PPC.cpp revision 1.3 
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
17using namespace llvm;
18using namespace llvm::support::endian;
19using namespace llvm::ELF;
20using namespace lld;
21using namespace lld::elf;
22
23namespace {
24class PPC final : public TargetInfo {
25public:
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, const Symbol &sym,
35                uint64_t pltEntryAddr) const override {
36    llvm_unreachable("should call writePPC32GlinkSection() instead");
37  }
38  void writeIplt(uint8_t *buf, const Symbol &sym,
39                 uint64_t pltEntryAddr) const override;
40  void writeGotPlt(uint8_t *buf, const Symbol &s) const override;
41  bool needsThunk(RelExpr expr, RelType relocType, const InputFile *file,
42                  uint64_t branchAddr, const Symbol &s,
43                  int64_t a) const override;
44  uint32_t getThunkSectionSpacing() const override;
45  bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override;
46  void relocate(uint8_t *loc, const Relocation &rel,
47                uint64_t val) const override;
48  RelExpr adjustRelaxExpr(RelType type, const uint8_t *data,
49                          RelExpr expr) const override;
50  int getTlsGdRelaxSkip(RelType type) const override;
51  void relaxTlsGdToIe(uint8_t *loc, const Relocation &rel,
52                      uint64_t val) const override;
53  void relaxTlsGdToLe(uint8_t *loc, const Relocation &rel,
54                      uint64_t val) const override;
55  void relaxTlsLdToLe(uint8_t *loc, const Relocation &rel,
56                      uint64_t val) const override;
57  void relaxTlsIeToLe(uint8_t *loc, const Relocation &rel,
58                      uint64_t val) const override;
59};
60} // namespace
61
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(const uint8_t *loc) {
66  return read32(config->isLE ? loc : loc - 2);
67}
68
69static void writeFromHalf16(uint8_t *loc, uint32_t insn) {
70  write32(config->isLE ? loc : loc - 2, insn);
71}
72
73void elf::writePPC32GlinkSection(uint8_t *buf, size_t numEntries) {
74  // Create canonical PLT entries for non-PIE code. Compilers don't generate
75  // non-GOT-non-PLT relocations referencing external functions for -fpie/-fPIE.
76  uint32_t glink = in.plt->getVA(); // VA of .glink
77  if (!config->isPic) {
78    for (const Symbol *sym : cast<PPC32GlinkSection>(in.plt)->canonical_plts) {
79      writePPC32PltCallStub(buf, sym->getGotPltVA(), nullptr, 0);
80      buf += 16;
81      glink += 16;
82    }
83  }
84
85  // On PPC Secure PLT ABI, bl foo@plt jumps to a call stub, which loads an
86  // absolute address from a specific .plt slot (usually called .got.plt on
87  // other targets) and jumps there.
88  //
89  // a) With immediate binding (BIND_NOW), the .plt entry is resolved at load
90  // time. The .glink section is not used.
91  // b) With lazy binding, the .plt entry points to a `b PLTresolve`
92  // instruction in .glink, filled in by PPC::writeGotPlt().
93
94  // Write N `b PLTresolve` first.
95  for (size_t i = 0; i != numEntries; ++i)
96    write32(buf + 4 * i, 0x48000000 | 4 * (numEntries - i));
97  buf += 4 * numEntries;
98
99  // Then write PLTresolve(), which has two forms: PIC and non-PIC. PLTresolve()
100  // computes the PLT index (by computing the distance from the landing b to
101  // itself) and calls _dl_runtime_resolve() (in glibc).
102  uint32_t got = in.got->getVA();
103  const uint8_t *end = buf + 64;
104  if (config->isPic) {
105    uint32_t afterBcl = 4 * in.plt->getNumEntries() + 12;
106    uint32_t gotBcl = got + 4 - (glink + afterBcl);
107    write32(buf + 0, 0x3d6b0000 | ha(afterBcl));  // addis r11,r11,1f-glink@ha
108    write32(buf + 4, 0x7c0802a6);                 // mflr r0
109    write32(buf + 8, 0x429f0005);                 // bcl 20,30,.+4
110    write32(buf + 12, 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-glink@l
111    write32(buf + 16, 0x7d8802a6);                // mflr r12
112    write32(buf + 20, 0x7c0803a6);                // mtlr r0
113    write32(buf + 24, 0x7d6c5850);                // sub r11,r11,r12
114    write32(buf + 28, 0x3d8c0000 | ha(gotBcl));   // addis 12,12,GOT+4-1b@ha
115    if (ha(gotBcl) == ha(gotBcl + 4)) {
116      write32(buf + 32, 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12)
117      write32(buf + 36,
118              0x818c0000 | lo(gotBcl + 4));       // lwz r12,r12,GOT+8-1b@l(r12)
119    } else {
120      write32(buf + 32, 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12)
121      write32(buf + 36, 0x818c0000 | 4);          // lwz r12,r12,4(r12)
122    }
123    write32(buf + 40, 0x7c0903a6);                // mtctr 0
124    write32(buf + 44, 0x7c0b5a14);                // add r0,11,11
125    write32(buf + 48, 0x7d605a14);                // add r11,0,11
126    write32(buf + 52, 0x4e800420);                // bctr
127    buf += 56;
128  } else {
129    write32(buf + 0, 0x3d800000 | ha(got + 4));   // lis     r12,GOT+4@ha
130    write32(buf + 4, 0x3d6b0000 | ha(-glink));    // addis   r11,r11,-glink@ha
131    if (ha(got + 4) == ha(got + 8))
132      write32(buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12)
133    else
134      write32(buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12)
135    write32(buf + 12, 0x396b0000 | lo(-glink));   // addi    r11,r11,-glink@l
136    write32(buf + 16, 0x7c0903a6);                // mtctr   r0
137    write32(buf + 20, 0x7c0b5a14);                // add     r0,r11,r11
138    if (ha(got + 4) == ha(got + 8))
139      write32(buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@l(r12)
140    else
141      write32(buf + 24, 0x818c0000 | 4);          // lwz r12,4(r12)
142    write32(buf + 28, 0x7d605a14);                // add     r11,r0,r11
143    write32(buf + 32, 0x4e800420);                // bctr
144    buf += 36;
145  }
146
147  // Pad with nop. They should not be executed.
148  for (; buf < end; buf += 4)
149    write32(buf, 0x60000000);
150}
151
152PPC::PPC() {
153  copyRel = R_PPC_COPY;
154  gotRel = R_PPC_GLOB_DAT;
155  noneRel = R_PPC_NONE;
156  pltRel = R_PPC_JMP_SLOT;
157  relativeRel = R_PPC_RELATIVE;
158  iRelativeRel = R_PPC_IRELATIVE;
159  symbolicRel = R_PPC_ADDR32;
160  gotBaseSymInGotPlt = false;
161  gotHeaderEntriesNum = 3;
162  gotPltHeaderEntriesNum = 0;
163  pltHeaderSize = 0;
164  pltEntrySize = 4;
165  ipltEntrySize = 16;
166
167  needsThunks = true;
168
169  tlsModuleIndexRel = R_PPC_DTPMOD32;
170  tlsOffsetRel = R_PPC_DTPREL32;
171  tlsGotRel = R_PPC_TPREL32;
172
173  defaultMaxPageSize = 65536;
174  defaultImageBase = 0x10000000;
175
176  write32(trapInstr.data(), 0x7fe00008);
177}
178
179void PPC::writeIplt(uint8_t *buf, const Symbol &sym,
180                    uint64_t /*pltEntryAddr*/) const {
181  // In -pie or -shared mode, assume r30 points to .got2+0x8000, and use a
182  // .got2.plt_pic32. thunk.
183  writePPC32PltCallStub(buf, sym.getGotPltVA(), sym.file, 0x8000);
184}
185
186void PPC::writeGotHeader(uint8_t *buf) const {
187  // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC
188  // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1],
189  // link_map in _GLOBAL_OFFSET_TABLE_[2].
190  write32(buf, mainPart->dynamic->getVA());
191}
192
193void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const {
194  // Address of the symbol resolver stub in .glink .
195  write32(buf, in.plt->getVA() + in.plt->headerSize + 4 * s.pltIndex);
196}
197
198bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file,
199                     uint64_t branchAddr, const Symbol &s, int64_t a) const {
200  if (type != R_PPC_LOCAL24PC && type != R_PPC_REL24 && type != R_PPC_PLTREL24)
201    return false;
202  if (s.isInPlt())
203    return true;
204  if (s.isUndefWeak())
205    return false;
206  return !PPC::inBranchRange(type, branchAddr, s.getVA(a));
207}
208
209uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; }
210
211bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const {
212  uint64_t offset = dst - src;
213  if (type == R_PPC_LOCAL24PC || type == R_PPC_REL24 || type == R_PPC_PLTREL24)
214    return isInt<26>(offset);
215  llvm_unreachable("unsupported relocation type used in branch");
216}
217
218RelExpr PPC::getRelExpr(RelType type, const Symbol &s,
219                        const uint8_t *loc) const {
220  switch (type) {
221  case R_PPC_NONE:
222    return R_NONE;
223  case R_PPC_ADDR16_HA:
224  case R_PPC_ADDR16_HI:
225  case R_PPC_ADDR16_LO:
226  case R_PPC_ADDR24:
227  case R_PPC_ADDR32:
228    return R_ABS;
229  case R_PPC_DTPREL16:
230  case R_PPC_DTPREL16_HA:
231  case R_PPC_DTPREL16_HI:
232  case R_PPC_DTPREL16_LO:
233  case R_PPC_DTPREL32:
234    return R_DTPREL;
235  case R_PPC_REL14:
236  case R_PPC_REL32:
237  case R_PPC_REL16_LO:
238  case R_PPC_REL16_HI:
239  case R_PPC_REL16_HA:
240    return R_PC;
241  case R_PPC_GOT16:
242    return R_GOT_OFF;
243  case R_PPC_LOCAL24PC:
244  case R_PPC_REL24:
245    return R_PLT_PC;
246  case R_PPC_PLTREL24:
247    return R_PPC32_PLTREL;
248  case R_PPC_GOT_TLSGD16:
249    return R_TLSGD_GOT;
250  case R_PPC_GOT_TLSLD16:
251    return R_TLSLD_GOT;
252  case R_PPC_GOT_TPREL16:
253    return R_GOT_OFF;
254  case R_PPC_TLS:
255    return R_TLSIE_HINT;
256  case R_PPC_TLSGD:
257    return R_TLSDESC_CALL;
258  case R_PPC_TLSLD:
259    return R_TLSLD_HINT;
260  case R_PPC_TPREL16:
261  case R_PPC_TPREL16_HA:
262  case R_PPC_TPREL16_LO:
263  case R_PPC_TPREL16_HI:
264    return R_TLS;
265  default:
266    error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) +
267          ") against symbol " + toString(s));
268    return R_NONE;
269  }
270}
271
272RelType PPC::getDynRel(RelType type) const {
273  if (type == R_PPC_ADDR32)
274    return type;
275  return R_PPC_NONE;
276}
277
278static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) {
279  uint64_t dtpBiasedVal = val - 0x8000;
280  switch (type) {
281  case R_PPC_DTPREL16:
282    return {R_PPC64_ADDR16, dtpBiasedVal};
283  case R_PPC_DTPREL16_HA:
284    return {R_PPC_ADDR16_HA, dtpBiasedVal};
285  case R_PPC_DTPREL16_HI:
286    return {R_PPC_ADDR16_HI, dtpBiasedVal};
287  case R_PPC_DTPREL16_LO:
288    return {R_PPC_ADDR16_LO, dtpBiasedVal};
289  case R_PPC_DTPREL32:
290    return {R_PPC_ADDR32, dtpBiasedVal};
291  default:
292    return {type, val};
293  }
294}
295
296void PPC::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {
297  RelType newType;
298  std::tie(newType, val) = fromDTPREL(rel.type, val);
299  switch (newType) {
300  case R_PPC_ADDR16:
301    checkIntUInt(loc, val, 16, rel);
302    write16(loc, val);
303    break;
304  case R_PPC_GOT16:
305  case R_PPC_GOT_TLSGD16:
306  case R_PPC_GOT_TLSLD16:
307  case R_PPC_GOT_TPREL16:
308  case R_PPC_TPREL16:
309    checkInt(loc, val, 16, rel);
310    write16(loc, val);
311    break;
312  case R_PPC_ADDR16_HA:
313  case R_PPC_DTPREL16_HA:
314  case R_PPC_GOT_TLSGD16_HA:
315  case R_PPC_GOT_TLSLD16_HA:
316  case R_PPC_GOT_TPREL16_HA:
317  case R_PPC_REL16_HA:
318  case R_PPC_TPREL16_HA:
319    write16(loc, ha(val));
320    break;
321  case R_PPC_ADDR16_HI:
322  case R_PPC_DTPREL16_HI:
323  case R_PPC_GOT_TLSGD16_HI:
324  case R_PPC_GOT_TLSLD16_HI:
325  case R_PPC_GOT_TPREL16_HI:
326  case R_PPC_REL16_HI:
327  case R_PPC_TPREL16_HI:
328    write16(loc, val >> 16);
329    break;
330  case R_PPC_ADDR16_LO:
331  case R_PPC_DTPREL16_LO:
332  case R_PPC_GOT_TLSGD16_LO:
333  case R_PPC_GOT_TLSLD16_LO:
334  case R_PPC_GOT_TPREL16_LO:
335  case R_PPC_REL16_LO:
336  case R_PPC_TPREL16_LO:
337    write16(loc, val);
338    break;
339  case R_PPC_ADDR32:
340  case R_PPC_REL32:
341    write32(loc, val);
342    break;
343  case R_PPC_REL14: {
344    uint32_t mask = 0x0000FFFC;
345    checkInt(loc, val, 16, rel);
346    checkAlignment(loc, val, 4, rel);
347    write32(loc, (read32(loc) & ~mask) | (val & mask));
348    break;
349  }
350  case R_PPC_REL24:
351  case R_PPC_ADDR24:
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
365RelExpr 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
374int 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
388void 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
407void 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
423void 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
447void 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
472TargetInfo *elf::getPPCTargetInfo() {
473  static PPC target;
474  return &target;
475}
476