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 using namespace lld;
21 using namespace lld::elf;
22
23 // Undefine the macro predefined by GCC powerpc32.
24 #undef PPC
25
26 namespace {
27 class PPC final : public TargetInfo {
28 public:
29 PPC();
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;
writePltHeader(uint8_t * buf) const35 void writePltHeader(uint8_t *buf) const override {
36 llvm_unreachable("should call writePPC32GlinkSection() instead");
37 }
writePlt(uint8_t * buf,const Symbol & sym,uint64_t pltEntryAddr) const38 void writePlt(uint8_t *buf, const Symbol &sym,
39 uint64_t pltEntryAddr) const override {
40 llvm_unreachable("should call writePPC32GlinkSection() instead");
41 }
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 relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
55 uint64_t val) const override;
56 void relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
57 uint64_t val) const override;
58 void relaxTlsLdToLe(uint8_t *loc, const Relocation &rel,
59 uint64_t val) const override;
60 void relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,
61 uint64_t val) const override;
62 };
63 } // namespace
64
lo(uint32_t v)65 static uint16_t lo(uint32_t v) { return v; }
ha(uint32_t v)66 static uint16_t ha(uint32_t v) { return (v + 0x8000) >> 16; }
67
readFromHalf16(const uint8_t * loc)68 static uint32_t readFromHalf16(const uint8_t *loc) {
69 return read32(config->isLE ? loc : loc - 2);
70 }
71
writeFromHalf16(uint8_t * loc,uint32_t insn)72 static void writeFromHalf16(uint8_t *loc, uint32_t insn) {
73 write32(config->isLE ? loc : loc - 2, insn);
74 }
75
writePPC32GlinkSection(uint8_t * buf,size_t numEntries)76 void elf::writePPC32GlinkSection(uint8_t *buf, size_t numEntries) {
77 // Create canonical PLT entries for non-PIE code. Compilers don't generate
78 // non-GOT-non-PLT relocations referencing external functions for -fpie/-fPIE.
79 uint32_t glink = in.plt->getVA(); // VA of .glink
80 if (!config->isPic) {
81 for (const Symbol *sym : cast<PPC32GlinkSection>(*in.plt).canonical_plts) {
82 writePPC32PltCallStub(buf, sym->getGotPltVA(), nullptr, 0);
83 buf += 16;
84 glink += 16;
85 }
86 }
87
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.
91 //
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().
96
97 // Write N `b PLTresolve` first.
98 for (size_t i = 0; i != numEntries; ++i)
99 write32(buf + 4 * i, 0x48000000 | 4 * (numEntries - i));
100 buf += 4 * numEntries;
101
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 = in.got->getVA();
106 const uint8_t *end = buf + 64;
107 if (config->isPic) {
108 uint32_t afterBcl = 4 * in.plt->getNumEntries() + 12;
109 uint32_t gotBcl = got + 4 - (glink + afterBcl);
110 write32(buf + 0, 0x3d6b0000 | ha(afterBcl)); // addis r11,r11,1f-glink@ha
111 write32(buf + 4, 0x7c0802a6); // mflr r0
112 write32(buf + 8, 0x429f0005); // bcl 20,30,.+4
113 write32(buf + 12, 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-glink@l
114 write32(buf + 16, 0x7d8802a6); // mflr r12
115 write32(buf + 20, 0x7c0803a6); // mtlr r0
116 write32(buf + 24, 0x7d6c5850); // sub r11,r11,r12
117 write32(buf + 28, 0x3d8c0000 | ha(gotBcl)); // addis 12,12,GOT+4-1b@ha
118 if (ha(gotBcl) == ha(gotBcl + 4)) {
119 write32(buf + 32, 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12)
120 write32(buf + 36,
121 0x818c0000 | lo(gotBcl + 4)); // lwz r12,r12,GOT+8-1b@l(r12)
122 } else {
123 write32(buf + 32, 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12)
124 write32(buf + 36, 0x818c0000 | 4); // lwz r12,r12,4(r12)
125 }
126 write32(buf + 40, 0x7c0903a6); // mtctr 0
127 write32(buf + 44, 0x7c0b5a14); // add r0,11,11
128 write32(buf + 48, 0x7d605a14); // add r11,0,11
129 write32(buf + 52, 0x4e800420); // bctr
130 buf += 56;
131 } else {
132 write32(buf + 0, 0x3d800000 | ha(got + 4)); // lis r12,GOT+4@ha
133 write32(buf + 4, 0x3d6b0000 | ha(-glink)); // addis r11,r11,-glink@ha
134 if (ha(got + 4) == ha(got + 8))
135 write32(buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12)
136 else
137 write32(buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12)
138 write32(buf + 12, 0x396b0000 | lo(-glink)); // addi r11,r11,-glink@l
139 write32(buf + 16, 0x7c0903a6); // mtctr r0
140 write32(buf + 20, 0x7c0b5a14); // add r0,r11,r11
141 if (ha(got + 4) == ha(got + 8))
142 write32(buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@l(r12)
143 else
144 write32(buf + 24, 0x818c0000 | 4); // lwz r12,4(r12)
145 write32(buf + 28, 0x7d605a14); // add r11,r0,r11
146 write32(buf + 32, 0x4e800420); // bctr
147 buf += 36;
148 }
149
150 // Pad with nop. They should not be executed.
151 for (; buf < end; buf += 4)
152 write32(buf, 0x60000000);
153 }
154
PPC()155 PPC::PPC() {
156 copyRel = R_PPC_COPY;
157 gotRel = R_PPC_GLOB_DAT;
158 pltRel = R_PPC_JMP_SLOT;
159 relativeRel = R_PPC_RELATIVE;
160 iRelativeRel = R_PPC_IRELATIVE;
161 symbolicRel = R_PPC_ADDR32;
162 gotHeaderEntriesNum = 3;
163 gotPltHeaderEntriesNum = 0;
164 pltHeaderSize = 0;
165 pltEntrySize = 4;
166 ipltEntrySize = 16;
167
168 needsThunks = true;
169
170 tlsModuleIndexRel = R_PPC_DTPMOD32;
171 tlsOffsetRel = R_PPC_DTPREL32;
172 tlsGotRel = R_PPC_TPREL32;
173
174 defaultMaxPageSize = 65536;
175 defaultImageBase = 0x10000000;
176
177 write32(trapInstr.data(), 0x7fe00008);
178 }
179
writeIplt(uint8_t * buf,const Symbol & sym,uint64_t) const180 void PPC::writeIplt(uint8_t *buf, const Symbol &sym,
181 uint64_t /*pltEntryAddr*/) const {
182 // In -pie or -shared mode, assume r30 points to .got2+0x8000, and use a
183 // .got2.plt_pic32. thunk.
184 writePPC32PltCallStub(buf, sym.getGotPltVA(), sym.file, 0x8000);
185 }
186
writeGotHeader(uint8_t * buf) const187 void PPC::writeGotHeader(uint8_t *buf) const {
188 // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC
189 // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1],
190 // link_map in _GLOBAL_OFFSET_TABLE_[2].
191 write32(buf, mainPart->dynamic->getVA());
192 }
193
writeGotPlt(uint8_t * buf,const Symbol & s) const194 void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const {
195 // Address of the symbol resolver stub in .glink .
196 write32(buf, in.plt->getVA() + in.plt->headerSize + 4 * s.getPltIdx());
197 }
198
needsThunk(RelExpr expr,RelType type,const InputFile * file,uint64_t branchAddr,const Symbol & s,int64_t a) const199 bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file,
200 uint64_t branchAddr, const Symbol &s, int64_t a) const {
201 if (type != R_PPC_LOCAL24PC && type != R_PPC_REL24 && type != R_PPC_PLTREL24)
202 return false;
203 if (s.isInPlt())
204 return true;
205 if (s.isUndefWeak())
206 return false;
207 return !PPC::inBranchRange(type, branchAddr, s.getVA(a));
208 }
209
getThunkSectionSpacing() const210 uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; }
211
inBranchRange(RelType type,uint64_t src,uint64_t dst) const212 bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const {
213 uint64_t offset = dst - src;
214 if (type == R_PPC_LOCAL24PC || type == R_PPC_REL24 || type == R_PPC_PLTREL24)
215 return isInt<26>(offset);
216 llvm_unreachable("unsupported relocation type used in branch");
217 }
218
getRelExpr(RelType type,const Symbol & s,const uint8_t * loc) const219 RelExpr PPC::getRelExpr(RelType type, const Symbol &s,
220 const uint8_t *loc) const {
221 switch (type) {
222 case R_PPC_NONE:
223 return R_NONE;
224 case R_PPC_ADDR16_HA:
225 case R_PPC_ADDR16_HI:
226 case R_PPC_ADDR16_LO:
227 case R_PPC_ADDR24:
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_TPREL;
266 default:
267 error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) +
268 ") against symbol " + toString(s));
269 return R_NONE;
270 }
271 }
272
getDynRel(RelType type) const273 RelType PPC::getDynRel(RelType type) const {
274 if (type == R_PPC_ADDR32)
275 return type;
276 return R_PPC_NONE;
277 }
278
getImplicitAddend(const uint8_t * buf,RelType type) const279 int64_t PPC::getImplicitAddend(const uint8_t *buf, RelType type) const {
280 switch (type) {
281 case R_PPC_NONE:
282 return 0;
283 case R_PPC_ADDR32:
284 case R_PPC_REL32:
285 return SignExtend64<32>(read32(buf));
286 default:
287 internalLinkerError(getErrorLocation(buf),
288 "cannot read addend for relocation " + toString(type));
289 return 0;
290 }
291 }
292
fromDTPREL(RelType type,uint64_t val)293 static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) {
294 uint64_t dtpBiasedVal = val - 0x8000;
295 switch (type) {
296 case R_PPC_DTPREL16:
297 return {R_PPC64_ADDR16, dtpBiasedVal};
298 case R_PPC_DTPREL16_HA:
299 return {R_PPC_ADDR16_HA, dtpBiasedVal};
300 case R_PPC_DTPREL16_HI:
301 return {R_PPC_ADDR16_HI, dtpBiasedVal};
302 case R_PPC_DTPREL16_LO:
303 return {R_PPC_ADDR16_LO, dtpBiasedVal};
304 case R_PPC_DTPREL32:
305 return {R_PPC_ADDR32, dtpBiasedVal};
306 default:
307 return {type, val};
308 }
309 }
310
relocate(uint8_t * loc,const Relocation & rel,uint64_t val) const311 void PPC::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {
312 RelType newType;
313 std::tie(newType, val) = fromDTPREL(rel.type, val);
314 switch (newType) {
315 case R_PPC_ADDR16:
316 checkIntUInt(loc, val, 16, rel);
317 write16(loc, val);
318 break;
319 case R_PPC_GOT16:
320 case R_PPC_GOT_TLSGD16:
321 case R_PPC_GOT_TLSLD16:
322 case R_PPC_GOT_TPREL16:
323 case R_PPC_TPREL16:
324 checkInt(loc, val, 16, rel);
325 write16(loc, val);
326 break;
327 case R_PPC_ADDR16_HA:
328 case R_PPC_DTPREL16_HA:
329 case R_PPC_GOT_TLSGD16_HA:
330 case R_PPC_GOT_TLSLD16_HA:
331 case R_PPC_GOT_TPREL16_HA:
332 case R_PPC_REL16_HA:
333 case R_PPC_TPREL16_HA:
334 write16(loc, ha(val));
335 break;
336 case R_PPC_ADDR16_HI:
337 case R_PPC_DTPREL16_HI:
338 case R_PPC_GOT_TLSGD16_HI:
339 case R_PPC_GOT_TLSLD16_HI:
340 case R_PPC_GOT_TPREL16_HI:
341 case R_PPC_REL16_HI:
342 case R_PPC_TPREL16_HI:
343 write16(loc, val >> 16);
344 break;
345 case R_PPC_ADDR16_LO:
346 case R_PPC_DTPREL16_LO:
347 case R_PPC_GOT_TLSGD16_LO:
348 case R_PPC_GOT_TLSLD16_LO:
349 case R_PPC_GOT_TPREL16_LO:
350 case R_PPC_REL16_LO:
351 case R_PPC_TPREL16_LO:
352 write16(loc, val);
353 break;
354 case R_PPC_ADDR32:
355 case R_PPC_REL32:
356 write32(loc, val);
357 break;
358 case R_PPC_REL14: {
359 uint32_t mask = 0x0000FFFC;
360 checkInt(loc, val, 16, rel);
361 checkAlignment(loc, val, 4, rel);
362 write32(loc, (read32(loc) & ~mask) | (val & mask));
363 break;
364 }
365 case R_PPC_ADDR24:
366 case R_PPC_REL24:
367 case R_PPC_LOCAL24PC:
368 case R_PPC_PLTREL24: {
369 uint32_t mask = 0x03FFFFFC;
370 checkInt(loc, val, 26, rel);
371 checkAlignment(loc, val, 4, rel);
372 write32(loc, (read32(loc) & ~mask) | (val & mask));
373 break;
374 }
375 default:
376 llvm_unreachable("unknown relocation");
377 }
378 }
379
adjustTlsExpr(RelType type,RelExpr expr) const380 RelExpr PPC::adjustTlsExpr(RelType type, RelExpr expr) const {
381 if (expr == R_RELAX_TLS_GD_TO_IE)
382 return R_RELAX_TLS_GD_TO_IE_GOT_OFF;
383 if (expr == R_RELAX_TLS_LD_TO_LE)
384 return R_RELAX_TLS_LD_TO_LE_ABS;
385 return expr;
386 }
387
getTlsGdRelaxSkip(RelType type) const388 int PPC::getTlsGdRelaxSkip(RelType type) const {
389 // A __tls_get_addr call instruction is marked with 2 relocations:
390 //
391 // R_PPC_TLSGD / R_PPC_TLSLD: marker relocation
392 // R_PPC_REL24: __tls_get_addr
393 //
394 // After the relaxation we no longer call __tls_get_addr and should skip both
395 // relocations to not create a false dependence on __tls_get_addr being
396 // defined.
397 if (type == R_PPC_TLSGD || type == R_PPC_TLSLD)
398 return 2;
399 return 1;
400 }
401
relaxTlsGdToIe(uint8_t * loc,const Relocation & rel,uint64_t val) const402 void PPC::relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
403 uint64_t val) const {
404 switch (rel.type) {
405 case R_PPC_GOT_TLSGD16: {
406 // addi rT, rA, x@got@tlsgd --> lwz rT, x@got@tprel(rA)
407 uint32_t insn = readFromHalf16(loc);
408 writeFromHalf16(loc, 0x80000000 | (insn & 0x03ff0000));
409 relocateNoSym(loc, R_PPC_GOT_TPREL16, val);
410 break;
411 }
412 case R_PPC_TLSGD:
413 // bl __tls_get_addr(x@tldgd) --> add r3, r3, r2
414 write32(loc, 0x7c631214);
415 break;
416 default:
417 llvm_unreachable("unsupported relocation for TLS GD to IE relaxation");
418 }
419 }
420
relaxTlsGdToLe(uint8_t * loc,const Relocation & rel,uint64_t val) const421 void PPC::relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
422 uint64_t val) const {
423 switch (rel.type) {
424 case R_PPC_GOT_TLSGD16:
425 // addi r3, r31, x@got@tlsgd --> addis r3, r2, x@tprel@ha
426 writeFromHalf16(loc, 0x3c620000 | ha(val));
427 break;
428 case R_PPC_TLSGD:
429 // bl __tls_get_addr(x@tldgd) --> add r3, r3, x@tprel@l
430 write32(loc, 0x38630000 | lo(val));
431 break;
432 default:
433 llvm_unreachable("unsupported relocation for TLS GD to LE relaxation");
434 }
435 }
436
relaxTlsLdToLe(uint8_t * loc,const Relocation & rel,uint64_t val) const437 void PPC::relaxTlsLdToLe(uint8_t *loc, const Relocation &rel,
438 uint64_t val) const {
439 switch (rel.type) {
440 case R_PPC_GOT_TLSLD16:
441 // addi r3, rA, x@got@tlsgd --> addis r3, r2, 0
442 writeFromHalf16(loc, 0x3c620000);
443 break;
444 case R_PPC_TLSLD:
445 // r3+x@dtprel computes r3+x-0x8000, while we want it to compute r3+x@tprel
446 // = r3+x-0x7000, so add 4096 to r3.
447 // bl __tls_get_addr(x@tlsld) --> addi r3, r3, 4096
448 write32(loc, 0x38631000);
449 break;
450 case R_PPC_DTPREL16:
451 case R_PPC_DTPREL16_HA:
452 case R_PPC_DTPREL16_HI:
453 case R_PPC_DTPREL16_LO:
454 relocate(loc, rel, val);
455 break;
456 default:
457 llvm_unreachable("unsupported relocation for TLS LD to LE relaxation");
458 }
459 }
460
relaxTlsIeToLe(uint8_t * loc,const Relocation & rel,uint64_t val) const461 void PPC::relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,
462 uint64_t val) const {
463 switch (rel.type) {
464 case R_PPC_GOT_TPREL16: {
465 // lwz rT, x@got@tprel(rA) --> addis rT, r2, x@tprel@ha
466 uint32_t rt = readFromHalf16(loc) & 0x03e00000;
467 writeFromHalf16(loc, 0x3c020000 | rt | ha(val));
468 break;
469 }
470 case R_PPC_TLS: {
471 uint32_t insn = read32(loc);
472 if (insn >> 26 != 31)
473 error("unrecognized instruction for IE to LE R_PPC_TLS");
474 // addi rT, rT, x@tls --> addi rT, rT, x@tprel@l
475 uint32_t dFormOp = getPPCDFormOp((read32(loc) & 0x000007fe) >> 1);
476 if (dFormOp == 0)
477 error("unrecognized instruction for IE to LE R_PPC_TLS");
478 write32(loc, (dFormOp << 26) | (insn & 0x03ff0000) | lo(val));
479 break;
480 }
481 default:
482 llvm_unreachable("unsupported relocation for TLS IE to LE relaxation");
483 }
484 }
485
getPPCTargetInfo()486 TargetInfo *elf::getPPCTargetInfo() {
487 static PPC target;
488 return ⌖
489 }
490