xref: /freebsd/contrib/llvm-project/lld/ELF/Arch/PPC.cpp (revision 2f513db72b034fd5ef7f080b11be5c711c15186a)
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 "lld/Common/ErrorHandler.h"
14 #include "llvm/Support/Endian.h"
15 
16 using namespace llvm;
17 using namespace llvm::support::endian;
18 using namespace llvm::ELF;
19 using namespace lld;
20 using namespace lld::elf;
21 
22 namespace {
23 class PPC final : public TargetInfo {
24 public:
25   PPC();
26   RelExpr getRelExpr(RelType type, const Symbol &s,
27                      const uint8_t *loc) const override;
28   RelType getDynRel(RelType type) const override;
29   void writeGotHeader(uint8_t *buf) const override;
30   void writePltHeader(uint8_t *buf) const override {
31     llvm_unreachable("should call writePPC32GlinkSection() instead");
32   }
33   void writePlt(uint8_t *buf, uint64_t gotPltEntryAddr, uint64_t pltEntryAddr,
34     int32_t index, unsigned relOff) const override {
35     llvm_unreachable("should call writePPC32GlinkSection() instead");
36   }
37   void writeGotPlt(uint8_t *buf, const Symbol &s) const override;
38   bool needsThunk(RelExpr expr, RelType relocType, const InputFile *file,
39                   uint64_t branchAddr, const Symbol &s) const override;
40   uint32_t getThunkSectionSpacing() const override;
41   bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override;
42   void relocateOne(uint8_t *loc, RelType type, uint64_t val) const override;
43   RelExpr adjustRelaxExpr(RelType type, const uint8_t *data,
44                           RelExpr expr) const override;
45   int getTlsGdRelaxSkip(RelType type) const override;
46   void relaxTlsGdToIe(uint8_t *loc, RelType type, uint64_t val) const override;
47   void relaxTlsGdToLe(uint8_t *loc, RelType type, uint64_t val) const override;
48   void relaxTlsLdToLe(uint8_t *loc, RelType type, uint64_t val) const override;
49   void relaxTlsIeToLe(uint8_t *loc, RelType type, uint64_t val) const override;
50 };
51 } // namespace
52 
53 static uint16_t lo(uint32_t v) { return v; }
54 static uint16_t ha(uint32_t v) { return (v + 0x8000) >> 16; }
55 
56 static uint32_t readFromHalf16(const uint8_t *loc) {
57   return read32(config->isLE ? loc : loc - 2);
58 }
59 
60 static void writeFromHalf16(uint8_t *loc, uint32_t insn) {
61   write32(config->isLE ? loc : loc - 2, insn);
62 }
63 
64 void elf::writePPC32GlinkSection(uint8_t *buf, size_t numEntries) {
65   // On PPC Secure PLT ABI, bl foo@plt jumps to a call stub, which loads an
66   // absolute address from a specific .plt slot (usually called .got.plt on
67   // other targets) and jumps there.
68   //
69   // a) With immediate binding (BIND_NOW), the .plt entry is resolved at load
70   // time. The .glink section is not used.
71   // b) With lazy binding, the .plt entry points to a `b PLTresolve`
72   // instruction in .glink, filled in by PPC::writeGotPlt().
73 
74   // Write N `b PLTresolve` first.
75   for (size_t i = 0; i != numEntries; ++i)
76     write32(buf + 4 * i, 0x48000000 | 4 * (numEntries - i));
77   buf += 4 * numEntries;
78 
79   // Then write PLTresolve(), which has two forms: PIC and non-PIC. PLTresolve()
80   // computes the PLT index (by computing the distance from the landing b to
81   // itself) and calls _dl_runtime_resolve() (in glibc).
82   uint32_t got = in.got->getVA();
83   uint32_t glink = in.plt->getVA(); // VA of .glink
84   const uint8_t *end = buf + 64;
85   if (config->isPic) {
86     uint32_t afterBcl = in.plt->getSize() - target->pltHeaderSize + 12;
87     uint32_t gotBcl = got + 4 - (glink + afterBcl);
88     write32(buf + 0, 0x3d6b0000 | ha(afterBcl));  // addis r11,r11,1f-glink@ha
89     write32(buf + 4, 0x7c0802a6);                 // mflr r0
90     write32(buf + 8, 0x429f0005);                 // bcl 20,30,.+4
91     write32(buf + 12, 0x396b0000 | lo(afterBcl)); // 1: addi r11,r11,1b-.glink@l
92     write32(buf + 16, 0x7d8802a6);                // mflr r12
93     write32(buf + 20, 0x7c0803a6);                // mtlr r0
94     write32(buf + 24, 0x7d6c5850);                // sub r11,r11,r12
95     write32(buf + 28, 0x3d8c0000 | ha(gotBcl));   // addis 12,12,GOT+4-1b@ha
96     if (ha(gotBcl) == ha(gotBcl + 4)) {
97       write32(buf + 32, 0x800c0000 | lo(gotBcl)); // lwz r0,r12,GOT+4-1b@l(r12)
98       write32(buf + 36,
99               0x818c0000 | lo(gotBcl + 4));       // lwz r12,r12,GOT+8-1b@l(r12)
100     } else {
101       write32(buf + 32, 0x840c0000 | lo(gotBcl)); // lwzu r0,r12,GOT+4-1b@l(r12)
102       write32(buf + 36, 0x818c0000 | 4);          // lwz r12,r12,4(r12)
103     }
104     write32(buf + 40, 0x7c0903a6);                // mtctr 0
105     write32(buf + 44, 0x7c0b5a14);                // add r0,11,11
106     write32(buf + 48, 0x7d605a14);                // add r11,0,11
107     write32(buf + 52, 0x4e800420);                // bctr
108     buf += 56;
109   } else {
110     write32(buf + 0, 0x3d800000 | ha(got + 4));   // lis     r12,GOT+4@ha
111     write32(buf + 4, 0x3d6b0000 | ha(-glink));    // addis   r11,r11,-Glink@ha
112     if (ha(got + 4) == ha(got + 8))
113       write32(buf + 8, 0x800c0000 | lo(got + 4)); // lwz r0,GOT+4@l(r12)
114     else
115       write32(buf + 8, 0x840c0000 | lo(got + 4)); // lwzu r0,GOT+4@l(r12)
116     write32(buf + 12, 0x396b0000 | lo(-glink));   // addi    r11,r11,-Glink@l
117     write32(buf + 16, 0x7c0903a6);                // mtctr   r0
118     write32(buf + 20, 0x7c0b5a14);                // add     r0,r11,r11
119     if (ha(got + 4) == ha(got + 8))
120       write32(buf + 24, 0x818c0000 | lo(got + 8)); // lwz r12,GOT+8@ha(r12)
121     else
122       write32(buf + 24, 0x818c0000 | 4);          // lwz r12,4(r12)
123     write32(buf + 28, 0x7d605a14);                // add     r11,r0,r11
124     write32(buf + 32, 0x4e800420);                // bctr
125     buf += 36;
126   }
127 
128   // Pad with nop. They should not be executed.
129   for (; buf < end; buf += 4)
130     write32(buf, 0x60000000);
131 }
132 
133 PPC::PPC() {
134   gotRel = R_PPC_GLOB_DAT;
135   noneRel = R_PPC_NONE;
136   pltRel = R_PPC_JMP_SLOT;
137   relativeRel = R_PPC_RELATIVE;
138   iRelativeRel = R_PPC_IRELATIVE;
139   symbolicRel = R_PPC_ADDR32;
140   gotBaseSymInGotPlt = false;
141   gotHeaderEntriesNum = 3;
142   gotPltHeaderEntriesNum = 0;
143   pltHeaderSize = 64; // size of PLTresolve in .glink
144   pltEntrySize = 4;
145 
146   needsThunks = true;
147 
148   tlsModuleIndexRel = R_PPC_DTPMOD32;
149   tlsOffsetRel = R_PPC_DTPREL32;
150   tlsGotRel = R_PPC_TPREL32;
151 
152   defaultMaxPageSize = 65536;
153   defaultImageBase = 0x10000000;
154 
155   write32(trapInstr.data(), 0x7fe00008);
156 }
157 
158 void PPC::writeGotHeader(uint8_t *buf) const {
159   // _GLOBAL_OFFSET_TABLE_[0] = _DYNAMIC
160   // glibc stores _dl_runtime_resolve in _GLOBAL_OFFSET_TABLE_[1],
161   // link_map in _GLOBAL_OFFSET_TABLE_[2].
162   write32(buf, mainPart->dynamic->getVA());
163 }
164 
165 void PPC::writeGotPlt(uint8_t *buf, const Symbol &s) const {
166   // Address of the symbol resolver stub in .glink .
167   write32(buf, in.plt->getVA() + 4 * s.pltIndex);
168 }
169 
170 bool PPC::needsThunk(RelExpr expr, RelType type, const InputFile *file,
171                      uint64_t branchAddr, const Symbol &s) const {
172   if (type != R_PPC_REL24 && type != R_PPC_PLTREL24)
173     return false;
174   if (s.isInPlt())
175     return true;
176   if (s.isUndefWeak())
177     return false;
178   return !(expr == R_PC && PPC::inBranchRange(type, branchAddr, s.getVA()));
179 }
180 
181 uint32_t PPC::getThunkSectionSpacing() const { return 0x2000000; }
182 
183 bool PPC::inBranchRange(RelType type, uint64_t src, uint64_t dst) const {
184   uint64_t offset = dst - src;
185   if (type == R_PPC_REL24 || type == R_PPC_PLTREL24)
186     return isInt<26>(offset);
187   llvm_unreachable("unsupported relocation type used in branch");
188 }
189 
190 RelExpr PPC::getRelExpr(RelType type, const Symbol &s,
191                         const uint8_t *loc) const {
192   switch (type) {
193   case R_PPC_NONE:
194     return R_NONE;
195   case R_PPC_ADDR16_HA:
196   case R_PPC_ADDR16_HI:
197   case R_PPC_ADDR16_LO:
198   case R_PPC_ADDR32:
199     return R_ABS;
200   case R_PPC_DTPREL16:
201   case R_PPC_DTPREL16_HA:
202   case R_PPC_DTPREL16_HI:
203   case R_PPC_DTPREL16_LO:
204   case R_PPC_DTPREL32:
205     return R_DTPREL;
206   case R_PPC_REL14:
207   case R_PPC_REL32:
208   case R_PPC_LOCAL24PC:
209   case R_PPC_REL16_LO:
210   case R_PPC_REL16_HI:
211   case R_PPC_REL16_HA:
212     return R_PC;
213   case R_PPC_GOT16:
214     return R_GOT_OFF;
215   case R_PPC_REL24:
216     return R_PLT_PC;
217   case R_PPC_PLTREL24:
218     return R_PPC32_PLTREL;
219   case R_PPC_GOT_TLSGD16:
220     return R_TLSGD_GOT;
221   case R_PPC_GOT_TLSLD16:
222     return R_TLSLD_GOT;
223   case R_PPC_GOT_TPREL16:
224     return R_GOT_OFF;
225   case R_PPC_TLS:
226     return R_TLSIE_HINT;
227   case R_PPC_TLSGD:
228     return R_TLSDESC_CALL;
229   case R_PPC_TLSLD:
230     return R_TLSLD_HINT;
231   case R_PPC_TPREL16:
232   case R_PPC_TPREL16_HA:
233   case R_PPC_TPREL16_LO:
234   case R_PPC_TPREL16_HI:
235     return R_TLS;
236   default:
237     error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) +
238           ") against symbol " + toString(s));
239     return R_NONE;
240   }
241 }
242 
243 RelType PPC::getDynRel(RelType type) const {
244   if (type == R_PPC_ADDR32)
245     return type;
246   return R_PPC_NONE;
247 }
248 
249 static std::pair<RelType, uint64_t> fromDTPREL(RelType type, uint64_t val) {
250   uint64_t dtpBiasedVal = val - 0x8000;
251   switch (type) {
252   case R_PPC_DTPREL16:
253     return {R_PPC64_ADDR16, dtpBiasedVal};
254   case R_PPC_DTPREL16_HA:
255     return {R_PPC_ADDR16_HA, dtpBiasedVal};
256   case R_PPC_DTPREL16_HI:
257     return {R_PPC_ADDR16_HI, dtpBiasedVal};
258   case R_PPC_DTPREL16_LO:
259     return {R_PPC_ADDR16_LO, dtpBiasedVal};
260   case R_PPC_DTPREL32:
261     return {R_PPC_ADDR32, dtpBiasedVal};
262   default:
263     return {type, val};
264   }
265 }
266 
267 void PPC::relocateOne(uint8_t *loc, RelType type, uint64_t val) const {
268   RelType newType;
269   std::tie(newType, val) = fromDTPREL(type, val);
270   switch (newType) {
271   case R_PPC_ADDR16:
272     checkIntUInt(loc, val, 16, type);
273     write16(loc, val);
274     break;
275   case R_PPC_GOT16:
276   case R_PPC_GOT_TLSGD16:
277   case R_PPC_GOT_TLSLD16:
278   case R_PPC_GOT_TPREL16:
279   case R_PPC_TPREL16:
280     checkInt(loc, val, 16, type);
281     write16(loc, val);
282     break;
283   case R_PPC_ADDR16_HA:
284   case R_PPC_DTPREL16_HA:
285   case R_PPC_GOT_TLSGD16_HA:
286   case R_PPC_GOT_TLSLD16_HA:
287   case R_PPC_GOT_TPREL16_HA:
288   case R_PPC_REL16_HA:
289   case R_PPC_TPREL16_HA:
290     write16(loc, ha(val));
291     break;
292   case R_PPC_ADDR16_HI:
293   case R_PPC_DTPREL16_HI:
294   case R_PPC_GOT_TLSGD16_HI:
295   case R_PPC_GOT_TLSLD16_HI:
296   case R_PPC_GOT_TPREL16_HI:
297   case R_PPC_REL16_HI:
298   case R_PPC_TPREL16_HI:
299     write16(loc, val >> 16);
300     break;
301   case R_PPC_ADDR16_LO:
302   case R_PPC_DTPREL16_LO:
303   case R_PPC_GOT_TLSGD16_LO:
304   case R_PPC_GOT_TLSLD16_LO:
305   case R_PPC_GOT_TPREL16_LO:
306   case R_PPC_REL16_LO:
307   case R_PPC_TPREL16_LO:
308     write16(loc, val);
309     break;
310   case R_PPC_ADDR32:
311   case R_PPC_REL32:
312     write32(loc, val);
313     break;
314   case R_PPC_REL14: {
315     uint32_t mask = 0x0000FFFC;
316     checkInt(loc, val, 16, type);
317     checkAlignment(loc, val, 4, type);
318     write32(loc, (read32(loc) & ~mask) | (val & mask));
319     break;
320   }
321   case R_PPC_REL24:
322   case R_PPC_LOCAL24PC:
323   case R_PPC_PLTREL24: {
324     uint32_t mask = 0x03FFFFFC;
325     checkInt(loc, val, 26, type);
326     checkAlignment(loc, val, 4, type);
327     write32(loc, (read32(loc) & ~mask) | (val & mask));
328     break;
329   }
330   default:
331     llvm_unreachable("unknown relocation");
332   }
333 }
334 
335 RelExpr PPC::adjustRelaxExpr(RelType type, const uint8_t *data,
336                              RelExpr expr) const {
337   if (expr == R_RELAX_TLS_GD_TO_IE)
338     return R_RELAX_TLS_GD_TO_IE_GOT_OFF;
339   if (expr == R_RELAX_TLS_LD_TO_LE)
340     return R_RELAX_TLS_LD_TO_LE_ABS;
341   return expr;
342 }
343 
344 int PPC::getTlsGdRelaxSkip(RelType type) const {
345   // A __tls_get_addr call instruction is marked with 2 relocations:
346   //
347   //   R_PPC_TLSGD / R_PPC_TLSLD: marker relocation
348   //   R_PPC_REL24: __tls_get_addr
349   //
350   // After the relaxation we no longer call __tls_get_addr and should skip both
351   // relocations to not create a false dependence on __tls_get_addr being
352   // defined.
353   if (type == R_PPC_TLSGD || type == R_PPC_TLSLD)
354     return 2;
355   return 1;
356 }
357 
358 void PPC::relaxTlsGdToIe(uint8_t *loc, RelType type, uint64_t val) const {
359   switch (type) {
360   case R_PPC_GOT_TLSGD16: {
361     // addi rT, rA, x@got@tlsgd --> lwz rT, x@got@tprel(rA)
362     uint32_t insn = readFromHalf16(loc);
363     writeFromHalf16(loc, 0x80000000 | (insn & 0x03ff0000));
364     relocateOne(loc, R_PPC_GOT_TPREL16, val);
365     break;
366   }
367   case R_PPC_TLSGD:
368     // bl __tls_get_addr(x@tldgd) --> add r3, r3, r2
369     write32(loc, 0x7c631214);
370     break;
371   default:
372     llvm_unreachable("unsupported relocation for TLS GD to IE relaxation");
373   }
374 }
375 
376 void PPC::relaxTlsGdToLe(uint8_t *loc, RelType type, uint64_t val) const {
377   switch (type) {
378   case R_PPC_GOT_TLSGD16:
379     // addi r3, r31, x@got@tlsgd --> addis r3, r2, x@tprel@ha
380     writeFromHalf16(loc, 0x3c620000 | ha(val));
381     break;
382   case R_PPC_TLSGD:
383     // bl __tls_get_addr(x@tldgd) --> add r3, r3, x@tprel@l
384     write32(loc, 0x38630000 | lo(val));
385     break;
386   default:
387     llvm_unreachable("unsupported relocation for TLS GD to LE relaxation");
388   }
389 }
390 
391 void PPC::relaxTlsLdToLe(uint8_t *loc, RelType type, uint64_t val) const {
392   switch (type) {
393   case R_PPC_GOT_TLSLD16:
394     // addi r3, rA, x@got@tlsgd --> addis r3, r2, 0
395     writeFromHalf16(loc, 0x3c620000);
396     break;
397   case R_PPC_TLSLD:
398     // r3+x@dtprel computes r3+x-0x8000, while we want it to compute r3+x@tprel
399     // = r3+x-0x7000, so add 4096 to r3.
400     // bl __tls_get_addr(x@tlsld) --> addi r3, r3, 4096
401     write32(loc, 0x38631000);
402     break;
403   case R_PPC_DTPREL16:
404   case R_PPC_DTPREL16_HA:
405   case R_PPC_DTPREL16_HI:
406   case R_PPC_DTPREL16_LO:
407     relocateOne(loc, type, val);
408     break;
409   default:
410     llvm_unreachable("unsupported relocation for TLS LD to LE relaxation");
411   }
412 }
413 
414 void PPC::relaxTlsIeToLe(uint8_t *loc, RelType type, uint64_t val) const {
415   switch (type) {
416   case R_PPC_GOT_TPREL16: {
417     // lwz rT, x@got@tprel(rA) --> addis rT, r2, x@tprel@ha
418     uint32_t rt = readFromHalf16(loc) & 0x03e00000;
419     writeFromHalf16(loc, 0x3c020000 | rt | ha(val));
420     break;
421   }
422   case R_PPC_TLS: {
423     uint32_t insn = read32(loc);
424     if (insn >> 26 != 31)
425       error("unrecognized instruction for IE to LE R_PPC_TLS");
426     // addi rT, rT, x@tls --> addi rT, rT, x@tprel@l
427     uint32_t dFormOp = getPPCDFormOp((read32(loc) & 0x000007fe) >> 1);
428     if (dFormOp == 0)
429       error("unrecognized instruction for IE to LE R_PPC_TLS");
430     write32(loc, (dFormOp << 26) | (insn & 0x03ff0000) | lo(val));
431     break;
432   }
433   default:
434     llvm_unreachable("unsupported relocation for TLS IE to LE relaxation");
435   }
436 }
437 
438 TargetInfo *elf::getPPCTargetInfo() {
439   static PPC target;
440   return &target;
441 }
442