[mlir][py] Enable loading only specified dialects during creation. (#121421)
[llvm-project.git] / lld / ELF / Arch / PPC.cpp
blob3203e27d82fa2ec1c5d40a07d9aa6dba9fa05eae
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 //===----------------------------------------------------------------------===//
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"
17 using namespace llvm;
18 using namespace llvm::support::endian;
19 using namespace llvm::ELF;
20 using namespace lld;
21 using namespace lld::elf;
23 // Undefine the macro predefined by GCC powerpc32.
24 #undef PPC
26 namespace {
27 class PPC final : public TargetInfo {
28 public:
29 PPC(Ctx &);
30 RelExpr getRelExpr(RelType type, const Symbol &s,
31 const uint8_t *loc) const override;
32 RelType getDynRel(RelType type) const override;
33 int64_t getImplicitAddend(const uint8_t *buf, RelType type) const override;
34 void writeGotHeader(uint8_t *buf) const override;
35 void writePltHeader(uint8_t *buf) const override {
36 llvm_unreachable("should call writePPC32GlinkSection() instead");
38 void writePlt(uint8_t *buf, const Symbol &sym,
39 uint64_t pltEntryAddr) const override {
40 llvm_unreachable("should call writePPC32GlinkSection() instead");
42 void writeIplt(uint8_t *buf, const Symbol &sym,
43 uint64_t pltEntryAddr) const override;
44 void writeGotPlt(uint8_t *buf, const Symbol &s) const override;
45 bool needsThunk(RelExpr expr, RelType relocType, const InputFile *file,
46 uint64_t branchAddr, const Symbol &s,
47 int64_t a) const override;
48 uint32_t getThunkSectionSpacing() const override;
49 bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override;
50 void relocate(uint8_t *loc, const Relocation &rel,
51 uint64_t val) const override;
52 RelExpr adjustTlsExpr(RelType type, RelExpr expr) const override;
53 int getTlsGdRelaxSkip(RelType type) const override;
54 void relocateAlloc(InputSectionBase &sec, uint8_t *buf) const override;
56 private:
57 void relaxTlsGdToIe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
58 void relaxTlsGdToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
59 void relaxTlsLdToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
60 void relaxTlsIeToLe(uint8_t *loc, const Relocation &rel, uint64_t val) const;
62 } // namespace
64 static uint16_t lo(uint32_t v) { return v; }
65 static uint16_t ha(uint32_t v) { return (v + 0x8000) >> 16; }
67 static uint32_t readFromHalf16(Ctx &ctx, const uint8_t *loc) {
68 return read32(ctx, ctx.arg.isLE ? loc : loc - 2);
71 static void writeFromHalf16(Ctx &ctx, uint8_t *loc, uint32_t insn) {
72 write32(ctx, ctx.arg.isLE ? loc : loc - 2, insn);
75 void elf::writePPC32GlinkSection(Ctx &ctx, uint8_t *buf, size_t numEntries) {
76 // Create canonical PLT entries for non-PIE code. Compilers don't generate
77 // non-GOT-non-PLT relocations referencing external functions for -fpie/-fPIE.
78 uint32_t glink = ctx.in.plt->getVA(); // VA of .glink
79 if (!ctx.arg.isPic) {
80 for (const Symbol *sym :
81 cast<PPC32GlinkSection>(*ctx.in.plt).canonical_plts) {
82 writePPC32PltCallStub(ctx, buf, sym->getGotPltVA(ctx), nullptr, 0);
83 buf += 16;
84 glink += 16;
88 // On PPC Secure PLT ABI, bl foo@plt jumps to a call stub, which loads an
89 // absolute address from a specific .plt slot (usually called .got.plt on
90 // other targets) and jumps there.
92 // a) With immediate binding (BIND_NOW), the .plt entry is resolved at load
93 // time. The .glink section is not used.
94 // b) With lazy binding, the .plt entry points to a `b PLTresolve`
95 // instruction in .glink, filled in by PPC::writeGotPlt().
97 // Write N `b PLTresolve` first.
98 for (size_t i = 0; i != numEntries; ++i)
99 write32(ctx, buf + 4 * i, 0x48000000 | 4 * (numEntries - i));
100 buf += 4 * numEntries;
102 // Then write PLTresolve(), which has two forms: PIC and non-PIC. PLTresolve()
103 // computes the PLT index (by computing the distance from the landing b to
104 // itself) and calls _dl_runtime_resolve() (in glibc).
105 uint32_t got = ctx.in.got->getVA();
106 const uint8_t *end = buf + 64;
107 if (ctx.arg.isPic) {
108 uint32_t afterBcl = 4 * ctx.in.plt->getNumEntries() + 12;
109 uint32_t gotBcl = got + 4 - (glink + afterBcl);
110 write32(ctx, buf + 0,
111 0x3d6b0000 | ha(afterBcl)); // addis r11,r11,1f-glink@ha
112 write32(ctx, buf + 4, 0x7c0802a6); // mflr r0
113 write32(ctx, buf + 8, 0x429f0005); // bcl 20,30,.+4
114 write32(ctx, buf + 12,
115 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-glink@l
116 write32(ctx, buf + 16, 0x7d8802a6); // mflr r12
117 write32(ctx, buf + 20, 0x7c0803a6); // mtlr r0
118 write32(ctx, buf + 24, 0x7d6c5850); // sub r11,r11,r12
119 write32(ctx, buf + 28, 0x3d8c0000 | ha(gotBcl)); // addis 12,12,GOT+4-1b@ha
120 if (ha(gotBcl) == ha(gotBcl + 4)) {
121 write32(ctx, buf + 32,
122 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12)
123 write32(ctx, buf + 36,
124 0x818c0000 | lo(gotBcl + 4)); // lwz r12,r12,GOT+8-1b@l(r12)
125 } else {
126 write32(ctx, buf + 32,
127 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12)
128 write32(ctx, buf + 36, 0x818c0000 | 4); // lwz r12,r12,4(r12)
130 write32(ctx, buf + 40, 0x7c0903a6); // mtctr 0
131 write32(ctx, buf + 44, 0x7c0b5a14); // add r0,11,11
132 write32(ctx, buf + 48, 0x7d605a14); // add r11,0,11
133 write32(ctx, buf + 52, 0x4e800420); // bctr
134 buf += 56;
135 } else {
136 write32(ctx, buf + 0, 0x3d800000 | ha(got + 4)); // lis r12,GOT+4@ha
137 write32(ctx, buf + 4, 0x3d6b0000 | ha(-glink)); // addis r11,r11,-glink@ha
138 if (ha(got + 4) == ha(got + 8))
139 write32(ctx, buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12)
140 else
141 write32(ctx, buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12)
142 write32(ctx, buf + 12, 0x396b0000 | lo(-glink)); // addi r11,r11,-glink@l
143 write32(ctx, buf + 16, 0x7c0903a6); // mtctr r0
144 write32(ctx, buf + 20, 0x7c0b5a14); // add r0,r11,r11
145 if (ha(got + 4) == ha(got + 8))
146 write32(ctx, buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@l(r12)
147 else
148 write32(ctx, buf + 24, 0x818c0000 | 4); // lwz r12,4(r12)
149 write32(ctx, buf + 28, 0x7d605a14); // add r11,r0,r11
150 write32(ctx, buf + 32, 0x4e800420); // bctr
151 buf += 36;
154 // Pad with nop. They should not be executed.
155 for (; buf < end; buf += 4)
156 write32(ctx, buf, 0x60000000);
159 PPC::PPC(Ctx &ctx) : TargetInfo(ctx) {
160 copyRel = R_PPC_COPY;
161 gotRel = R_PPC_GLOB_DAT;
162 pltRel = R_PPC_JMP_SLOT;
163 relativeRel = R_PPC_RELATIVE;
164 iRelativeRel = R_PPC_IRELATIVE;
165 symbolicRel = R_PPC_ADDR32;
166 gotHeaderEntriesNum = 3;
167 gotPltHeaderEntriesNum = 0;
168 pltHeaderSize = 0;
169 pltEntrySize = 4;
170 ipltEntrySize = 16;
172 needsThunks = true;
174 tlsModuleIndexRel = R_PPC_DTPMOD32;
175 tlsOffsetRel = R_PPC_DTPREL32;
176 tlsGotRel = R_PPC_TPREL32;
178 defaultMaxPageSize = 65536;
179 defaultImageBase = 0x10000000;
181 write32(ctx, trapInstr.data(), 0x7fe00008);
184 void PPC::writeIplt(uint8_t *buf, const Symbol &sym,
185 uint64_t /*pltEntryAddr*/) const {
186 // In -pie or -shared mode, assume r30 points to .got2+0x8000, and use a
187 // .got2.plt_pic32. thunk.
188 writePPC32PltCallStub(ctx, buf, sym.getGotPltVA(ctx), sym.file, 0x8000);
191 void PPC::writeGotHeader(uint8_t *buf) const {
192 // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC
193 // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1],
194 // link_map in _GLOBAL_OFFSET_TABLE_[2].
195 write32(ctx, buf, ctx.mainPart->dynamic->getVA());
198 void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const {
199 // Address of the symbol resolver stub in .glink .
200 write32(ctx, buf,
201 ctx.in.plt->getVA() + ctx.in.plt->headerSize + 4 * s.getPltIdx(ctx));
204 bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file,
205 uint64_t branchAddr, const Symbol &s, int64_t a) const {
206 if (type != R_PPC_LOCAL24PC && type != R_PPC_REL24 && type != R_PPC_PLTREL24)
207 return false;
208 if (s.isInPlt(ctx))
209 return true;
210 if (s.isUndefWeak())
211 return false;
212 return !PPC::inBranchRange(type, branchAddr, s.getVA(ctx, a));
215 uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; }
217 bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const {
218 uint64_t offset = dst - src;
219 if (type == R_PPC_LOCAL24PC || type == R_PPC_REL24 || type == R_PPC_PLTREL24)
220 return isInt<26>(offset);
221 llvm_unreachable("unsupported relocation type used in branch");
224 RelExpr PPC::getRelExpr(RelType type, const Symbol &s,
225 const uint8_t *loc) const {
226 switch (type) {
227 case R_PPC_NONE:
228 return R_NONE;
229 case R_PPC_ADDR16_HA:
230 case R_PPC_ADDR16_HI:
231 case R_PPC_ADDR16_LO:
232 case R_PPC_ADDR24:
233 case R_PPC_ADDR32:
234 return R_ABS;
235 case R_PPC_DTPREL16:
236 case R_PPC_DTPREL16_HA:
237 case R_PPC_DTPREL16_HI:
238 case R_PPC_DTPREL16_LO:
239 case R_PPC_DTPREL32:
240 return R_DTPREL;
241 case R_PPC_REL14:
242 case R_PPC_REL32:
243 case R_PPC_REL16_LO:
244 case R_PPC_REL16_HI:
245 case R_PPC_REL16_HA:
246 return R_PC;
247 case R_PPC_GOT16:
248 return R_GOT_OFF;
249 case R_PPC_LOCAL24PC:
250 case R_PPC_REL24:
251 return R_PLT_PC;
252 case R_PPC_PLTREL24:
253 return RE_PPC32_PLTREL;
254 case R_PPC_GOT_TLSGD16:
255 return R_TLSGD_GOT;
256 case R_PPC_GOT_TLSLD16:
257 return R_TLSLD_GOT;
258 case R_PPC_GOT_TPREL16:
259 return R_GOT_OFF;
260 case R_PPC_TLS:
261 return R_TLSIE_HINT;
262 case R_PPC_TLSGD:
263 return R_TLSDESC_CALL;
264 case R_PPC_TLSLD:
265 return R_TLSLD_HINT;
266 case R_PPC_TPREL16:
267 case R_PPC_TPREL16_HA:
268 case R_PPC_TPREL16_LO:
269 case R_PPC_TPREL16_HI:
270 return R_TPREL;
271 default:
272 Err(ctx) << getErrorLoc(ctx, loc) << "unknown relocation (" << type.v
273 << ") against symbol " << &s;
274 return R_NONE;
278 RelType PPC::getDynRel(RelType type) const {
279 if (type == R_PPC_ADDR32)
280 return type;
281 return R_PPC_NONE;
284 int64_t PPC::getImplicitAddend(const uint8_t *buf, RelType type) const {
285 switch (type) {
286 case R_PPC_NONE:
287 case R_PPC_GLOB_DAT:
288 case R_PPC_JMP_SLOT:
289 return 0;
290 case R_PPC_ADDR32:
291 case R_PPC_REL32:
292 case R_PPC_RELATIVE:
293 case R_PPC_IRELATIVE:
294 case R_PPC_DTPMOD32:
295 case R_PPC_DTPREL32:
296 case R_PPC_TPREL32:
297 return SignExtend64<32>(read32(ctx, buf));
298 default:
299 InternalErr(ctx, buf) << "cannot read addend for relocation " << type;
300 return 0;
304 static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) {
305 uint64_t dtpBiasedVal = val - 0x8000;
306 switch (type) {
307 case R_PPC_DTPREL16:
308 return {R_PPC64_ADDR16, dtpBiasedVal};
309 case R_PPC_DTPREL16_HA:
310 return {R_PPC_ADDR16_HA, dtpBiasedVal};
311 case R_PPC_DTPREL16_HI:
312 return {R_PPC_ADDR16_HI, dtpBiasedVal};
313 case R_PPC_DTPREL16_LO:
314 return {R_PPC_ADDR16_LO, dtpBiasedVal};
315 case R_PPC_DTPREL32:
316 return {R_PPC_ADDR32, dtpBiasedVal};
317 default:
318 return {type, val};
322 void PPC::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {
323 RelType newType;
324 std::tie(newType, val) = fromDTPREL(rel.type, val);
325 switch (newType) {
326 case R_PPC_ADDR16:
327 checkIntUInt(ctx, loc, val, 16, rel);
328 write16(ctx, loc, val);
329 break;
330 case R_PPC_GOT16:
331 case R_PPC_GOT_TLSGD16:
332 case R_PPC_GOT_TLSLD16:
333 case R_PPC_GOT_TPREL16:
334 case R_PPC_TPREL16:
335 checkInt(ctx, loc, val, 16, rel);
336 write16(ctx, loc, val);
337 break;
338 case R_PPC_ADDR16_HA:
339 case R_PPC_DTPREL16_HA:
340 case R_PPC_GOT_TLSGD16_HA:
341 case R_PPC_GOT_TLSLD16_HA:
342 case R_PPC_GOT_TPREL16_HA:
343 case R_PPC_REL16_HA:
344 case R_PPC_TPREL16_HA:
345 write16(ctx, loc, ha(val));
346 break;
347 case R_PPC_ADDR16_HI:
348 case R_PPC_DTPREL16_HI:
349 case R_PPC_GOT_TLSGD16_HI:
350 case R_PPC_GOT_TLSLD16_HI:
351 case R_PPC_GOT_TPREL16_HI:
352 case R_PPC_REL16_HI:
353 case R_PPC_TPREL16_HI:
354 write16(ctx, loc, val >> 16);
355 break;
356 case R_PPC_ADDR16_LO:
357 case R_PPC_DTPREL16_LO:
358 case R_PPC_GOT_TLSGD16_LO:
359 case R_PPC_GOT_TLSLD16_LO:
360 case R_PPC_GOT_TPREL16_LO:
361 case R_PPC_REL16_LO:
362 case R_PPC_TPREL16_LO:
363 write16(ctx, loc, val);
364 break;
365 case R_PPC_ADDR32:
366 case R_PPC_REL32:
367 write32(ctx, loc, val);
368 break;
369 case R_PPC_REL14: {
370 uint32_t mask = 0x0000FFFC;
371 checkInt(ctx, loc, val, 16, rel);
372 checkAlignment(ctx, loc, val, 4, rel);
373 write32(ctx, loc, (read32(ctx, loc) & ~mask) | (val & mask));
374 break;
376 case R_PPC_ADDR24:
377 case R_PPC_REL24:
378 case R_PPC_LOCAL24PC:
379 case R_PPC_PLTREL24: {
380 uint32_t mask = 0x03FFFFFC;
381 checkInt(ctx, loc, val, 26, rel);
382 checkAlignment(ctx, loc, val, 4, rel);
383 write32(ctx, loc, (read32(ctx, loc) & ~mask) | (val & mask));
384 break;
386 default:
387 llvm_unreachable("unknown relocation");
391 RelExpr PPC::adjustTlsExpr(RelType type, RelExpr expr) const {
392 if (expr == R_RELAX_TLS_GD_TO_IE)
393 return R_RELAX_TLS_GD_TO_IE_GOT_OFF;
394 if (expr == R_RELAX_TLS_LD_TO_LE)
395 return R_RELAX_TLS_LD_TO_LE_ABS;
396 return expr;
399 int PPC::getTlsGdRelaxSkip(RelType type) const {
400 // A __tls_get_addr call instruction is marked with 2 relocations:
402 // R_PPC_TLSGD / R_PPC_TLSLD: marker relocation
403 // R_PPC_REL24: __tls_get_addr
405 // After the relaxation we no longer call __tls_get_addr and should skip both
406 // relocations to not create a false dependence on __tls_get_addr being
407 // defined.
408 if (type == R_PPC_TLSGD || type == R_PPC_TLSLD)
409 return 2;
410 return 1;
413 void PPC::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
414 uint64_t val) const {
415 switch (rel.type) {
416 case R_PPC_GOT_TLSGD16: {
417 // addi rT, rA, x@got@tlsgd --> lwz rT, x@got@tprel(rA)
418 uint32_t insn = readFromHalf16(ctx, loc);
419 writeFromHalf16(ctx, loc, 0x80000000 | (insn & 0x03ff0000));
420 relocateNoSym(loc, R_PPC_GOT_TPREL16, val);
421 break;
423 case R_PPC_TLSGD:
424 // bl __tls_get_addr(x@tldgd) --> add r3, r3, r2
425 write32(ctx, loc, 0x7c631214);
426 break;
427 default:
428 llvm_unreachable("unsupported relocation for TLS GD to IE relaxation");
432 void PPC::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
433 uint64_t val) const {
434 switch (rel.type) {
435 case R_PPC_GOT_TLSGD16:
436 // addi r3, r31, x@got@tlsgd --> addis r3, r2, x@tprel@ha
437 writeFromHalf16(ctx, loc, 0x3c620000 | ha(val));
438 break;
439 case R_PPC_TLSGD:
440 // bl __tls_get_addr(x@tldgd) --> add r3, r3, x@tprel@l
441 write32(ctx, loc, 0x38630000 | lo(val));
442 break;
443 default:
444 llvm_unreachable("unsupported relocation for TLS GD to LE relaxation");
448 void PPC::relaxTlsLdToLe(uint8_t *loc, const Relocation &rel,
449 uint64_t val) const {
450 switch (rel.type) {
451 case R_PPC_GOT_TLSLD16:
452 // addi r3, rA, x@got@tlsgd --> addis r3, r2, 0
453 writeFromHalf16(ctx, loc, 0x3c620000);
454 break;
455 case R_PPC_TLSLD:
456 // r3+x@dtprel computes r3+x-0x8000, while we want it to compute r3+x@tprel
457 // = r3+x-0x7000, so add 4096 to r3.
458 // bl __tls_get_addr(x@tlsld) --> addi r3, r3, 4096
459 write32(ctx, loc, 0x38631000);
460 break;
461 case R_PPC_DTPREL16:
462 case R_PPC_DTPREL16_HA:
463 case R_PPC_DTPREL16_HI:
464 case R_PPC_DTPREL16_LO:
465 relocate(loc, rel, val);
466 break;
467 default:
468 llvm_unreachable("unsupported relocation for TLS LD to LE relaxation");
472 void PPC::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,
473 uint64_t val) const {
474 switch (rel.type) {
475 case R_PPC_GOT_TPREL16: {
476 // lwz rT, x@got@tprel(rA) --> addis rT, r2, x@tprel@ha
477 uint32_t rt = readFromHalf16(ctx, loc) & 0x03e00000;
478 writeFromHalf16(ctx, loc, 0x3c020000 | rt | ha(val));
479 break;
481 case R_PPC_TLS: {
482 uint32_t insn = read32(ctx, loc);
483 if (insn >> 26 != 31)
484 ErrAlways(ctx) << "unrecognized instruction for IE to LE R_PPC_TLS";
485 // addi rT, rT, x@tls --> addi rT, rT, x@tprel@l
486 unsigned secondaryOp = (read32(ctx, loc) & 0x000007fe) >> 1;
487 uint32_t dFormOp = getPPCDFormOp(secondaryOp);
488 if (dFormOp == 0) { // Expecting a DS-Form instruction.
489 dFormOp = getPPCDSFormOp(secondaryOp);
490 if (dFormOp == 0)
491 ErrAlways(ctx) << "unrecognized instruction for IE to LE R_PPC_TLS";
493 write32(ctx, loc, (dFormOp | (insn & 0x03ff0000) | lo(val)));
494 break;
496 default:
497 llvm_unreachable("unsupported relocation for TLS IE to LE relaxation");
501 void PPC::relocateAlloc(InputSectionBase &sec, uint8_t *buf) const {
502 uint64_t secAddr = sec.getOutputSection()->addr;
503 if (auto *s = dyn_cast<InputSection>(&sec))
504 secAddr += s->outSecOff;
505 for (const Relocation &rel : sec.relocs()) {
506 uint8_t *loc = buf + rel.offset;
507 const uint64_t val =
508 SignExtend64(sec.getRelocTargetVA(ctx, rel, secAddr + rel.offset), 32);
509 switch (rel.expr) {
510 case R_RELAX_TLS_GD_TO_IE_GOT_OFF:
511 relaxTlsGdToIe(loc, rel, val);
512 break;
513 case R_RELAX_TLS_GD_TO_LE:
514 relaxTlsGdToLe(loc, rel, val);
515 break;
516 case R_RELAX_TLS_LD_TO_LE_ABS:
517 relaxTlsLdToLe(loc, rel, val);
518 break;
519 case R_RELAX_TLS_IE_TO_LE:
520 relaxTlsIeToLe(loc, rel, val);
521 break;
522 default:
523 relocate(loc, rel, val);
524 break;
529 void elf::setPPCTargetInfo(Ctx &ctx) { ctx.target.reset(new PPC(ctx)); }