1 //===- PPC.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 #include "OutputSections.h" 10 #include "Symbols.h" 11 #include "SyntheticSections.h" 12 #include "Target.h" 13 #include "Thunks.h" 14 #include "lld/Common/ErrorHandler.h" 15 #include "llvm/Support/Endian.h" 16 17 using namespace llvm; 18 using namespace llvm::support::endian; 19 using namespace llvm::ELF; 20 21 namespace lld { 22 namespace elf { 23 24 namespace { 25 class PPC final : public TargetInfo { 26 public: 27 PPC(); 28 RelExpr getRelExpr(RelType type, const Symbol &s, 29 const uint8_t *loc) const override; 30 RelType getDynRel(RelType type) const override; 31 void writeGotHeader(uint8_t *buf) const override; 32 void writePltHeader(uint8_t *buf) const override { 33 llvm_unreachable("should call writePPC32GlinkSection() instead"); 34 } 35 void writePlt(uint8_t *buf, const Symbol &sym, 36 uint64_t pltEntryAddr) const override { 37 llvm_unreachable("should call writePPC32GlinkSection() instead"); 38 } 39 void writeIplt(uint8_t *buf, const Symbol &sym, 40 uint64_t pltEntryAddr) const override; 41 void writeGotPlt(uint8_t *buf, const Symbol &s) const override; 42 bool needsThunk(RelExpr expr, RelType relocType, const InputFile *file, 43 uint64_t branchAddr, const Symbol &s, 44 int64_t a) const override; 45 uint32_t getThunkSectionSpacing() const override; 46 bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override; 47 void relocate(uint8_t *loc, const Relocation &rel, 48 uint64_t val) const override; 49 RelExpr adjustRelaxExpr(RelType type, const uint8_t *data, 50 RelExpr expr) const override; 51 int getTlsGdRelaxSkip(RelType type) const override; 52 void relaxTlsGdToIe(uint8_t *loc, const Relocation &rel, 53 uint64_t val) const override; 54 void relaxTlsGdToLe(uint8_t *loc, const Relocation &rel, 55 uint64_t val) const override; 56 void relaxTlsLdToLe(uint8_t *loc, const Relocation &rel, 57 uint64_t val) const override; 58 void relaxTlsIeToLe(uint8_t *loc, const Relocation &rel, 59 uint64_t val) const override; 60 }; 61 } // namespace 62 63 static uint16_t lo(uint32_t v) { return v; } 64 static uint16_t ha(uint32_t v) { return (v + 0x8000) >> 16; } 65 66 static uint32_t readFromHalf16(const uint8_t *loc) { 67 return read32(config->isLE ? loc : loc - 2); 68 } 69 70 static void writeFromHalf16(uint8_t *loc, uint32_t insn) { 71 write32(config->isLE ? loc : loc - 2, insn); 72 } 73 74 void writePPC32GlinkSection(uint8_t *buf, size_t numEntries) { 75 // Create canonical PLT entries for non-PIE code. Compilers don't generate 76 // non-GOT-non-PLT relocations referencing external functions for -fpie/-fPIE. 77 uint32_t glink = in.plt->getVA(); // VA of .glink 78 if (!config->isPic) { 79 for (const Symbol *sym : in.plt->entries) 80 if (sym->needsPltAddr) { 81 writePPC32PltCallStub(buf, sym->getGotPltVA(), nullptr, 0); 82 buf += 16; 83 glink += 16; 84 } 85 } 86 87 // On PPC Secure PLT ABI, bl foo@plt jumps to a call stub, which loads an 88 // absolute address from a specific .plt slot (usually called .got.plt on 89 // other targets) and jumps there. 90 // 91 // a) With immediate binding (BIND_NOW), the .plt entry is resolved at load 92 // time. The .glink section is not used. 93 // b) With lazy binding, the .plt entry points to a `b PLTresolve` 94 // instruction in .glink, filled in by PPC::writeGotPlt(). 95 96 // Write N `b PLTresolve` first. 97 for (size_t i = 0; i != numEntries; ++i) 98 write32(buf + 4 * i, 0x48000000 | 4 * (numEntries - i)); 99 buf += 4 * numEntries; 100 101 // Then write PLTresolve(), which has two forms: PIC and non-PIC. PLTresolve() 102 // computes the PLT index (by computing the distance from the landing b to 103 // itself) and calls _dl_runtime_resolve() (in glibc). 104 uint32_t got = in.got->getVA(); 105 const uint8_t *end = buf + 64; 106 if (config->isPic) { 107 uint32_t afterBcl = 4 * in.plt->getNumEntries() + 12; 108 uint32_t gotBcl = got + 4 - (glink + afterBcl); 109 write32(buf + 0, 0x3d6b0000 | ha(afterBcl)); // addis r11,r11,1f-glink@ha 110 write32(buf + 4, 0x7c0802a6); // mflr r0 111 write32(buf + 8, 0x429f0005); // bcl 20,30,.+4 112 write32(buf + 12, 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-glink@l 113 write32(buf + 16, 0x7d8802a6); // mflr r12 114 write32(buf + 20, 0x7c0803a6); // mtlr r0 115 write32(buf + 24, 0x7d6c5850); // sub r11,r11,r12 116 write32(buf + 28, 0x3d8c0000 | ha(gotBcl)); // addis 12,12,GOT+4-1b@ha 117 if (ha(gotBcl) == ha(gotBcl + 4)) { 118 write32(buf + 32, 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12) 119 write32(buf + 36, 120 0x818c0000 | lo(gotBcl + 4)); // lwz r12,r12,GOT+8-1b@l(r12) 121 } else { 122 write32(buf + 32, 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12) 123 write32(buf + 36, 0x818c0000 | 4); // lwz r12,r12,4(r12) 124 } 125 write32(buf + 40, 0x7c0903a6); // mtctr 0 126 write32(buf + 44, 0x7c0b5a14); // add r0,11,11 127 write32(buf + 48, 0x7d605a14); // add r11,0,11 128 write32(buf + 52, 0x4e800420); // bctr 129 buf += 56; 130 } else { 131 write32(buf + 0, 0x3d800000 | ha(got + 4)); // lis r12,GOT+4@ha 132 write32(buf + 4, 0x3d6b0000 | ha(-glink)); // addis r11,r11,-glink@ha 133 if (ha(got + 4) == ha(got + 8)) 134 write32(buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12) 135 else 136 write32(buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12) 137 write32(buf + 12, 0x396b0000 | lo(-glink)); // addi r11,r11,-glink@l 138 write32(buf + 16, 0x7c0903a6); // mtctr r0 139 write32(buf + 20, 0x7c0b5a14); // add r0,r11,r11 140 if (ha(got + 4) == ha(got + 8)) 141 write32(buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@l(r12) 142 else 143 write32(buf + 24, 0x818c0000 | 4); // lwz r12,4(r12) 144 write32(buf + 28, 0x7d605a14); // add r11,r0,r11 145 write32(buf + 32, 0x4e800420); // bctr 146 buf += 36; 147 } 148 149 // Pad with nop. They should not be executed. 150 for (; buf < end; buf += 4) 151 write32(buf, 0x60000000); 152 } 153 154 PPC::PPC() { 155 copyRel = R_PPC_COPY; 156 gotRel = R_PPC_GLOB_DAT; 157 noneRel = R_PPC_NONE; 158 pltRel = R_PPC_JMP_SLOT; 159 relativeRel = R_PPC_RELATIVE; 160 iRelativeRel = R_PPC_IRELATIVE; 161 symbolicRel = R_PPC_ADDR32; 162 gotBaseSymInGotPlt = false; 163 gotHeaderEntriesNum = 3; 164 gotPltHeaderEntriesNum = 0; 165 pltHeaderSize = 0; 166 pltEntrySize = 4; 167 ipltEntrySize = 16; 168 169 needsThunks = true; 170 171 tlsModuleIndexRel = R_PPC_DTPMOD32; 172 tlsOffsetRel = R_PPC_DTPREL32; 173 tlsGotRel = R_PPC_TPREL32; 174 175 defaultMaxPageSize = 65536; 176 defaultImageBase = 0x10000000; 177 178 write32(trapInstr.data(), 0x7fe00008); 179 } 180 181 void PPC::writeIplt(uint8_t *buf, const Symbol &sym, 182 uint64_t /*pltEntryAddr*/) const { 183 // In -pie or -shared mode, assume r30 points to .got2+0x8000, and use a 184 // .got2.plt_pic32. thunk. 185 writePPC32PltCallStub(buf, sym.getGotPltVA(), sym.file, 0x8000); 186 } 187 188 void PPC::writeGotHeader(uint8_t *buf) const { 189 // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC 190 // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1], 191 // link_map in _GLOBAL_OFFSET_TABLE_[2]. 192 write32(buf, mainPart->dynamic->getVA()); 193 } 194 195 void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const { 196 // Address of the symbol resolver stub in .glink . 197 write32(buf, in.plt->getVA() + in.plt->headerSize + 4 * s.pltIndex); 198 } 199 200 bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file, 201 uint64_t branchAddr, const Symbol &s, int64_t a) const { 202 if (type != R_PPC_LOCAL24PC && type != R_PPC_REL24 && type != R_PPC_PLTREL24) 203 return false; 204 if (s.isInPlt()) 205 return true; 206 if (s.isUndefWeak()) 207 return false; 208 return !PPC::inBranchRange(type, branchAddr, s.getVA(a)); 209 } 210 211 uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; } 212 213 bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const { 214 uint64_t offset = dst - src; 215 if (type == R_PPC_LOCAL24PC || type == R_PPC_REL24 || type == R_PPC_PLTREL24) 216 return isInt<26>(offset); 217 llvm_unreachable("unsupported relocation type used in branch"); 218 } 219 220 RelExpr PPC::getRelExpr(RelType type, const Symbol &s, 221 const uint8_t *loc) const { 222 switch (type) { 223 case R_PPC_NONE: 224 return R_NONE; 225 case R_PPC_ADDR16_HA: 226 case R_PPC_ADDR16_HI: 227 case R_PPC_ADDR16_LO: 228 case R_PPC_ADDR32: 229 return R_ABS; 230 case R_PPC_DTPREL16: 231 case R_PPC_DTPREL16_HA: 232 case R_PPC_DTPREL16_HI: 233 case R_PPC_DTPREL16_LO: 234 case R_PPC_DTPREL32: 235 return R_DTPREL; 236 case R_PPC_REL14: 237 case R_PPC_REL32: 238 case R_PPC_REL16_LO: 239 case R_PPC_REL16_HI: 240 case R_PPC_REL16_HA: 241 return R_PC; 242 case R_PPC_GOT16: 243 return R_GOT_OFF; 244 case R_PPC_LOCAL24PC: 245 case R_PPC_REL24: 246 return R_PLT_PC; 247 case R_PPC_PLTREL24: 248 return R_PPC32_PLTREL; 249 case R_PPC_GOT_TLSGD16: 250 return R_TLSGD_GOT; 251 case R_PPC_GOT_TLSLD16: 252 return R_TLSLD_GOT; 253 case R_PPC_GOT_TPREL16: 254 return R_GOT_OFF; 255 case R_PPC_TLS: 256 return R_TLSIE_HINT; 257 case R_PPC_TLSGD: 258 return R_TLSDESC_CALL; 259 case R_PPC_TLSLD: 260 return R_TLSLD_HINT; 261 case R_PPC_TPREL16: 262 case R_PPC_TPREL16_HA: 263 case R_PPC_TPREL16_LO: 264 case R_PPC_TPREL16_HI: 265 return R_TLS; 266 default: 267 error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) + 268 ") against symbol " + toString(s)); 269 return R_NONE; 270 } 271 } 272 273 RelType PPC::getDynRel(RelType type) const { 274 if (type == R_PPC_ADDR32) 275 return type; 276 return R_PPC_NONE; 277 } 278 279 static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) { 280 uint64_t dtpBiasedVal = val - 0x8000; 281 switch (type) { 282 case R_PPC_DTPREL16: 283 return {R_PPC64_ADDR16, dtpBiasedVal}; 284 case R_PPC_DTPREL16_HA: 285 return {R_PPC_ADDR16_HA, dtpBiasedVal}; 286 case R_PPC_DTPREL16_HI: 287 return {R_PPC_ADDR16_HI, dtpBiasedVal}; 288 case R_PPC_DTPREL16_LO: 289 return {R_PPC_ADDR16_LO, dtpBiasedVal}; 290 case R_PPC_DTPREL32: 291 return {R_PPC_ADDR32, dtpBiasedVal}; 292 default: 293 return {type, val}; 294 } 295 } 296 297 void PPC::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const { 298 RelType newType; 299 std::tie(newType, val) = fromDTPREL(rel.type, val); 300 switch (newType) { 301 case R_PPC_ADDR16: 302 checkIntUInt(loc, val, 16, rel); 303 write16(loc, val); 304 break; 305 case R_PPC_GOT16: 306 case R_PPC_GOT_TLSGD16: 307 case R_PPC_GOT_TLSLD16: 308 case R_PPC_GOT_TPREL16: 309 case R_PPC_TPREL16: 310 checkInt(loc, val, 16, rel); 311 write16(loc, val); 312 break; 313 case R_PPC_ADDR16_HA: 314 case R_PPC_DTPREL16_HA: 315 case R_PPC_GOT_TLSGD16_HA: 316 case R_PPC_GOT_TLSLD16_HA: 317 case R_PPC_GOT_TPREL16_HA: 318 case R_PPC_REL16_HA: 319 case R_PPC_TPREL16_HA: 320 write16(loc, ha(val)); 321 break; 322 case R_PPC_ADDR16_HI: 323 case R_PPC_DTPREL16_HI: 324 case R_PPC_GOT_TLSGD16_HI: 325 case R_PPC_GOT_TLSLD16_HI: 326 case R_PPC_GOT_TPREL16_HI: 327 case R_PPC_REL16_HI: 328 case R_PPC_TPREL16_HI: 329 write16(loc, val >> 16); 330 break; 331 case R_PPC_ADDR16_LO: 332 case R_PPC_DTPREL16_LO: 333 case R_PPC_GOT_TLSGD16_LO: 334 case R_PPC_GOT_TLSLD16_LO: 335 case R_PPC_GOT_TPREL16_LO: 336 case R_PPC_REL16_LO: 337 case R_PPC_TPREL16_LO: 338 write16(loc, val); 339 break; 340 case R_PPC_ADDR32: 341 case R_PPC_REL32: 342 write32(loc, val); 343 break; 344 case R_PPC_REL14: { 345 uint32_t mask = 0x0000FFFC; 346 checkInt(loc, val, 16, rel); 347 checkAlignment(loc, val, 4, rel); 348 write32(loc, (read32(loc) & ~mask) | (val & mask)); 349 break; 350 } 351 case R_PPC_REL24: 352 case R_PPC_LOCAL24PC: 353 case R_PPC_PLTREL24: { 354 uint32_t mask = 0x03FFFFFC; 355 checkInt(loc, val, 26, rel); 356 checkAlignment(loc, val, 4, rel); 357 write32(loc, (read32(loc) & ~mask) | (val & mask)); 358 break; 359 } 360 default: 361 llvm_unreachable("unknown relocation"); 362 } 363 } 364 365 RelExpr PPC::adjustRelaxExpr(RelType type, const uint8_t *data, 366 RelExpr expr) const { 367 if (expr == R_RELAX_TLS_GD_TO_IE) 368 return R_RELAX_TLS_GD_TO_IE_GOT_OFF; 369 if (expr == R_RELAX_TLS_LD_TO_LE) 370 return R_RELAX_TLS_LD_TO_LE_ABS; 371 return expr; 372 } 373 374 int PPC::getTlsGdRelaxSkip(RelType type) const { 375 // A __tls_get_addr call instruction is marked with 2 relocations: 376 // 377 // R_PPC_TLSGD / R_PPC_TLSLD: marker relocation 378 // R_PPC_REL24: __tls_get_addr 379 // 380 // After the relaxation we no longer call __tls_get_addr and should skip both 381 // relocations to not create a false dependence on __tls_get_addr being 382 // defined. 383 if (type == R_PPC_TLSGD || type == R_PPC_TLSLD) 384 return 2; 385 return 1; 386 } 387 388 void PPC::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel, 389 uint64_t val) const { 390 switch (rel.type) { 391 case R_PPC_GOT_TLSGD16: { 392 // addi rT, rA, x@got@tlsgd --> lwz rT, x@got@tprel(rA) 393 uint32_t insn = readFromHalf16(loc); 394 writeFromHalf16(loc, 0x80000000 | (insn & 0x03ff0000)); 395 relocateNoSym(loc, R_PPC_GOT_TPREL16, val); 396 break; 397 } 398 case R_PPC_TLSGD: 399 // bl __tls_get_addr(x@tldgd) --> add r3, r3, r2 400 write32(loc, 0x7c631214); 401 break; 402 default: 403 llvm_unreachable("unsupported relocation for TLS GD to IE relaxation"); 404 } 405 } 406 407 void PPC::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel, 408 uint64_t val) const { 409 switch (rel.type) { 410 case R_PPC_GOT_TLSGD16: 411 // addi r3, r31, x@got@tlsgd --> addis r3, r2, x@tprel@ha 412 writeFromHalf16(loc, 0x3c620000 | ha(val)); 413 break; 414 case R_PPC_TLSGD: 415 // bl __tls_get_addr(x@tldgd) --> add r3, r3, x@tprel@l 416 write32(loc, 0x38630000 | lo(val)); 417 break; 418 default: 419 llvm_unreachable("unsupported relocation for TLS GD to LE relaxation"); 420 } 421 } 422 423 void PPC::relaxTlsLdToLe(uint8_t *loc, const Relocation &rel, 424 uint64_t val) const { 425 switch (rel.type) { 426 case R_PPC_GOT_TLSLD16: 427 // addi r3, rA, x@got@tlsgd --> addis r3, r2, 0 428 writeFromHalf16(loc, 0x3c620000); 429 break; 430 case R_PPC_TLSLD: 431 // r3+x@dtprel computes r3+x-0x8000, while we want it to compute r3+x@tprel 432 // = r3+x-0x7000, so add 4096 to r3. 433 // bl __tls_get_addr(x@tlsld) --> addi r3, r3, 4096 434 write32(loc, 0x38631000); 435 break; 436 case R_PPC_DTPREL16: 437 case R_PPC_DTPREL16_HA: 438 case R_PPC_DTPREL16_HI: 439 case R_PPC_DTPREL16_LO: 440 relocate(loc, rel, val); 441 break; 442 default: 443 llvm_unreachable("unsupported relocation for TLS LD to LE relaxation"); 444 } 445 } 446 447 void PPC::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel, 448 uint64_t val) const { 449 switch (rel.type) { 450 case R_PPC_GOT_TPREL16: { 451 // lwz rT, x@got@tprel(rA) --> addis rT, r2, x@tprel@ha 452 uint32_t rt = readFromHalf16(loc) & 0x03e00000; 453 writeFromHalf16(loc, 0x3c020000 | rt | ha(val)); 454 break; 455 } 456 case R_PPC_TLS: { 457 uint32_t insn = read32(loc); 458 if (insn >> 26 != 31) 459 error("unrecognized instruction for IE to LE R_PPC_TLS"); 460 // addi rT, rT, x@tls --> addi rT, rT, x@tprel@l 461 uint32_t dFormOp = getPPCDFormOp((read32(loc) & 0x000007fe) >> 1); 462 if (dFormOp == 0) 463 error("unrecognized instruction for IE to LE R_PPC_TLS"); 464 write32(loc, (dFormOp << 26) | (insn & 0x03ff0000) | lo(val)); 465 break; 466 } 467 default: 468 llvm_unreachable("unsupported relocation for TLS IE to LE relaxation"); 469 } 470 } 471 472 TargetInfo *getPPCTargetInfo() { 473 static PPC target; 474 return ⌖ 475 } 476 477 } // namespace elf 478 } // namespace lld 479