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 
21 namespace lld {
22 namespace elf {
23 
24 namespace {
25 class PPC final : public TargetInfo {
26 public:
27   PPC();
28   RelExpr getRelExpr(RelType type, const Symbol &s,
29                      const uint8_t *loc) const override;
30   RelType getDynRel(RelType type) const override;
31   void writeGotHeader(uint8_t *buf) const override;
32   void writePltHeader(uint8_t *buf) const override {
33     llvm_unreachable("should call writePPC32GlinkSection() instead");
34   }
35   void writePlt(uint8_t *buf, const Symbol &sym,
36                 uint64_t pltEntryAddr) const override {
37     llvm_unreachable("should call writePPC32GlinkSection() instead");
38   }
39   void writeIplt(uint8_t *buf, const Symbol &sym,
40                  uint64_t pltEntryAddr) const override;
41   void writeGotPlt(uint8_t *buf, const Symbol &s) const override;
42   bool needsThunk(RelExpr expr, RelType relocType, const InputFile *file,
43                   uint64_t branchAddr, const Symbol &s,
44                   int64_t a) const override;
45   uint32_t getThunkSectionSpacing() const override;
46   bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override;
47   void relocateOne(uint8_t *loc, RelType type, 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, RelType type, uint64_t val) const override;
52   void relaxTlsGdToLe(uint8_t *loc, RelType type, uint64_t val) const override;
53   void relaxTlsLdToLe(uint8_t *loc, RelType type, uint64_t val) const override;
54   void relaxTlsIeToLe(uint8_t *loc, RelType type, uint64_t val) const override;
55 };
56 } // namespace
57 
58 static uint16_t lo(uint32_t v) { return v; }
59 static uint16_t ha(uint32_t v) { return (v + 0x8000) >> 16; }
60 
61 static uint32_t readFromHalf16(const uint8_t *loc) {
62   return read32(config->isLE ? loc : loc - 2);
63 }
64 
65 static void writeFromHalf16(uint8_t *loc, uint32_t insn) {
66   write32(config->isLE ? loc : loc - 2, insn);
67 }
68 
69 void writePPC32GlinkSection(uint8_t *buf, size_t numEntries) {
70   // Create canonical PLT entries for non-PIE code. Compilers don't generate
71   // non-GOT-non-PLT relocations referencing external functions for -fpie/-fPIE.
72   uint32_t glink = in.plt->getVA(); // VA of .glink
73   if (!config->isPic) {
74     for (const Symbol *sym : cast<PPC32GlinkSection>(in.plt)->canonical_plts) {
75       writePPC32PltCallStub(buf, sym->getGotPltVA(), nullptr, 0);
76       buf += 16;
77       glink += 16;
78     }
79   }
80 
81   // On PPC Secure PLT ABI, bl foo@plt jumps to a call stub, which loads an
82   // absolute address from a specific .plt slot (usually called .got.plt on
83   // other targets) and jumps there.
84   //
85   // a) With immediate binding (BIND_NOW), the .plt entry is resolved at load
86   // time. The .glink section is not used.
87   // b) With lazy binding, the .plt entry points to a `b PLTresolve`
88   // instruction in .glink, filled in by PPC::writeGotPlt().
89 
90   // Write N `b PLTresolve` first.
91   for (size_t i = 0; i != numEntries; ++i)
92     write32(buf + 4 * i, 0x48000000 | 4 * (numEntries - i));
93   buf += 4 * numEntries;
94 
95   // Then write PLTresolve(), which has two forms: PIC and non-PIC. PLTresolve()
96   // computes the PLT index (by computing the distance from the landing b to
97   // itself) and calls _dl_runtime_resolve() (in glibc).
98   uint32_t got = in.got->getVA();
99   const uint8_t *end = buf + 64;
100   if (config->isPic) {
101     uint32_t afterBcl = 4 * in.plt->getNumEntries() + 12;
102     uint32_t gotBcl = got + 4 - (glink + afterBcl);
103     write32(buf + 0, 0x3d6b0000 | ha(afterBcl));  // addis r11,r11,1f-glink@ha
104     write32(buf + 4, 0x7c0802a6);                 // mflr r0
105     write32(buf + 8, 0x429f0005);                 // bcl 20,30,.+4
106     write32(buf + 12, 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-glink@l
107     write32(buf + 16, 0x7d8802a6);                // mflr r12
108     write32(buf + 20, 0x7c0803a6);                // mtlr r0
109     write32(buf + 24, 0x7d6c5850);                // sub r11,r11,r12
110     write32(buf + 28, 0x3d8c0000 | ha(gotBcl));   // addis 12,12,GOT+4-1b@ha
111     if (ha(gotBcl) == ha(gotBcl + 4)) {
112       write32(buf + 32, 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12)
113       write32(buf + 36,
114               0x818c0000 | lo(gotBcl + 4));       // lwz r12,r12,GOT+8-1b@l(r12)
115     } else {
116       write32(buf + 32, 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12)
117       write32(buf + 36, 0x818c0000 | 4);          // lwz r12,r12,4(r12)
118     }
119     write32(buf + 40, 0x7c0903a6);                // mtctr 0
120     write32(buf + 44, 0x7c0b5a14);                // add r0,11,11
121     write32(buf + 48, 0x7d605a14);                // add r11,0,11
122     write32(buf + 52, 0x4e800420);                // bctr
123     buf += 56;
124   } else {
125     write32(buf + 0, 0x3d800000 | ha(got + 4));   // lis     r12,GOT+4@ha
126     write32(buf + 4, 0x3d6b0000 | ha(-glink));    // addis   r11,r11,-glink@ha
127     if (ha(got + 4) == ha(got + 8))
128       write32(buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12)
129     else
130       write32(buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12)
131     write32(buf + 12, 0x396b0000 | lo(-glink));   // addi    r11,r11,-glink@l
132     write32(buf + 16, 0x7c0903a6);                // mtctr   r0
133     write32(buf + 20, 0x7c0b5a14);                // add     r0,r11,r11
134     if (ha(got + 4) == ha(got + 8))
135       write32(buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@l(r12)
136     else
137       write32(buf + 24, 0x818c0000 | 4);          // lwz r12,4(r12)
138     write32(buf + 28, 0x7d605a14);                // add     r11,r0,r11
139     write32(buf + 32, 0x4e800420);                // bctr
140     buf += 36;
141   }
142 
143   // Pad with nop. They should not be executed.
144   for (; buf < end; buf += 4)
145     write32(buf, 0x60000000);
146 }
147 
148 PPC::PPC() {
149   copyRel = R_PPC_COPY;
150   gotRel = R_PPC_GLOB_DAT;
151   noneRel = R_PPC_NONE;
152   pltRel = R_PPC_JMP_SLOT;
153   relativeRel = R_PPC_RELATIVE;
154   iRelativeRel = R_PPC_IRELATIVE;
155   symbolicRel = R_PPC_ADDR32;
156   gotBaseSymInGotPlt = false;
157   gotHeaderEntriesNum = 3;
158   gotPltHeaderEntriesNum = 0;
159   pltHeaderSize = 0;
160   pltEntrySize = 4;
161   ipltEntrySize = 16;
162 
163   needsThunks = true;
164 
165   tlsModuleIndexRel = R_PPC_DTPMOD32;
166   tlsOffsetRel = R_PPC_DTPREL32;
167   tlsGotRel = R_PPC_TPREL32;
168 
169   defaultMaxPageSize = 65536;
170   defaultImageBase = 0x10000000;
171 
172   write32(trapInstr.data(), 0x7fe00008);
173 }
174 
175 void PPC::writeIplt(uint8_t *buf, const Symbol &sym,
176                     uint64_t /*pltEntryAddr*/) const {
177   // In -pie or -shared mode, assume r30 points to .got2+0x8000, and use a
178   // .got2.plt_pic32. thunk.
179   writePPC32PltCallStub(buf, sym.getGotPltVA(), sym.file, 0x8000);
180 }
181 
182 void PPC::writeGotHeader(uint8_t *buf) const {
183   // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC
184   // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1],
185   // link_map in _GLOBAL_OFFSET_TABLE_[2].
186   write32(buf, mainPart->dynamic->getVA());
187 }
188 
189 void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const {
190   // Address of the symbol resolver stub in .glink .
191   write32(buf, in.plt->getVA() + in.plt->headerSize + 4 * s.pltIndex);
192 }
193 
194 bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file,
195                      uint64_t branchAddr, const Symbol &s, int64_t a) const {
196   if (type != R_PPC_LOCAL24PC && type != R_PPC_REL24 && type != R_PPC_PLTREL24)
197     return false;
198   if (s.isInPlt())
199     return true;
200   if (s.isUndefWeak())
201     return false;
202   return !PPC::inBranchRange(type, branchAddr, s.getVA(a));
203 }
204 
205 uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; }
206 
207 bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const {
208   uint64_t offset = dst - src;
209   if (type == R_PPC_LOCAL24PC || type == R_PPC_REL24 || type == R_PPC_PLTREL24)
210     return isInt<26>(offset);
211   llvm_unreachable("unsupported relocation type used in branch");
212 }
213 
214 RelExpr PPC::getRelExpr(RelType type, const Symbol &s,
215                         const uint8_t *loc) const {
216   switch (type) {
217   case R_PPC_NONE:
218     return R_NONE;
219   case R_PPC_ADDR16_HA:
220   case R_PPC_ADDR16_HI:
221   case R_PPC_ADDR16_LO:
222   case R_PPC_ADDR32:
223     return R_ABS;
224   case R_PPC_DTPREL16:
225   case R_PPC_DTPREL16_HA:
226   case R_PPC_DTPREL16_HI:
227   case R_PPC_DTPREL16_LO:
228   case R_PPC_DTPREL32:
229     return R_DTPREL;
230   case R_PPC_REL14:
231   case R_PPC_REL32:
232   case R_PPC_REL16_LO:
233   case R_PPC_REL16_HI:
234   case R_PPC_REL16_HA:
235     return R_PC;
236   case R_PPC_GOT16:
237     return R_GOT_OFF;
238   case R_PPC_LOCAL24PC:
239   case R_PPC_REL24:
240     return R_PLT_PC;
241   case R_PPC_PLTREL24:
242     return R_PPC32_PLTREL;
243   case R_PPC_GOT_TLSGD16:
244     return R_TLSGD_GOT;
245   case R_PPC_GOT_TLSLD16:
246     return R_TLSLD_GOT;
247   case R_PPC_GOT_TPREL16:
248     return R_GOT_OFF;
249   case R_PPC_TLS:
250     return R_TLSIE_HINT;
251   case R_PPC_TLSGD:
252     return R_TLSDESC_CALL;
253   case R_PPC_TLSLD:
254     return R_TLSLD_HINT;
255   case R_PPC_TPREL16:
256   case R_PPC_TPREL16_HA:
257   case R_PPC_TPREL16_LO:
258   case R_PPC_TPREL16_HI:
259     return R_TLS;
260   default:
261     error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) +
262           ") against symbol " + toString(s));
263     return R_NONE;
264   }
265 }
266 
267 RelType PPC::getDynRel(RelType type) const {
268   if (type == R_PPC_ADDR32)
269     return type;
270   return R_PPC_NONE;
271 }
272 
273 static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) {
274   uint64_t dtpBiasedVal = val - 0x8000;
275   switch (type) {
276   case R_PPC_DTPREL16:
277     return {R_PPC64_ADDR16, dtpBiasedVal};
278   case R_PPC_DTPREL16_HA:
279     return {R_PPC_ADDR16_HA, dtpBiasedVal};
280   case R_PPC_DTPREL16_HI:
281     return {R_PPC_ADDR16_HI, dtpBiasedVal};
282   case R_PPC_DTPREL16_LO:
283     return {R_PPC_ADDR16_LO, dtpBiasedVal};
284   case R_PPC_DTPREL32:
285     return {R_PPC_ADDR32, dtpBiasedVal};
286   default:
287     return {type, val};
288   }
289 }
290 
291 void PPC::relocateOne(uint8_t *loc, RelType type, uint64_t val) const {
292   RelType newType;
293   std::tie(newType, val) = fromDTPREL(type, val);
294   switch (newType) {
295   case R_PPC_ADDR16:
296     checkIntUInt(loc, val, 16, type);
297     write16(loc, val);
298     break;
299   case R_PPC_GOT16:
300   case R_PPC_GOT_TLSGD16:
301   case R_PPC_GOT_TLSLD16:
302   case R_PPC_GOT_TPREL16:
303   case R_PPC_TPREL16:
304     checkInt(loc, val, 16, type);
305     write16(loc, val);
306     break;
307   case R_PPC_ADDR16_HA:
308   case R_PPC_DTPREL16_HA:
309   case R_PPC_GOT_TLSGD16_HA:
310   case R_PPC_GOT_TLSLD16_HA:
311   case R_PPC_GOT_TPREL16_HA:
312   case R_PPC_REL16_HA:
313   case R_PPC_TPREL16_HA:
314     write16(loc, ha(val));
315     break;
316   case R_PPC_ADDR16_HI:
317   case R_PPC_DTPREL16_HI:
318   case R_PPC_GOT_TLSGD16_HI:
319   case R_PPC_GOT_TLSLD16_HI:
320   case R_PPC_GOT_TPREL16_HI:
321   case R_PPC_REL16_HI:
322   case R_PPC_TPREL16_HI:
323     write16(loc, val >> 16);
324     break;
325   case R_PPC_ADDR16_LO:
326   case R_PPC_DTPREL16_LO:
327   case R_PPC_GOT_TLSGD16_LO:
328   case R_PPC_GOT_TLSLD16_LO:
329   case R_PPC_GOT_TPREL16_LO:
330   case R_PPC_REL16_LO:
331   case R_PPC_TPREL16_LO:
332     write16(loc, val);
333     break;
334   case R_PPC_ADDR32:
335   case R_PPC_REL32:
336     write32(loc, val);
337     break;
338   case R_PPC_REL14: {
339     uint32_t mask = 0x0000FFFC;
340     checkInt(loc, val, 16, type);
341     checkAlignment(loc, val, 4, type);
342     write32(loc, (read32(loc) & ~mask) | (val & mask));
343     break;
344   }
345   case R_PPC_REL24:
346   case R_PPC_LOCAL24PC:
347   case R_PPC_PLTREL24: {
348     uint32_t mask = 0x03FFFFFC;
349     checkInt(loc, val, 26, type);
350     checkAlignment(loc, val, 4, type);
351     write32(loc, (read32(loc) & ~mask) | (val & mask));
352     break;
353   }
354   default:
355     llvm_unreachable("unknown relocation");
356   }
357 }
358 
359 RelExpr PPC::adjustRelaxExpr(RelType type, const uint8_t *data,
360                              RelExpr expr) const {
361   if (expr == R_RELAX_TLS_GD_TO_IE)
362     return R_RELAX_TLS_GD_TO_IE_GOT_OFF;
363   if (expr == R_RELAX_TLS_LD_TO_LE)
364     return R_RELAX_TLS_LD_TO_LE_ABS;
365   return expr;
366 }
367 
368 int PPC::getTlsGdRelaxSkip(RelType type) const {
369   // A __tls_get_addr call instruction is marked with 2 relocations:
370   //
371   //   R_PPC_TLSGD / R_PPC_TLSLD: marker relocation
372   //   R_PPC_REL24: __tls_get_addr
373   //
374   // After the relaxation we no longer call __tls_get_addr and should skip both
375   // relocations to not create a false dependence on __tls_get_addr being
376   // defined.
377   if (type == R_PPC_TLSGD || type == R_PPC_TLSLD)
378     return 2;
379   return 1;
380 }
381 
382 void PPC::relaxTlsGdToIe(uint8_t *loc, RelType type, uint64_t val) const {
383   switch (type) {
384   case R_PPC_GOT_TLSGD16: {
385     // addi rT, rA, x@got@tlsgd --> lwz rT, x@got@tprel(rA)
386     uint32_t insn = readFromHalf16(loc);
387     writeFromHalf16(loc, 0x80000000 | (insn & 0x03ff0000));
388     relocateOne(loc, R_PPC_GOT_TPREL16, val);
389     break;
390   }
391   case R_PPC_TLSGD:
392     // bl __tls_get_addr(x@tldgd) --> add r3, r3, r2
393     write32(loc, 0x7c631214);
394     break;
395   default:
396     llvm_unreachable("unsupported relocation for TLS GD to IE relaxation");
397   }
398 }
399 
400 void PPC::relaxTlsGdToLe(uint8_t *loc, RelType type, uint64_t val) const {
401   switch (type) {
402   case R_PPC_GOT_TLSGD16:
403     // addi r3, r31, x@got@tlsgd --> addis r3, r2, x@tprel@ha
404     writeFromHalf16(loc, 0x3c620000 | ha(val));
405     break;
406   case R_PPC_TLSGD:
407     // bl __tls_get_addr(x@tldgd) --> add r3, r3, x@tprel@l
408     write32(loc, 0x38630000 | lo(val));
409     break;
410   default:
411     llvm_unreachable("unsupported relocation for TLS GD to LE relaxation");
412   }
413 }
414 
415 void PPC::relaxTlsLdToLe(uint8_t *loc, RelType type, uint64_t val) const {
416   switch (type) {
417   case R_PPC_GOT_TLSLD16:
418     // addi r3, rA, x@got@tlsgd --> addis r3, r2, 0
419     writeFromHalf16(loc, 0x3c620000);
420     break;
421   case R_PPC_TLSLD:
422     // r3+x@dtprel computes r3+x-0x8000, while we want it to compute r3+x@tprel
423     // = r3+x-0x7000, so add 4096 to r3.
424     // bl __tls_get_addr(x@tlsld) --> addi r3, r3, 4096
425     write32(loc, 0x38631000);
426     break;
427   case R_PPC_DTPREL16:
428   case R_PPC_DTPREL16_HA:
429   case R_PPC_DTPREL16_HI:
430   case R_PPC_DTPREL16_LO:
431     relocateOne(loc, type, val);
432     break;
433   default:
434     llvm_unreachable("unsupported relocation for TLS LD to LE relaxation");
435   }
436 }
437 
438 void PPC::relaxTlsIeToLe(uint8_t *loc, RelType type, uint64_t val) const {
439   switch (type) {
440   case R_PPC_GOT_TPREL16: {
441     // lwz rT, x@got@tprel(rA) --> addis rT, r2, x@tprel@ha
442     uint32_t rt = readFromHalf16(loc) & 0x03e00000;
443     writeFromHalf16(loc, 0x3c020000 | rt | ha(val));
444     break;
445   }
446   case R_PPC_TLS: {
447     uint32_t insn = read32(loc);
448     if (insn >> 26 != 31)
449       error("unrecognized instruction for IE to LE R_PPC_TLS");
450     // addi rT, rT, x@tls --> addi rT, rT, x@tprel@l
451     uint32_t dFormOp = getPPCDFormOp((read32(loc) & 0x000007fe) >> 1);
452     if (dFormOp == 0)
453       error("unrecognized instruction for IE to LE R_PPC_TLS");
454     write32(loc, (dFormOp << 26) | (insn & 0x03ff0000) | lo(val));
455     break;
456   }
457   default:
458     llvm_unreachable("unsupported relocation for TLS IE to LE relaxation");
459   }
460 }
461 
462 TargetInfo *getPPCTargetInfo() {
463   static PPC target;
464   return &target;
465 }
466 
467 } // namespace elf
468 } // namespace lld
469