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 "lld/Common/ErrorHandler.h" 14 #include "llvm/Support/Endian.h" 15 16 using namespace llvm; 17 using namespace llvm::support::endian; 18 using namespace llvm::ELF; 19 20 namespace lld { 21 namespace elf { 22 23 namespace { 24 class PPC final : public TargetInfo { 25 public: 26 PPC(); 27 RelExpr getRelExpr(RelType type, const Symbol &s, 28 const uint8_t *loc) const override; 29 RelType getDynRel(RelType type) const override; 30 void writeGotHeader(uint8_t *buf) const override; 31 void writePltHeader(uint8_t *buf) const override { 32 llvm_unreachable("should call writePPC32GlinkSection() instead"); 33 } 34 void writePlt(uint8_t *buf, uint64_t gotPltEntryAddr, uint64_t pltEntryAddr, 35 int32_t index, unsigned relOff) const override { 36 llvm_unreachable("should call writePPC32GlinkSection() instead"); 37 } 38 void writeGotPlt(uint8_t *buf, const Symbol &s) const override; 39 bool needsThunk(RelExpr expr, RelType relocType, const InputFile *file, 40 uint64_t branchAddr, const Symbol &s) const override; 41 uint32_t getThunkSectionSpacing() const override; 42 bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override; 43 void relocateOne(uint8_t *loc, RelType type, uint64_t val) const override; 44 RelExpr adjustRelaxExpr(RelType type, const uint8_t *data, 45 RelExpr expr) const override; 46 int getTlsGdRelaxSkip(RelType type) const override; 47 void relaxTlsGdToIe(uint8_t *loc, RelType type, uint64_t val) const override; 48 void relaxTlsGdToLe(uint8_t *loc, RelType type, uint64_t val) const override; 49 void relaxTlsLdToLe(uint8_t *loc, RelType type, uint64_t val) const override; 50 void relaxTlsIeToLe(uint8_t *loc, RelType type, uint64_t val) const override; 51 }; 52 } // namespace 53 54 static uint16_t lo(uint32_t v) { return v; } 55 static uint16_t ha(uint32_t v) { return (v + 0x8000) >> 16; } 56 57 static uint32_t readFromHalf16(const uint8_t *loc) { 58 return read32(config->isLE ? loc : loc - 2); 59 } 60 61 static void writeFromHalf16(uint8_t *loc, uint32_t insn) { 62 write32(config->isLE ? loc : loc - 2, insn); 63 } 64 65 void writePPC32GlinkSection(uint8_t *buf, size_t numEntries) { 66 // On PPC Secure PLT ABI, bl foo@plt jumps to a call stub, which loads an 67 // absolute address from a specific .plt slot (usually called .got.plt on 68 // other targets) and jumps there. 69 // 70 // a) With immediate binding (BIND_NOW), the .plt entry is resolved at load 71 // time. The .glink section is not used. 72 // b) With lazy binding, the .plt entry points to a `b PLTresolve` 73 // instruction in .glink, filled in by PPC::writeGotPlt(). 74 75 // Write N `b PLTresolve` first. 76 for (size_t i = 0; i != numEntries; ++i) 77 write32(buf + 4 * i, 0x48000000 | 4 * (numEntries - i)); 78 buf += 4 * numEntries; 79 80 // Then write PLTresolve(), which has two forms: PIC and non-PIC. PLTresolve() 81 // computes the PLT index (by computing the distance from the landing b to 82 // itself) and calls _dl_runtime_resolve() (in glibc). 83 uint32_t got = in.got->getVA(); 84 uint32_t glink = in.plt->getVA(); // VA of .glink 85 const uint8_t *end = buf + 64; 86 if (config->isPic) { 87 uint32_t afterBcl = in.plt->getSize() - target->pltHeaderSize + 12; 88 uint32_t gotBcl = got + 4 - (glink + afterBcl); 89 write32(buf + 0, 0x3d6b0000 | ha(afterBcl)); // addis r11,r11,1f-glink@ha 90 write32(buf + 4, 0x7c0802a6); // mflr r0 91 write32(buf + 8, 0x429f0005); // bcl 20,30,.+4 92 write32(buf + 12, 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-.glink@l 93 write32(buf + 16, 0x7d8802a6); // mflr r12 94 write32(buf + 20, 0x7c0803a6); // mtlr r0 95 write32(buf + 24, 0x7d6c5850); // sub r11,r11,r12 96 write32(buf + 28, 0x3d8c0000 | ha(gotBcl)); // addis 12,12,GOT+4-1b@ha 97 if (ha(gotBcl) == ha(gotBcl + 4)) { 98 write32(buf + 32, 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12) 99 write32(buf + 36, 100 0x818c0000 | lo(gotBcl + 4)); // lwz r12,r12,GOT+8-1b@l(r12) 101 } else { 102 write32(buf + 32, 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12) 103 write32(buf + 36, 0x818c0000 | 4); // lwz r12,r12,4(r12) 104 } 105 write32(buf + 40, 0x7c0903a6); // mtctr 0 106 write32(buf + 44, 0x7c0b5a14); // add r0,11,11 107 write32(buf + 48, 0x7d605a14); // add r11,0,11 108 write32(buf + 52, 0x4e800420); // bctr 109 buf += 56; 110 } else { 111 write32(buf + 0, 0x3d800000 | ha(got + 4)); // lis r12,GOT+4@ha 112 write32(buf + 4, 0x3d6b0000 | ha(-glink)); // addis r11,r11,-Glink@ha 113 if (ha(got + 4) == ha(got + 8)) 114 write32(buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12) 115 else 116 write32(buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12) 117 write32(buf + 12, 0x396b0000 | lo(-glink)); // addi r11,r11,-Glink@l 118 write32(buf + 16, 0x7c0903a6); // mtctr r0 119 write32(buf + 20, 0x7c0b5a14); // add r0,r11,r11 120 if (ha(got + 4) == ha(got + 8)) 121 write32(buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@ha(r12) 122 else 123 write32(buf + 24, 0x818c0000 | 4); // lwz r12,4(r12) 124 write32(buf + 28, 0x7d605a14); // add r11,r0,r11 125 write32(buf + 32, 0x4e800420); // bctr 126 buf += 36; 127 } 128 129 // Pad with nop. They should not be executed. 130 for (; buf < end; buf += 4) 131 write32(buf, 0x60000000); 132 } 133 134 PPC::PPC() { 135 gotRel = R_PPC_GLOB_DAT; 136 noneRel = R_PPC_NONE; 137 pltRel = R_PPC_JMP_SLOT; 138 relativeRel = R_PPC_RELATIVE; 139 iRelativeRel = R_PPC_IRELATIVE; 140 symbolicRel = R_PPC_ADDR32; 141 gotBaseSymInGotPlt = false; 142 gotHeaderEntriesNum = 3; 143 gotPltHeaderEntriesNum = 0; 144 pltHeaderSize = 64; // size of PLTresolve in .glink 145 pltEntrySize = 4; 146 147 needsThunks = true; 148 149 tlsModuleIndexRel = R_PPC_DTPMOD32; 150 tlsOffsetRel = R_PPC_DTPREL32; 151 tlsGotRel = R_PPC_TPREL32; 152 153 defaultMaxPageSize = 65536; 154 defaultImageBase = 0x10000000; 155 156 write32(trapInstr.data(), 0x7fe00008); 157 } 158 159 void PPC::writeGotHeader(uint8_t *buf) const { 160 // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC 161 // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1], 162 // link_map in _GLOBAL_OFFSET_TABLE_[2]. 163 write32(buf, mainPart->dynamic->getVA()); 164 } 165 166 void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const { 167 // Address of the symbol resolver stub in .glink . 168 write32(buf, in.plt->getVA() + 4 * s.pltIndex); 169 } 170 171 bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file, 172 uint64_t branchAddr, const Symbol &s) const { 173 if (type != R_PPC_REL24 && type != R_PPC_PLTREL24) 174 return false; 175 if (s.isInPlt()) 176 return true; 177 if (s.isUndefWeak()) 178 return false; 179 return !(expr == R_PC && PPC::inBranchRange(type, branchAddr, s.getVA())); 180 } 181 182 uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; } 183 184 bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const { 185 uint64_t offset = dst - src; 186 if (type == R_PPC_REL24 || type == R_PPC_PLTREL24) 187 return isInt<26>(offset); 188 llvm_unreachable("unsupported relocation type used in branch"); 189 } 190 191 RelExpr PPC::getRelExpr(RelType type, const Symbol &s, 192 const uint8_t *loc) const { 193 switch (type) { 194 case R_PPC_NONE: 195 return R_NONE; 196 case R_PPC_ADDR16_HA: 197 case R_PPC_ADDR16_HI: 198 case R_PPC_ADDR16_LO: 199 case R_PPC_ADDR32: 200 return R_ABS; 201 case R_PPC_DTPREL16: 202 case R_PPC_DTPREL16_HA: 203 case R_PPC_DTPREL16_HI: 204 case R_PPC_DTPREL16_LO: 205 case R_PPC_DTPREL32: 206 return R_DTPREL; 207 case R_PPC_REL14: 208 case R_PPC_REL32: 209 case R_PPC_LOCAL24PC: 210 case R_PPC_REL16_LO: 211 case R_PPC_REL16_HI: 212 case R_PPC_REL16_HA: 213 return R_PC; 214 case R_PPC_GOT16: 215 return R_GOT_OFF; 216 case R_PPC_REL24: 217 return R_PLT_PC; 218 case R_PPC_PLTREL24: 219 return R_PPC32_PLTREL; 220 case R_PPC_GOT_TLSGD16: 221 return R_TLSGD_GOT; 222 case R_PPC_GOT_TLSLD16: 223 return R_TLSLD_GOT; 224 case R_PPC_GOT_TPREL16: 225 return R_GOT_OFF; 226 case R_PPC_TLS: 227 return R_TLSIE_HINT; 228 case R_PPC_TLSGD: 229 return R_TLSDESC_CALL; 230 case R_PPC_TLSLD: 231 return R_TLSLD_HINT; 232 case R_PPC_TPREL16: 233 case R_PPC_TPREL16_HA: 234 case R_PPC_TPREL16_LO: 235 case R_PPC_TPREL16_HI: 236 return R_TLS; 237 default: 238 error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) + 239 ") against symbol " + toString(s)); 240 return R_NONE; 241 } 242 } 243 244 RelType PPC::getDynRel(RelType type) const { 245 if (type == R_PPC_ADDR32) 246 return type; 247 return R_PPC_NONE; 248 } 249 250 static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) { 251 uint64_t dtpBiasedVal = val - 0x8000; 252 switch (type) { 253 case R_PPC_DTPREL16: 254 return {R_PPC64_ADDR16, dtpBiasedVal}; 255 case R_PPC_DTPREL16_HA: 256 return {R_PPC_ADDR16_HA, dtpBiasedVal}; 257 case R_PPC_DTPREL16_HI: 258 return {R_PPC_ADDR16_HI, dtpBiasedVal}; 259 case R_PPC_DTPREL16_LO: 260 return {R_PPC_ADDR16_LO, dtpBiasedVal}; 261 case R_PPC_DTPREL32: 262 return {R_PPC_ADDR32, dtpBiasedVal}; 263 default: 264 return {type, val}; 265 } 266 } 267 268 void PPC::relocateOne(uint8_t *loc, RelType type, uint64_t val) const { 269 RelType newType; 270 std::tie(newType, val) = fromDTPREL(type, val); 271 switch (newType) { 272 case R_PPC_ADDR16: 273 checkIntUInt(loc, val, 16, type); 274 write16(loc, val); 275 break; 276 case R_PPC_GOT16: 277 case R_PPC_GOT_TLSGD16: 278 case R_PPC_GOT_TLSLD16: 279 case R_PPC_GOT_TPREL16: 280 case R_PPC_TPREL16: 281 checkInt(loc, val, 16, type); 282 write16(loc, val); 283 break; 284 case R_PPC_ADDR16_HA: 285 case R_PPC_DTPREL16_HA: 286 case R_PPC_GOT_TLSGD16_HA: 287 case R_PPC_GOT_TLSLD16_HA: 288 case R_PPC_GOT_TPREL16_HA: 289 case R_PPC_REL16_HA: 290 case R_PPC_TPREL16_HA: 291 write16(loc, ha(val)); 292 break; 293 case R_PPC_ADDR16_HI: 294 case R_PPC_DTPREL16_HI: 295 case R_PPC_GOT_TLSGD16_HI: 296 case R_PPC_GOT_TLSLD16_HI: 297 case R_PPC_GOT_TPREL16_HI: 298 case R_PPC_REL16_HI: 299 case R_PPC_TPREL16_HI: 300 write16(loc, val >> 16); 301 break; 302 case R_PPC_ADDR16_LO: 303 case R_PPC_DTPREL16_LO: 304 case R_PPC_GOT_TLSGD16_LO: 305 case R_PPC_GOT_TLSLD16_LO: 306 case R_PPC_GOT_TPREL16_LO: 307 case R_PPC_REL16_LO: 308 case R_PPC_TPREL16_LO: 309 write16(loc, val); 310 break; 311 case R_PPC_ADDR32: 312 case R_PPC_REL32: 313 write32(loc, val); 314 break; 315 case R_PPC_REL14: { 316 uint32_t mask = 0x0000FFFC; 317 checkInt(loc, val, 16, type); 318 checkAlignment(loc, val, 4, type); 319 write32(loc, (read32(loc) & ~mask) | (val & mask)); 320 break; 321 } 322 case R_PPC_REL24: 323 case R_PPC_LOCAL24PC: 324 case R_PPC_PLTREL24: { 325 uint32_t mask = 0x03FFFFFC; 326 checkInt(loc, val, 26, type); 327 checkAlignment(loc, val, 4, type); 328 write32(loc, (read32(loc) & ~mask) | (val & mask)); 329 break; 330 } 331 default: 332 llvm_unreachable("unknown relocation"); 333 } 334 } 335 336 RelExpr PPC::adjustRelaxExpr(RelType type, const uint8_t *data, 337 RelExpr expr) const { 338 if (expr == R_RELAX_TLS_GD_TO_IE) 339 return R_RELAX_TLS_GD_TO_IE_GOT_OFF; 340 if (expr == R_RELAX_TLS_LD_TO_LE) 341 return R_RELAX_TLS_LD_TO_LE_ABS; 342 return expr; 343 } 344 345 int PPC::getTlsGdRelaxSkip(RelType type) const { 346 // A __tls_get_addr call instruction is marked with 2 relocations: 347 // 348 // R_PPC_TLSGD / R_PPC_TLSLD: marker relocation 349 // R_PPC_REL24: __tls_get_addr 350 // 351 // After the relaxation we no longer call __tls_get_addr and should skip both 352 // relocations to not create a false dependence on __tls_get_addr being 353 // defined. 354 if (type == R_PPC_TLSGD || type == R_PPC_TLSLD) 355 return 2; 356 return 1; 357 } 358 359 void PPC::relaxTlsGdToIe(uint8_t *loc, RelType type, uint64_t val) const { 360 switch (type) { 361 case R_PPC_GOT_TLSGD16: { 362 // addi rT, rA, x@got@tlsgd --> lwz rT, x@got@tprel(rA) 363 uint32_t insn = readFromHalf16(loc); 364 writeFromHalf16(loc, 0x80000000 | (insn & 0x03ff0000)); 365 relocateOne(loc, R_PPC_GOT_TPREL16, val); 366 break; 367 } 368 case R_PPC_TLSGD: 369 // bl __tls_get_addr(x@tldgd) --> add r3, r3, r2 370 write32(loc, 0x7c631214); 371 break; 372 default: 373 llvm_unreachable("unsupported relocation for TLS GD to IE relaxation"); 374 } 375 } 376 377 void PPC::relaxTlsGdToLe(uint8_t *loc, RelType type, uint64_t val) const { 378 switch (type) { 379 case R_PPC_GOT_TLSGD16: 380 // addi r3, r31, x@got@tlsgd --> addis r3, r2, x@tprel@ha 381 writeFromHalf16(loc, 0x3c620000 | ha(val)); 382 break; 383 case R_PPC_TLSGD: 384 // bl __tls_get_addr(x@tldgd) --> add r3, r3, x@tprel@l 385 write32(loc, 0x38630000 | lo(val)); 386 break; 387 default: 388 llvm_unreachable("unsupported relocation for TLS GD to LE relaxation"); 389 } 390 } 391 392 void PPC::relaxTlsLdToLe(uint8_t *loc, RelType type, uint64_t val) const { 393 switch (type) { 394 case R_PPC_GOT_TLSLD16: 395 // addi r3, rA, x@got@tlsgd --> addis r3, r2, 0 396 writeFromHalf16(loc, 0x3c620000); 397 break; 398 case R_PPC_TLSLD: 399 // r3+x@dtprel computes r3+x-0x8000, while we want it to compute r3+x@tprel 400 // = r3+x-0x7000, so add 4096 to r3. 401 // bl __tls_get_addr(x@tlsld) --> addi r3, r3, 4096 402 write32(loc, 0x38631000); 403 break; 404 case R_PPC_DTPREL16: 405 case R_PPC_DTPREL16_HA: 406 case R_PPC_DTPREL16_HI: 407 case R_PPC_DTPREL16_LO: 408 relocateOne(loc, type, val); 409 break; 410 default: 411 llvm_unreachable("unsupported relocation for TLS LD to LE relaxation"); 412 } 413 } 414 415 void PPC::relaxTlsIeToLe(uint8_t *loc, RelType type, uint64_t val) const { 416 switch (type) { 417 case R_PPC_GOT_TPREL16: { 418 // lwz rT, x@got@tprel(rA) --> addis rT, r2, x@tprel@ha 419 uint32_t rt = readFromHalf16(loc) & 0x03e00000; 420 writeFromHalf16(loc, 0x3c020000 | rt | ha(val)); 421 break; 422 } 423 case R_PPC_TLS: { 424 uint32_t insn = read32(loc); 425 if (insn >> 26 != 31) 426 error("unrecognized instruction for IE to LE R_PPC_TLS"); 427 // addi rT, rT, x@tls --> addi rT, rT, x@tprel@l 428 uint32_t dFormOp = getPPCDFormOp((read32(loc) & 0x000007fe) >> 1); 429 if (dFormOp == 0) 430 error("unrecognized instruction for IE to LE R_PPC_TLS"); 431 write32(loc, (dFormOp << 26) | (insn & 0x03ff0000) | lo(val)); 432 break; 433 } 434 default: 435 llvm_unreachable("unsupported relocation for TLS IE to LE relaxation"); 436 } 437 } 438 439 TargetInfo *getPPCTargetInfo() { 440 static PPC target; 441 return ⌖ 442 } 443 444 } // namespace elf 445 } // namespace lld 446