526 lines · cpp
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-exception6//7//===----------------------------------------------------------------------===//8 9#include "OutputSections.h"10#include "Symbols.h"11#include "SyntheticSections.h"12#include "Target.h"13#include "Thunks.h"14 15using namespace llvm;16using namespace llvm::support::endian;17using namespace llvm::ELF;18using namespace lld;19using namespace lld::elf;20 21// Undefine the macro predefined by GCC powerpc32.22#undef PPC23 24namespace {25class PPC final : public TargetInfo {26public:27 PPC(Ctx &);28 RelExpr getRelExpr(RelType type, const Symbol &s,29 const uint8_t *loc) const override;30 RelType getDynRel(RelType type) const override;31 int64_t getImplicitAddend(const uint8_t *buf, RelType type) const override;32 void writeGotHeader(uint8_t *buf) const override;33 void writePltHeader(uint8_t *buf) const override {34 llvm_unreachable("should call writePPC32GlinkSection() instead");35 }36 void writePlt(uint8_t *buf, const Symbol &sym,37 uint64_t pltEntryAddr) const override {38 llvm_unreachable("should call writePPC32GlinkSection() instead");39 }40 void writeIplt(uint8_t *buf, const Symbol &sym,41 uint64_t pltEntryAddr) const override;42 void writeGotPlt(uint8_t *buf, const Symbol &s) const override;43 bool needsThunk(RelExpr expr, RelType relocType, const InputFile *file,44 uint64_t branchAddr, const Symbol &s,45 int64_t a) const override;46 uint32_t getThunkSectionSpacing() const override;47 bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override;48 void relocate(uint8_t *loc, const Relocation &rel,49 uint64_t val) const override;50 RelExpr adjustTlsExpr(RelType type, RelExpr expr) const override;51 int getTlsGdRelaxSkip(RelType type) const override;52 void relocateAlloc(InputSection &sec, uint8_t *buf) const override;53 54private:55 void relaxTlsGdToIe(uint8_t *loc, const Relocation &rel, uint64_t val) const;56 void relaxTlsGdToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;57 void relaxTlsLdToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;58 void relaxTlsIeToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;59};60} // namespace61 62static uint16_t lo(uint32_t v) { return v; }63static uint16_t ha(uint32_t v) { return (v + 0x8000) >> 16; }64 65static uint32_t readFromHalf16(Ctx &ctx, const uint8_t *loc) {66 return read32(ctx, ctx.arg.isLE ? loc : loc - 2);67}68 69static void writeFromHalf16(Ctx &ctx, uint8_t *loc, uint32_t insn) {70 write32(ctx, ctx.arg.isLE ? loc : loc - 2, insn);71}72 73void elf::writePPC32GlinkSection(Ctx &ctx, uint8_t *buf, size_t numEntries) {74 // Create canonical PLT entries for non-PIE code. Compilers don't generate75 // non-GOT-non-PLT relocations referencing external functions for -fpie/-fPIE.76 uint32_t glink = ctx.in.plt->getVA(); // VA of .glink77 if (!ctx.arg.isPic) {78 for (const Symbol *sym :79 cast<PPC32GlinkSection>(*ctx.in.plt).canonical_plts) {80 writePPC32PltCallStub(ctx, buf, sym->getGotPltVA(ctx), nullptr, 0);81 buf += 16;82 glink += 16;83 }84 }85 86 // On PPC Secure PLT ABI, bl foo@plt jumps to a call stub, which loads an87 // absolute address from a specific .plt slot (usually called .got.plt on88 // other targets) and jumps there.89 //90 // a) With immediate binding (BIND_NOW), the .plt entry is resolved at load91 // time. The .glink section is not used.92 // b) With lazy binding, the .plt entry points to a `b PLTresolve`93 // instruction in .glink, filled in by PPC::writeGotPlt().94 95 // Write N `b PLTresolve` first.96 for (size_t i = 0; i != numEntries; ++i)97 write32(ctx, buf + 4 * i, 0x48000000 | 4 * (numEntries - i));98 buf += 4 * numEntries;99 100 // Then write PLTresolve(), which has two forms: PIC and non-PIC. PLTresolve()101 // computes the PLT index (by computing the distance from the landing b to102 // itself) and calls _dl_runtime_resolve() (in glibc).103 uint32_t got = ctx.in.got->getVA();104 const uint8_t *end = buf + 64;105 if (ctx.arg.isPic) {106 uint32_t afterBcl = 4 * ctx.in.plt->getNumEntries() + 12;107 uint32_t gotBcl = got + 4 - (glink + afterBcl);108 write32(ctx, buf + 0,109 0x3d6b0000 | ha(afterBcl)); // addis r11,r11,1f-glink@ha110 write32(ctx, buf + 4, 0x7c0802a6); // mflr r0111 write32(ctx, buf + 8, 0x429f0005); // bcl 20,30,.+4112 write32(ctx, buf + 12,113 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-glink@l114 write32(ctx, buf + 16, 0x7d8802a6); // mflr r12115 write32(ctx, buf + 20, 0x7c0803a6); // mtlr r0116 write32(ctx, buf + 24, 0x7d6c5850); // sub r11,r11,r12117 write32(ctx, buf + 28, 0x3d8c0000 | ha(gotBcl)); // addis 12,12,GOT+4-1b@ha118 if (ha(gotBcl) == ha(gotBcl + 4)) {119 write32(ctx, buf + 32,120 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12)121 write32(ctx, buf + 36,122 0x818c0000 | lo(gotBcl + 4)); // lwz r12,r12,GOT+8-1b@l(r12)123 } else {124 write32(ctx, buf + 32,125 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12)126 write32(ctx, buf + 36, 0x818c0000 | 4); // lwz r12,r12,4(r12)127 }128 write32(ctx, buf + 40, 0x7c0903a6); // mtctr 0129 write32(ctx, buf + 44, 0x7c0b5a14); // add r0,11,11130 write32(ctx, buf + 48, 0x7d605a14); // add r11,0,11131 write32(ctx, buf + 52, 0x4e800420); // bctr132 buf += 56;133 } else {134 write32(ctx, buf + 0, 0x3d800000 | ha(got + 4)); // lis r12,GOT+4@ha135 write32(ctx, buf + 4, 0x3d6b0000 | ha(-glink)); // addis r11,r11,-glink@ha136 if (ha(got + 4) == ha(got + 8))137 write32(ctx, buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12)138 else139 write32(ctx, buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12)140 write32(ctx, buf + 12, 0x396b0000 | lo(-glink)); // addi r11,r11,-glink@l141 write32(ctx, buf + 16, 0x7c0903a6); // mtctr r0142 write32(ctx, buf + 20, 0x7c0b5a14); // add r0,r11,r11143 if (ha(got + 4) == ha(got + 8))144 write32(ctx, buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@l(r12)145 else146 write32(ctx, buf + 24, 0x818c0000 | 4); // lwz r12,4(r12)147 write32(ctx, buf + 28, 0x7d605a14); // add r11,r0,r11148 write32(ctx, buf + 32, 0x4e800420); // bctr149 buf += 36;150 }151 152 // Pad with nop. They should not be executed.153 for (; buf < end; buf += 4)154 write32(ctx, buf, 0x60000000);155}156 157PPC::PPC(Ctx &ctx) : TargetInfo(ctx) {158 copyRel = R_PPC_COPY;159 gotRel = R_PPC_GLOB_DAT;160 pltRel = R_PPC_JMP_SLOT;161 relativeRel = R_PPC_RELATIVE;162 iRelativeRel = R_PPC_IRELATIVE;163 symbolicRel = R_PPC_ADDR32;164 gotHeaderEntriesNum = 3;165 gotPltHeaderEntriesNum = 0;166 pltHeaderSize = 0;167 pltEntrySize = 4;168 ipltEntrySize = 16;169 170 needsThunks = true;171 172 tlsModuleIndexRel = R_PPC_DTPMOD32;173 tlsOffsetRel = R_PPC_DTPREL32;174 tlsGotRel = R_PPC_TPREL32;175 176 defaultMaxPageSize = 65536;177 defaultImageBase = 0x10000000;178 179 write32(ctx, trapInstr.data(), 0x7fe00008);180}181 182void PPC::writeIplt(uint8_t *buf, const Symbol &sym,183 uint64_t /*pltEntryAddr*/) const {184 // In -pie or -shared mode, assume r30 points to .got2+0x8000, and use a185 // .got2.plt_pic32. thunk.186 writePPC32PltCallStub(ctx, buf, sym.getGotPltVA(ctx), sym.file, 0x8000);187}188 189void PPC::writeGotHeader(uint8_t *buf) const {190 // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC191 // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1],192 // link_map in _GLOBAL_OFFSET_TABLE_[2].193 write32(ctx, buf, ctx.mainPart->dynamic->getVA());194}195 196void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const {197 // Address of the symbol resolver stub in .glink .198 write32(ctx, buf,199 ctx.in.plt->getVA() + ctx.in.plt->headerSize + 4 * s.getPltIdx(ctx));200}201 202bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file,203 uint64_t branchAddr, const Symbol &s, int64_t a) const {204 if (type != R_PPC_LOCAL24PC && type != R_PPC_REL24 && type != R_PPC_PLTREL24)205 return false;206 if (s.isInPlt(ctx))207 return true;208 if (s.isUndefWeak())209 return false;210 return !PPC::inBranchRange(type, branchAddr, s.getVA(ctx, a));211}212 213uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; }214 215bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const {216 uint64_t offset = dst - src;217 if (type == R_PPC_LOCAL24PC || type == R_PPC_REL24 || type == R_PPC_PLTREL24)218 return isInt<26>(offset);219 llvm_unreachable("unsupported relocation type used in branch");220}221 222RelExpr PPC::getRelExpr(RelType type, const Symbol &s,223 const uint8_t *loc) const {224 switch (type) {225 case R_PPC_NONE:226 return R_NONE;227 case R_PPC_ADDR16_HA:228 case R_PPC_ADDR16_HI:229 case R_PPC_ADDR16_LO:230 case R_PPC_ADDR24:231 case R_PPC_ADDR32:232 return R_ABS;233 case R_PPC_DTPREL16:234 case R_PPC_DTPREL16_HA:235 case R_PPC_DTPREL16_HI:236 case R_PPC_DTPREL16_LO:237 case R_PPC_DTPREL32:238 return R_DTPREL;239 case R_PPC_REL14:240 case R_PPC_REL32:241 case R_PPC_REL16_LO:242 case R_PPC_REL16_HI:243 case R_PPC_REL16_HA:244 return R_PC;245 case R_PPC_GOT16:246 return R_GOT_OFF;247 case R_PPC_LOCAL24PC:248 case R_PPC_REL24:249 return R_PLT_PC;250 case R_PPC_PLTREL24:251 return RE_PPC32_PLTREL;252 case R_PPC_GOT_TLSGD16:253 return R_TLSGD_GOT;254 case R_PPC_GOT_TLSLD16:255 return R_TLSLD_GOT;256 case R_PPC_GOT_TPREL16:257 return R_GOT_OFF;258 case R_PPC_TLS:259 return R_TLSIE_HINT;260 case R_PPC_TLSGD:261 return R_TLSDESC_CALL;262 case R_PPC_TLSLD:263 return R_TLSLD_HINT;264 case R_PPC_TPREL16:265 case R_PPC_TPREL16_HA:266 case R_PPC_TPREL16_LO:267 case R_PPC_TPREL16_HI:268 return R_TPREL;269 default:270 Err(ctx) << getErrorLoc(ctx, loc) << "unknown relocation (" << type.v271 << ") against symbol " << &s;272 return R_NONE;273 }274}275 276RelType PPC::getDynRel(RelType type) const {277 if (type == R_PPC_ADDR32)278 return type;279 return R_PPC_NONE;280}281 282int64_t PPC::getImplicitAddend(const uint8_t *buf, RelType type) const {283 switch (type) {284 case R_PPC_NONE:285 case R_PPC_GLOB_DAT:286 case R_PPC_JMP_SLOT:287 return 0;288 case R_PPC_ADDR32:289 case R_PPC_REL32:290 case R_PPC_RELATIVE:291 case R_PPC_IRELATIVE:292 case R_PPC_DTPMOD32:293 case R_PPC_DTPREL32:294 case R_PPC_TPREL32:295 return SignExtend64<32>(read32(ctx, buf));296 default:297 InternalErr(ctx, buf) << "cannot read addend for relocation " << type;298 return 0;299 }300}301 302static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) {303 uint64_t dtpBiasedVal = val - 0x8000;304 switch (type) {305 case R_PPC_DTPREL16:306 return {R_PPC64_ADDR16, dtpBiasedVal};307 case R_PPC_DTPREL16_HA:308 return {R_PPC_ADDR16_HA, dtpBiasedVal};309 case R_PPC_DTPREL16_HI:310 return {R_PPC_ADDR16_HI, dtpBiasedVal};311 case R_PPC_DTPREL16_LO:312 return {R_PPC_ADDR16_LO, dtpBiasedVal};313 case R_PPC_DTPREL32:314 return {R_PPC_ADDR32, dtpBiasedVal};315 default:316 return {type, val};317 }318}319 320void PPC::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {321 RelType newType;322 std::tie(newType, val) = fromDTPREL(rel.type, val);323 switch (newType) {324 case R_PPC_ADDR16:325 checkIntUInt(ctx, loc, val, 16, rel);326 write16(ctx, loc, val);327 break;328 case R_PPC_GOT16:329 case R_PPC_GOT_TLSGD16:330 case R_PPC_GOT_TLSLD16:331 case R_PPC_GOT_TPREL16:332 case R_PPC_TPREL16:333 checkInt(ctx, loc, val, 16, rel);334 write16(ctx, loc, val);335 break;336 case R_PPC_ADDR16_HA:337 case R_PPC_DTPREL16_HA:338 case R_PPC_GOT_TLSGD16_HA:339 case R_PPC_GOT_TLSLD16_HA:340 case R_PPC_GOT_TPREL16_HA:341 case R_PPC_REL16_HA:342 case R_PPC_TPREL16_HA:343 write16(ctx, loc, ha(val));344 break;345 case R_PPC_ADDR16_HI:346 case R_PPC_DTPREL16_HI:347 case R_PPC_GOT_TLSGD16_HI:348 case R_PPC_GOT_TLSLD16_HI:349 case R_PPC_GOT_TPREL16_HI:350 case R_PPC_REL16_HI:351 case R_PPC_TPREL16_HI:352 write16(ctx, loc, val >> 16);353 break;354 case R_PPC_ADDR16_LO:355 case R_PPC_DTPREL16_LO:356 case R_PPC_GOT_TLSGD16_LO:357 case R_PPC_GOT_TLSLD16_LO:358 case R_PPC_GOT_TPREL16_LO:359 case R_PPC_REL16_LO:360 case R_PPC_TPREL16_LO:361 write16(ctx, loc, val);362 break;363 case R_PPC_ADDR32:364 case R_PPC_REL32:365 write32(ctx, loc, val);366 break;367 case R_PPC_REL14: {368 uint32_t mask = 0x0000FFFC;369 checkInt(ctx, loc, val, 16, rel);370 checkAlignment(ctx, loc, val, 4, rel);371 write32(ctx, loc, (read32(ctx, loc) & ~mask) | (val & mask));372 break;373 }374 case R_PPC_ADDR24:375 case R_PPC_REL24:376 case R_PPC_LOCAL24PC:377 case R_PPC_PLTREL24: {378 uint32_t mask = 0x03FFFFFC;379 checkInt(ctx, loc, val, 26, rel);380 checkAlignment(ctx, loc, val, 4, rel);381 write32(ctx, loc, (read32(ctx, loc) & ~mask) | (val & mask));382 break;383 }384 default:385 llvm_unreachable("unknown relocation");386 }387}388 389RelExpr PPC::adjustTlsExpr(RelType type, RelExpr expr) const {390 if (expr == R_RELAX_TLS_GD_TO_IE)391 return R_RELAX_TLS_GD_TO_IE_GOT_OFF;392 if (expr == R_RELAX_TLS_LD_TO_LE)393 return R_RELAX_TLS_LD_TO_LE_ABS;394 return expr;395}396 397int PPC::getTlsGdRelaxSkip(RelType type) const {398 // A __tls_get_addr call instruction is marked with 2 relocations:399 //400 // R_PPC_TLSGD / R_PPC_TLSLD: marker relocation401 // R_PPC_REL24: __tls_get_addr402 //403 // After the relaxation we no longer call __tls_get_addr and should skip both404 // relocations to not create a false dependence on __tls_get_addr being405 // defined.406 if (type == R_PPC_TLSGD || type == R_PPC_TLSLD)407 return 2;408 return 1;409}410 411void PPC::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,412 uint64_t val) const {413 switch (rel.type) {414 case R_PPC_GOT_TLSGD16: {415 // addi rT, rA, x@got@tlsgd --> lwz rT, x@got@tprel(rA)416 uint32_t insn = readFromHalf16(ctx, loc);417 writeFromHalf16(ctx, loc, 0x80000000 | (insn & 0x03ff0000));418 relocateNoSym(loc, R_PPC_GOT_TPREL16, val);419 break;420 }421 case R_PPC_TLSGD:422 // bl __tls_get_addr(x@tldgd) --> add r3, r3, r2423 write32(ctx, loc, 0x7c631214);424 break;425 default:426 llvm_unreachable("unsupported relocation for TLS GD to IE relaxation");427 }428}429 430void PPC::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,431 uint64_t val) const {432 switch (rel.type) {433 case R_PPC_GOT_TLSGD16:434 // addi r3, r31, x@got@tlsgd --> addis r3, r2, x@tprel@ha435 writeFromHalf16(ctx, loc, 0x3c620000 | ha(val));436 break;437 case R_PPC_TLSGD:438 // bl __tls_get_addr(x@tldgd) --> add r3, r3, x@tprel@l439 write32(ctx, loc, 0x38630000 | lo(val));440 break;441 default:442 llvm_unreachable("unsupported relocation for TLS GD to LE relaxation");443 }444}445 446void PPC::relaxTlsLdToLe(uint8_t *loc, const Relocation &rel,447 uint64_t val) const {448 switch (rel.type) {449 case R_PPC_GOT_TLSLD16:450 // addi r3, rA, x@got@tlsgd --> addis r3, r2, 0451 writeFromHalf16(ctx, loc, 0x3c620000);452 break;453 case R_PPC_TLSLD:454 // r3+x@dtprel computes r3+x-0x8000, while we want it to compute r3+x@tprel455 // = r3+x-0x7000, so add 4096 to r3.456 // bl __tls_get_addr(x@tlsld) --> addi r3, r3, 4096457 write32(ctx, loc, 0x38631000);458 break;459 case R_PPC_DTPREL16:460 case R_PPC_DTPREL16_HA:461 case R_PPC_DTPREL16_HI:462 case R_PPC_DTPREL16_LO:463 relocate(loc, rel, val);464 break;465 default:466 llvm_unreachable("unsupported relocation for TLS LD to LE relaxation");467 }468}469 470void PPC::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,471 uint64_t val) const {472 switch (rel.type) {473 case R_PPC_GOT_TPREL16: {474 // lwz rT, x@got@tprel(rA) --> addis rT, r2, x@tprel@ha475 uint32_t rt = readFromHalf16(ctx, loc) & 0x03e00000;476 writeFromHalf16(ctx, loc, 0x3c020000 | rt | ha(val));477 break;478 }479 case R_PPC_TLS: {480 uint32_t insn = read32(ctx, loc);481 if (insn >> 26 != 31)482 ErrAlways(ctx) << "unrecognized instruction for IE to LE R_PPC_TLS";483 // addi rT, rT, x@tls --> addi rT, rT, x@tprel@l484 unsigned secondaryOp = (read32(ctx, loc) & 0x000007fe) >> 1;485 uint32_t dFormOp = getPPCDFormOp(secondaryOp);486 if (dFormOp == 0) { // Expecting a DS-Form instruction.487 dFormOp = getPPCDSFormOp(secondaryOp);488 if (dFormOp == 0)489 ErrAlways(ctx) << "unrecognized instruction for IE to LE R_PPC_TLS";490 }491 write32(ctx, loc, (dFormOp | (insn & 0x03ff0000) | lo(val)));492 break;493 }494 default:495 llvm_unreachable("unsupported relocation for TLS IE to LE relaxation");496 }497}498 499void PPC::relocateAlloc(InputSection &sec, uint8_t *buf) const {500 uint64_t secAddr = sec.getOutputSection()->addr + sec.outSecOff;501 for (const Relocation &rel : sec.relocs()) {502 uint8_t *loc = buf + rel.offset;503 const uint64_t val =504 SignExtend64(sec.getRelocTargetVA(ctx, rel, secAddr + rel.offset), 32);505 switch (rel.expr) {506 case R_RELAX_TLS_GD_TO_IE_GOT_OFF:507 relaxTlsGdToIe(loc, rel, val);508 break;509 case R_RELAX_TLS_GD_TO_LE:510 relaxTlsGdToLe(loc, rel, val);511 break;512 case R_RELAX_TLS_LD_TO_LE_ABS:513 relaxTlsLdToLe(loc, rel, val);514 break;515 case R_RELAX_TLS_IE_TO_LE:516 relaxTlsIeToLe(loc, rel, val);517 break;518 default:519 relocate(loc, rel, val);520 break;521 }522 }523}524 525void elf::setPPCTargetInfo(Ctx &ctx) { ctx.target.reset(new PPC(ctx)); }526