Add null terminator.
 767   dirs.push_back(make<NullChunk>(sizeof(ImportDirectoryTableEntry), 4));
 768 }
 769
 770 std::vector<Chunk *> DelayLoadContents::getChunks() {
 771   std::vector<Chunk *> v;
 772   v.insert(v.end(), dirs.begin(), dirs.end());
 773   v.insert(v.end(), names.begin(), names.end());
 774   v.insert(v.end(), hintNames.begin(), hintNames.end());
 775   v.insert(v.end(), dllNames.begin(), dllNames.end());
 776   return v;
 777 }
 778
 779 std::vector<Chunk *> DelayLoadContents::getDataChunks() {
 780   std::vector<Chunk *> v;
 781   v.insert(v.end(), moduleHandles.begin(), moduleHandles.end());
 782   v.insert(v.end(), addresses.begin(), addresses.end());
 783   return v;
 784 }
 785
 786 uint64_t DelayLoadContents::getDirSize() {
 787   return dirs.size() * sizeof(delay_import_directory_table_entry);
 788 }
 789
 790 void DelayLoadContents::create(Defined *h) {
 791   helper = h;
 792   std::vector<std::vector<DefinedImportData *>> v = binImports(ctx, imports);
 793
 794   Chunk *unwind = newTailMergeUnwindInfoChunk();
 795
 796   // Create .didat contents for each DLL.
 797   for (std::vector<DefinedImportData *> &syms : v) {
 798     // Create the delay import table header.
 799     dllNames.push_back(make<StringChunk>(syms[0]->getDLLName()));
 800     auto *dir = make<DelayDirectoryChunk>(dllNames.back());
 801
 802     size_t base = addresses.size();
 803     Chunk *tm = newTailMergeChunk(dir);
 804     Chunk *pdataChunk = unwind ? newTailMergePDataChunk(tm, unwind) : nullptr;
 805     for (DefinedImportData *s : syms) {
 806       Chunk *t = newThunkChunk(s, tm);
 807       auto *a = make<DelayAddressChunk>(ctx, t);
 808       addresses.push_back(a);
 809       thunks.push_back(t);
 810       StringRef extName = s->getExternalName();
 811       if (extName.empty()) {
 812         names.push_back(make<OrdinalOnlyChunk>(ctx, s->getOrdinal()));
 813       } else {
 814         auto *c = make<HintNameChunk>(extName, 0);
 815         names.push_back(make<LookupChunk>(ctx, c));
 816         hintNames.push_back(c);
 817         // Add a synthetic symbol for this load thunk, using the "__imp___load"
 818         // prefix, in case this thunk needs to be added to the list of valid
 819         // call targets for Control Flow Guard.
 820         StringRef symName = saver().save("__imp___load_" + extName);
 821         s->loadThunkSym =
 822             cast<DefinedSynthetic>(ctx.symtab.addSynthetic(symName, t));
 823       }
 824
 825       if (s->file->impECSym) {
 826         auto chunk = make<AuxImportChunk>(s->file);
 827         auxIat.push_back(chunk);
 828         s->file->impECSym->setLocation(chunk);
 829
 830         chunk = make<AuxImportChunk>(s->file);
 831         auxIatCopy.push_back(chunk);
 832         s->file->auxImpCopySym->setLocation(chunk);
 833       }
 834     }
 835     thunks.push_back(tm);
 836     if (pdataChunk)
 837       pdata.push_back(pdataChunk);
 838     StringRef tmName =
 839         saver().save("__tailMerge_" + syms[0]->getDLLName().lower());
 840     ctx.symtab.addSynthetic(tmName, tm);
 841     // Terminate with null values.
 842     addresses.push_back(make<NullChunk>(ctx, 8));
 843     names.push_back(make<NullChunk>(ctx, 8));
 844     if (ctx.config.machine == ARM64EC) {
 845       auxIat.push_back(make<NullChunk>(ctx, 8));
 846       auxIatCopy.push_back(make<NullChunk>(ctx, 8));
 847     }
 848
 849     for (int i = 0, e = syms.size(); i < e; ++i)
 850       syms[i]->setLocation(addresses[base + i]);
 851     auto *mh = make<NullChunk>(8, 8);
 852     moduleHandles.push_back(mh);
 853
 854     // Fill the delay import table header fields.
 855     dir->moduleHandle = mh;
 856     dir->addressTab = addresses[base];
 857     dir->nameTab = names[base];
 858     dirs.push_back(dir);
 859   }
 860
 861   if (unwind)
 862     unwindinfo.push_back(unwind);
 863   // Add null terminator.
 864   dirs.push_back(
 865       make<NullChunk>(sizeof(delay_import_directory_table_entry), 4));
 866 }
 867
 868 Chunk *DelayLoadContents::newTailMergeChunk(Chunk *dir) {
 869   switch (ctx.config.machine) {
 870   case AMD64:
 871   case ARM64EC:
 872     return make<TailMergeChunkX64>(dir, helper);
 873   case I386:
 874     return make<TailMergeChunkX86>(ctx, dir, helper);
 875   case ARMNT:
 876     return make<TailMergeChunkARM>(ctx, dir, helper);
 877   case ARM64:
 878     return make<TailMergeChunkARM64>(dir, helper);
 879   default:
 880     llvm_unreachable("unsupported machine type");
 881   }
 882 }
 883
 884 Chunk *DelayLoadContents::newTailMergeUnwindInfoChunk() {
 885   switch (ctx.config.machine) {
 886   case AMD64:
 887   case ARM64EC:
 888     return make<TailMergeUnwindInfoX64>();
 889     // FIXME: Add support for other architectures.
 890   default:
 891     return nullptr; // Just don't generate unwind info.
 892   }
 893 }
 894 Chunk *DelayLoadContents::newTailMergePDataChunk(Chunk *tm, Chunk *unwind) {
 895   switch (ctx.config.machine) {
 896   case AMD64:
 897   case ARM64EC:
 898     return make<TailMergePDataChunkX64>(tm, unwind);
 899     // FIXME: Add support for other architectures.
 900   default:
 901     return nullptr; // Just don't generate unwind info.
 902   }
 903 }
 904
 905 Chunk *DelayLoadContents::newThunkChunk(DefinedImportData *s,
 906                                         Chunk *tailMerge) {
 907   switch (ctx.config.machine) {
 908   case AMD64:
 909   case ARM64EC:
 910     return make<ThunkChunkX64>(s, tailMerge);
 911   case I386:
 912     return make<ThunkChunkX86>(ctx, s, tailMerge);
 913   case ARMNT:
 914     return make<ThunkChunkARM>(ctx, s, tailMerge);
 915   case ARM64:
 916     return make<ThunkChunkARM64>(s, tailMerge);
 917   default:
 918     llvm_unreachable("unsupported machine type");
 919   }
 920 }
 921
 922 EdataContents::EdataContents(COFFLinkerContext &ctx) : ctx(ctx) {
 923   unsigned baseOrdinal = 1 << 16, maxOrdinal = 0;
 924   for (Export &e : ctx.config.exports) {
 925     baseOrdinal = std::min(baseOrdinal, (unsigned)e.ordinal);
 926     maxOrdinal = std::max(maxOrdinal, (unsigned)e.ordinal);
 927   }
 928   // Ordinals must start at 1 as suggested in:
 929   // https://learn.microsoft.com/en-us/cpp/build/reference/export-exports-a-function?view=msvc-170
 930   assert(baseOrdinal >= 1);
 931
 932   auto *dllName = make<StringChunk>(sys::path::filename(ctx.config.outputFile));
 933   auto *addressTab = make<AddressTableChunk>(ctx, baseOrdinal, maxOrdinal);
 934   std::vector<Chunk *> names;
 935   for (Export &e : ctx.config.exports)
 936     if (!e.noname)
 937       names.push_back(make<StringChunk>(e.exportName));
 938
 939   std::vector<Chunk *> forwards;
 940   for (Export &e : ctx.config.exports) {
 941     if (e.forwardTo.empty())
 942       continue;
 943     e.forwardChunk = make<StringChunk>(e.forwardTo);
 944     forwards.push_back(e.forwardChunk);
 945   }
 946
 947   auto *nameTab = make<NamePointersChunk>(names);
 948   auto *ordinalTab = make<ExportOrdinalChunk>(ctx, baseOrdinal, names.size());
 949   auto *dir =
 950       make<ExportDirectoryChunk>(baseOrdinal, maxOrdinal, names.size(), dllName,
 951                                  addressTab, nameTab, ordinalTab);
 952   chunks.push_back(dir);
 953   chunks.push_back(dllName);
 954   chunks.push_back(addressTab);
 955   chunks.push_back(nameTab);
 956   chunks.push_back(ordinalTab);
 957   chunks.insert(chunks.end(), names.begin(), names.end());
 958   chunks.insert(chunks.end(), forwards.begin(), forwards.end());
 959 }
 960
 961 } // namespace lld::coff