xref: /llvm-project-15.0.7/lld/ELF/Arch/ARM.cpp (revision ff9e596b)
1 //===- ARM.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 "Symbols.h"
10 #include "SyntheticSections.h"
11 #include "Target.h"
12 #include "lld/Common/ErrorHandler.h"
13 #include "llvm/BinaryFormat/ELF.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 ARM final : public TargetInfo {
24 public:
25   ARM();
26   uint32_t calcEFlags() const override;
27   RelExpr getRelExpr(RelType type, const Symbol &s,
28                      const uint8_t *loc) const override;
29   RelType getDynRel(RelType type) const override;
30   int64_t getImplicitAddend(const uint8_t *buf, RelType type) const override;
31   void writeGotPlt(uint8_t *buf, const Symbol &s) const override;
32   void writeIgotPlt(uint8_t *buf, const Symbol &s) const override;
33   void writePltHeader(uint8_t *buf) const override;
34   void writePlt(uint8_t *buf, const Symbol &sym,
35                 uint64_t pltEntryAddr) const override;
36   void addPltSymbols(InputSection &isec, uint64_t off) const override;
37   void addPltHeaderSymbols(InputSection &isd) const override;
38   bool needsThunk(RelExpr expr, RelType type, const InputFile *file,
39                   uint64_t branchAddr, const Symbol &s,
40                   int64_t a) const override;
41   uint32_t getThunkSectionSpacing() const override;
42   bool inBranchRange(RelType type, uint64_t src, uint64_t dst) const override;
43   void relocate(uint8_t *loc, const Relocation &rel,
44                 uint64_t val) const override;
45 };
46 } // namespace
47 
48 ARM::ARM() {
49   copyRel = R_ARM_COPY;
50   relativeRel = R_ARM_RELATIVE;
51   iRelativeRel = R_ARM_IRELATIVE;
52   gotRel = R_ARM_GLOB_DAT;
53   pltRel = R_ARM_JUMP_SLOT;
54   symbolicRel = R_ARM_ABS32;
55   tlsGotRel = R_ARM_TLS_TPOFF32;
56   tlsModuleIndexRel = R_ARM_TLS_DTPMOD32;
57   tlsOffsetRel = R_ARM_TLS_DTPOFF32;
58   pltHeaderSize = 32;
59   pltEntrySize = 16;
60   ipltEntrySize = 16;
61   trapInstr = {0xd4, 0xd4, 0xd4, 0xd4};
62   needsThunks = true;
63   defaultMaxPageSize = 65536;
64 }
65 
66 uint32_t ARM::calcEFlags() const {
67   // The ABIFloatType is used by loaders to detect the floating point calling
68   // convention.
69   uint32_t abiFloatType = 0;
70   if (config->armVFPArgs == ARMVFPArgKind::Base ||
71       config->armVFPArgs == ARMVFPArgKind::Default)
72     abiFloatType = EF_ARM_ABI_FLOAT_SOFT;
73   else if (config->armVFPArgs == ARMVFPArgKind::VFP)
74     abiFloatType = EF_ARM_ABI_FLOAT_HARD;
75 
76   // We don't currently use any features incompatible with EF_ARM_EABI_VER5,
77   // but we don't have any firm guarantees of conformance. Linux AArch64
78   // kernels (as of 2016) require an EABI version to be set.
79   return EF_ARM_EABI_VER5 | abiFloatType;
80 }
81 
82 RelExpr ARM::getRelExpr(RelType type, const Symbol &s,
83                         const uint8_t *loc) const {
84   switch (type) {
85   case R_ARM_ABS32:
86   case R_ARM_MOVW_ABS_NC:
87   case R_ARM_MOVT_ABS:
88   case R_ARM_THM_MOVW_ABS_NC:
89   case R_ARM_THM_MOVT_ABS:
90     return R_ABS;
91   case R_ARM_THM_JUMP8:
92   case R_ARM_THM_JUMP11:
93     return R_PC;
94   case R_ARM_CALL:
95   case R_ARM_JUMP24:
96   case R_ARM_PC24:
97   case R_ARM_PLT32:
98   case R_ARM_PREL31:
99   case R_ARM_THM_JUMP19:
100   case R_ARM_THM_JUMP24:
101   case R_ARM_THM_CALL:
102     return R_PLT_PC;
103   case R_ARM_GOTOFF32:
104     // (S + A) - GOT_ORG
105     return R_GOTREL;
106   case R_ARM_GOT_BREL:
107     // GOT(S) + A - GOT_ORG
108     return R_GOT_OFF;
109   case R_ARM_GOT_PREL:
110   case R_ARM_TLS_IE32:
111     // GOT(S) + A - P
112     return R_GOT_PC;
113   case R_ARM_SBREL32:
114     return R_ARM_SBREL;
115   case R_ARM_TARGET1:
116     return config->target1Rel ? R_PC : R_ABS;
117   case R_ARM_TARGET2:
118     if (config->target2 == Target2Policy::Rel)
119       return R_PC;
120     if (config->target2 == Target2Policy::Abs)
121       return R_ABS;
122     return R_GOT_PC;
123   case R_ARM_TLS_GD32:
124     return R_TLSGD_PC;
125   case R_ARM_TLS_LDM32:
126     return R_TLSLD_PC;
127   case R_ARM_TLS_LDO32:
128     return R_DTPREL;
129   case R_ARM_BASE_PREL:
130     // B(S) + A - P
131     // FIXME: currently B(S) assumed to be .got, this may not hold for all
132     // platforms.
133     return R_GOTONLY_PC;
134   case R_ARM_MOVW_PREL_NC:
135   case R_ARM_MOVT_PREL:
136   case R_ARM_REL32:
137   case R_ARM_THM_MOVW_PREL_NC:
138   case R_ARM_THM_MOVT_PREL:
139     return R_PC;
140   case R_ARM_ALU_PC_G0:
141   case R_ARM_ALU_PC_G0_NC:
142   case R_ARM_ALU_PC_G1:
143   case R_ARM_ALU_PC_G1_NC:
144   case R_ARM_ALU_PC_G2:
145   case R_ARM_LDR_PC_G0:
146   case R_ARM_LDR_PC_G1:
147   case R_ARM_LDR_PC_G2:
148   case R_ARM_LDRS_PC_G0:
149   case R_ARM_LDRS_PC_G1:
150   case R_ARM_LDRS_PC_G2:
151   case R_ARM_THM_ALU_PREL_11_0:
152   case R_ARM_THM_PC8:
153   case R_ARM_THM_PC12:
154     return R_ARM_PCA;
155   case R_ARM_MOVW_BREL_NC:
156   case R_ARM_MOVW_BREL:
157   case R_ARM_MOVT_BREL:
158   case R_ARM_THM_MOVW_BREL_NC:
159   case R_ARM_THM_MOVW_BREL:
160   case R_ARM_THM_MOVT_BREL:
161     return R_ARM_SBREL;
162   case R_ARM_NONE:
163     return R_NONE;
164   case R_ARM_TLS_LE32:
165     return R_TPREL;
166   case R_ARM_V4BX:
167     // V4BX is just a marker to indicate there's a "bx rN" instruction at the
168     // given address. It can be used to implement a special linker mode which
169     // rewrites ARMv4T inputs to ARMv4. Since we support only ARMv4 input and
170     // not ARMv4 output, we can just ignore it.
171     return R_NONE;
172   default:
173     error(getErrorLocation(loc) + "unknown relocation (" + Twine(type) +
174           ") against symbol " + toString(s));
175     return R_NONE;
176   }
177 }
178 
179 RelType ARM::getDynRel(RelType type) const {
180   if ((type == R_ARM_ABS32) || (type == R_ARM_TARGET1 && !config->target1Rel))
181     return R_ARM_ABS32;
182   return R_ARM_NONE;
183 }
184 
185 void ARM::writeGotPlt(uint8_t *buf, const Symbol &) const {
186   write32le(buf, in.plt->getVA());
187 }
188 
189 void ARM::writeIgotPlt(uint8_t *buf, const Symbol &s) const {
190   // An ARM entry is the address of the ifunc resolver function.
191   write32le(buf, s.getVA());
192 }
193 
194 // Long form PLT Header that does not have any restrictions on the displacement
195 // of the .plt from the .plt.got.
196 static void writePltHeaderLong(uint8_t *buf) {
197   const uint8_t pltData[] = {
198       0x04, 0xe0, 0x2d, 0xe5, //     str lr, [sp,#-4]!
199       0x04, 0xe0, 0x9f, 0xe5, //     ldr lr, L2
200       0x0e, 0xe0, 0x8f, 0xe0, // L1: add lr, pc, lr
201       0x08, 0xf0, 0xbe, 0xe5, //     ldr pc, [lr, #8]
202       0x00, 0x00, 0x00, 0x00, // L2: .word   &(.got.plt) - L1 - 8
203       0xd4, 0xd4, 0xd4, 0xd4, //     Pad to 32-byte boundary
204       0xd4, 0xd4, 0xd4, 0xd4, //     Pad to 32-byte boundary
205       0xd4, 0xd4, 0xd4, 0xd4};
206   memcpy(buf, pltData, sizeof(pltData));
207   uint64_t gotPlt = in.gotPlt->getVA();
208   uint64_t l1 = in.plt->getVA() + 8;
209   write32le(buf + 16, gotPlt - l1 - 8);
210 }
211 
212 // The default PLT header requires the .plt.got to be within 128 Mb of the
213 // .plt in the positive direction.
214 void ARM::writePltHeader(uint8_t *buf) const {
215   // Use a similar sequence to that in writePlt(), the difference is the calling
216   // conventions mean we use lr instead of ip. The PLT entry is responsible for
217   // saving lr on the stack, the dynamic loader is responsible for reloading
218   // it.
219   const uint32_t pltData[] = {
220       0xe52de004, // L1: str lr, [sp,#-4]!
221       0xe28fe600, //     add lr, pc,  #0x0NN00000 &(.got.plt - L1 - 4)
222       0xe28eea00, //     add lr, lr,  #0x000NN000 &(.got.plt - L1 - 4)
223       0xe5bef000, //     ldr pc, [lr, #0x00000NNN] &(.got.plt -L1 - 4)
224   };
225 
226   uint64_t offset = in.gotPlt->getVA() - in.plt->getVA() - 4;
227   if (!llvm::isUInt<27>(offset)) {
228     // We cannot encode the Offset, use the long form.
229     writePltHeaderLong(buf);
230     return;
231   }
232   write32le(buf + 0, pltData[0]);
233   write32le(buf + 4, pltData[1] | ((offset >> 20) & 0xff));
234   write32le(buf + 8, pltData[2] | ((offset >> 12) & 0xff));
235   write32le(buf + 12, pltData[3] | (offset & 0xfff));
236   memcpy(buf + 16, trapInstr.data(), 4); // Pad to 32-byte boundary
237   memcpy(buf + 20, trapInstr.data(), 4);
238   memcpy(buf + 24, trapInstr.data(), 4);
239   memcpy(buf + 28, trapInstr.data(), 4);
240 }
241 
242 void ARM::addPltHeaderSymbols(InputSection &isec) const {
243   addSyntheticLocal("$a", STT_NOTYPE, 0, 0, isec);
244   addSyntheticLocal("$d", STT_NOTYPE, 16, 0, isec);
245 }
246 
247 // Long form PLT entries that do not have any restrictions on the displacement
248 // of the .plt from the .plt.got.
249 static void writePltLong(uint8_t *buf, uint64_t gotPltEntryAddr,
250                          uint64_t pltEntryAddr) {
251   const uint8_t pltData[] = {
252       0x04, 0xc0, 0x9f, 0xe5, //     ldr ip, L2
253       0x0f, 0xc0, 0x8c, 0xe0, // L1: add ip, ip, pc
254       0x00, 0xf0, 0x9c, 0xe5, //     ldr pc, [ip]
255       0x00, 0x00, 0x00, 0x00, // L2: .word   Offset(&(.plt.got) - L1 - 8
256   };
257   memcpy(buf, pltData, sizeof(pltData));
258   uint64_t l1 = pltEntryAddr + 4;
259   write32le(buf + 12, gotPltEntryAddr - l1 - 8);
260 }
261 
262 // The default PLT entries require the .plt.got to be within 128 Mb of the
263 // .plt in the positive direction.
264 void ARM::writePlt(uint8_t *buf, const Symbol &sym,
265                    uint64_t pltEntryAddr) const {
266   // The PLT entry is similar to the example given in Appendix A of ELF for
267   // the Arm Architecture. Instead of using the Group Relocations to find the
268   // optimal rotation for the 8-bit immediate used in the add instructions we
269   // hard code the most compact rotations for simplicity. This saves a load
270   // instruction over the long plt sequences.
271   const uint32_t pltData[] = {
272       0xe28fc600, // L1: add ip, pc,  #0x0NN00000  Offset(&(.plt.got) - L1 - 8
273       0xe28cca00, //     add ip, ip,  #0x000NN000  Offset(&(.plt.got) - L1 - 8
274       0xe5bcf000, //     ldr pc, [ip, #0x00000NNN] Offset(&(.plt.got) - L1 - 8
275   };
276 
277   uint64_t offset = sym.getGotPltVA() - pltEntryAddr - 8;
278   if (!llvm::isUInt<27>(offset)) {
279     // We cannot encode the Offset, use the long form.
280     writePltLong(buf, sym.getGotPltVA(), pltEntryAddr);
281     return;
282   }
283   write32le(buf + 0, pltData[0] | ((offset >> 20) & 0xff));
284   write32le(buf + 4, pltData[1] | ((offset >> 12) & 0xff));
285   write32le(buf + 8, pltData[2] | (offset & 0xfff));
286   memcpy(buf + 12, trapInstr.data(), 4); // Pad to 16-byte boundary
287 }
288 
289 void ARM::addPltSymbols(InputSection &isec, uint64_t off) const {
290   addSyntheticLocal("$a", STT_NOTYPE, off, 0, isec);
291   addSyntheticLocal("$d", STT_NOTYPE, off + 12, 0, isec);
292 }
293 
294 bool ARM::needsThunk(RelExpr expr, RelType type, const InputFile *file,
295                      uint64_t branchAddr, const Symbol &s,
296                      int64_t a) const {
297   // If S is an undefined weak symbol and does not have a PLT entry then it
298   // will be resolved as a branch to the next instruction.
299   if (s.isUndefWeak() && !s.isInPlt())
300     return false;
301   // A state change from ARM to Thumb and vice versa must go through an
302   // interworking thunk if the relocation type is not R_ARM_CALL or
303   // R_ARM_THM_CALL.
304   switch (type) {
305   case R_ARM_PC24:
306   case R_ARM_PLT32:
307   case R_ARM_JUMP24:
308     // Source is ARM, all PLT entries are ARM so no interworking required.
309     // Otherwise we need to interwork if STT_FUNC Symbol has bit 0 set (Thumb).
310     if (s.isFunc() && expr == R_PC && (s.getVA() & 1))
311       return true;
312     LLVM_FALLTHROUGH;
313   case R_ARM_CALL: {
314     uint64_t dst = (expr == R_PLT_PC) ? s.getPltVA() : s.getVA();
315     return !inBranchRange(type, branchAddr, dst + a);
316   }
317   case R_ARM_THM_JUMP19:
318   case R_ARM_THM_JUMP24:
319     // Source is Thumb, all PLT entries are ARM so interworking is required.
320     // Otherwise we need to interwork if STT_FUNC Symbol has bit 0 clear (ARM).
321     if (expr == R_PLT_PC || (s.isFunc() && (s.getVA() & 1) == 0))
322       return true;
323     LLVM_FALLTHROUGH;
324   case R_ARM_THM_CALL: {
325     uint64_t dst = (expr == R_PLT_PC) ? s.getPltVA() : s.getVA();
326     return !inBranchRange(type, branchAddr, dst + a);
327   }
328   }
329   return false;
330 }
331 
332 uint32_t ARM::getThunkSectionSpacing() const {
333   // The placing of pre-created ThunkSections is controlled by the value
334   // thunkSectionSpacing returned by getThunkSectionSpacing(). The aim is to
335   // place the ThunkSection such that all branches from the InputSections
336   // prior to the ThunkSection can reach a Thunk placed at the end of the
337   // ThunkSection. Graphically:
338   // | up to thunkSectionSpacing .text input sections |
339   // | ThunkSection                                   |
340   // | up to thunkSectionSpacing .text input sections |
341   // | ThunkSection                                   |
342 
343   // Pre-created ThunkSections are spaced roughly 16MiB apart on ARMv7. This
344   // is to match the most common expected case of a Thumb 2 encoded BL, BLX or
345   // B.W:
346   // ARM B, BL, BLX range +/- 32MiB
347   // Thumb B.W, BL, BLX range +/- 16MiB
348   // Thumb B<cc>.W range +/- 1MiB
349   // If a branch cannot reach a pre-created ThunkSection a new one will be
350   // created so we can handle the rare cases of a Thumb 2 conditional branch.
351   // We intentionally use a lower size for thunkSectionSpacing than the maximum
352   // branch range so the end of the ThunkSection is more likely to be within
353   // range of the branch instruction that is furthest away. The value we shorten
354   // thunkSectionSpacing by is set conservatively to allow us to create 16,384
355   // 12 byte Thunks at any offset in a ThunkSection without risk of a branch to
356   // one of the Thunks going out of range.
357 
358   // On Arm the thunkSectionSpacing depends on the range of the Thumb Branch
359   // range. On earlier Architectures such as ARMv4, ARMv5 and ARMv6 (except
360   // ARMv6T2) the range is +/- 4MiB.
361 
362   return (config->armJ1J2BranchEncoding) ? 0x1000000 - 0x30000
363                                          : 0x400000 - 0x7500;
364 }
365 
366 bool ARM::inBranchRange(RelType type, uint64_t src, uint64_t dst) const {
367   if ((dst & 0x1) == 0)
368     // Destination is ARM, if ARM caller then Src is already 4-byte aligned.
369     // If Thumb Caller (BLX) the Src address has bottom 2 bits cleared to ensure
370     // destination will be 4 byte aligned.
371     src &= ~0x3;
372   else
373     // Bit 0 == 1 denotes Thumb state, it is not part of the range.
374     dst &= ~0x1;
375 
376   int64_t offset = dst - src;
377   switch (type) {
378   case R_ARM_PC24:
379   case R_ARM_PLT32:
380   case R_ARM_JUMP24:
381   case R_ARM_CALL:
382     return llvm::isInt<26>(offset);
383   case R_ARM_THM_JUMP19:
384     return llvm::isInt<21>(offset);
385   case R_ARM_THM_JUMP24:
386   case R_ARM_THM_CALL:
387     return config->armJ1J2BranchEncoding ? llvm::isInt<25>(offset)
388                                          : llvm::isInt<23>(offset);
389   default:
390     return true;
391   }
392 }
393 
394 // Helper to produce message text when LLD detects that a CALL relocation to
395 // a non STT_FUNC symbol that may result in incorrect interworking between ARM
396 // or Thumb.
397 static void stateChangeWarning(uint8_t *loc, RelType relt, const Symbol &s) {
398   assert(!s.isFunc());
399   const ErrorPlace place = getErrorPlace(loc);
400   std::string hint;
401   if (!place.srcLoc.empty())
402     hint = "; " + place.srcLoc;
403   if (s.isSection()) {
404     // Section symbols must be defined and in a section. Users cannot change
405     // the type. Use the section name as getName() returns an empty string.
406     warn(place.loc + "branch and link relocation: " + toString(relt) +
407          " to STT_SECTION symbol " + cast<Defined>(s).section->name +
408          " ; interworking not performed" + hint);
409   } else {
410     // Warn with hint on how to alter the symbol type.
411     warn(getErrorLocation(loc) + "branch and link relocation: " +
412          toString(relt) + " to non STT_FUNC symbol: " + s.getName() +
413          " interworking not performed; consider using directive '.type " +
414          s.getName() +
415          ", %function' to give symbol type STT_FUNC if interworking between "
416          "ARM and Thumb is required" +
417          hint);
418   }
419 }
420 
421 // Rotate a 32-bit unsigned value right by a specified amt of bits.
422 static uint32_t rotr32(uint32_t val, uint32_t amt) {
423   assert(amt < 32 && "Invalid rotate amount");
424   return (val >> amt) | (val << ((32 - amt) & 31));
425 }
426 
427 static std::pair<uint32_t, uint32_t> getRemAndLZForGroup(unsigned group,
428                                                          uint32_t val) {
429   uint32_t rem, lz;
430   do {
431     lz = llvm::countLeadingZeros(val) & ~1;
432     rem = val;
433     if (lz == 32) // implies rem == 0
434       break;
435     val &= 0xffffff >> lz;
436   } while (group--);
437   return {rem, lz};
438 }
439 
440 static void encodeAluGroup(uint8_t *loc, const Relocation &rel, uint64_t val,
441                            int group, bool check) {
442   // ADD/SUB (immediate) add = bit23, sub = bit22
443   // immediate field carries is a 12-bit modified immediate, made up of a 4-bit
444   // even rotate right and an 8-bit immediate.
445   uint32_t opcode = 0x00800000;
446   if (val >> 63) {
447     opcode = 0x00400000;
448     val = -val;
449   }
450   uint32_t imm, lz;
451   std::tie(imm, lz) = getRemAndLZForGroup(group, val);
452   uint32_t rot = 0;
453   if (lz < 24) {
454     imm = rotr32(imm, 24 - lz);
455     rot = (lz + 8) << 7;
456   }
457   if (check && imm > 0xff)
458     error(getErrorLocation(loc) + "unencodeable immediate " + Twine(val).str() +
459           " for relocation " + toString(rel.type));
460   write32le(loc, (read32le(loc) & 0xff3ff000) | opcode | rot | (imm & 0xff));
461 }
462 
463 static void encodeLdrGroup(uint8_t *loc, const Relocation &rel, uint64_t val,
464                            int group) {
465   // R_ARM_LDR_PC_Gn is S + A - P, we have ((S + A) | T) - P, if S is a
466   // function then addr is 0 (modulo 2) and Pa is 0 (modulo 4) so we can clear
467   // bottom bit to recover S + A - P.
468   if (rel.sym->isFunc())
469     val &= ~0x1;
470   // LDR (literal) u = bit23
471   uint32_t opcode = 0x00800000;
472   if (val >> 63) {
473     opcode = 0x0;
474     val = -val;
475   }
476   uint32_t imm = getRemAndLZForGroup(group, val).first;
477   checkUInt(loc, imm, 12, rel);
478   write32le(loc, (read32le(loc) & 0xff7ff000) | opcode | imm);
479 }
480 
481 static void encodeLdrsGroup(uint8_t *loc, const Relocation &rel, uint64_t val,
482                             int group) {
483   // R_ARM_LDRS_PC_Gn is S + A - P, we have ((S + A) | T) - P, if S is a
484   // function then addr is 0 (modulo 2) and Pa is 0 (modulo 4) so we can clear
485   // bottom bit to recover S + A - P.
486   if (rel.sym->isFunc())
487     val &= ~0x1;
488   // LDRD/LDRH/LDRSB/LDRSH (literal) u = bit23
489   uint32_t opcode = 0x00800000;
490   if (val >> 63) {
491     opcode = 0x0;
492     val = -val;
493   }
494   uint32_t imm = getRemAndLZForGroup(group, val).first;
495   checkUInt(loc, imm, 8, rel);
496   write32le(loc, (read32le(loc) & 0xff7ff0f0) | opcode | ((imm & 0xf0) << 4) |
497                      (imm & 0xf));
498 }
499 
500 void ARM::relocate(uint8_t *loc, const Relocation &rel, uint64_t val) const {
501   switch (rel.type) {
502   case R_ARM_ABS32:
503   case R_ARM_BASE_PREL:
504   case R_ARM_GOTOFF32:
505   case R_ARM_GOT_BREL:
506   case R_ARM_GOT_PREL:
507   case R_ARM_REL32:
508   case R_ARM_RELATIVE:
509   case R_ARM_SBREL32:
510   case R_ARM_TARGET1:
511   case R_ARM_TARGET2:
512   case R_ARM_TLS_GD32:
513   case R_ARM_TLS_IE32:
514   case R_ARM_TLS_LDM32:
515   case R_ARM_TLS_LDO32:
516   case R_ARM_TLS_LE32:
517   case R_ARM_TLS_TPOFF32:
518   case R_ARM_TLS_DTPOFF32:
519     write32le(loc, val);
520     break;
521   case R_ARM_PREL31:
522     checkInt(loc, val, 31, rel);
523     write32le(loc, (read32le(loc) & 0x80000000) | (val & ~0x80000000));
524     break;
525   case R_ARM_CALL: {
526     // R_ARM_CALL is used for BL and BLX instructions, for symbols of type
527     // STT_FUNC we choose whether to write a BL or BLX depending on the
528     // value of bit 0 of Val. With bit 0 == 1 denoting Thumb. If the symbol is
529     // not of type STT_FUNC then we must preserve the original instruction.
530     // PLT entries are always ARM state so we know we don't need to interwork.
531     assert(rel.sym); // R_ARM_CALL is always reached via relocate().
532     bool bit0Thumb = val & 1;
533     bool isBlx = (read32le(loc) & 0xfe000000) == 0xfa000000;
534     // lld 10.0 and before always used bit0Thumb when deciding to write a BLX
535     // even when type not STT_FUNC.
536     if (!rel.sym->isFunc() && isBlx != bit0Thumb)
537       stateChangeWarning(loc, rel.type, *rel.sym);
538     if (rel.sym->isFunc() ? bit0Thumb : isBlx) {
539       // The BLX encoding is 0xfa:H:imm24 where Val = imm24:H:'1'
540       checkInt(loc, val, 26, rel);
541       write32le(loc, 0xfa000000 |                    // opcode
542                          ((val & 2) << 23) |         // H
543                          ((val >> 2) & 0x00ffffff)); // imm24
544       break;
545     }
546     // BLX (always unconditional) instruction to an ARM Target, select an
547     // unconditional BL.
548     write32le(loc, 0xeb000000 | (read32le(loc) & 0x00ffffff));
549     // fall through as BL encoding is shared with B
550   }
551     LLVM_FALLTHROUGH;
552   case R_ARM_JUMP24:
553   case R_ARM_PC24:
554   case R_ARM_PLT32:
555     checkInt(loc, val, 26, rel);
556     write32le(loc, (read32le(loc) & ~0x00ffffff) | ((val >> 2) & 0x00ffffff));
557     break;
558   case R_ARM_THM_JUMP8:
559     // We do a 9 bit check because val is right-shifted by 1 bit.
560     checkInt(loc, val, 9, rel);
561     write16le(loc, (read32le(loc) & 0xff00) | ((val >> 1) & 0x00ff));
562     break;
563   case R_ARM_THM_JUMP11:
564     // We do a 12 bit check because val is right-shifted by 1 bit.
565     checkInt(loc, val, 12, rel);
566     write16le(loc, (read32le(loc) & 0xf800) | ((val >> 1) & 0x07ff));
567     break;
568   case R_ARM_THM_JUMP19:
569     // Encoding T3: Val = S:J2:J1:imm6:imm11:0
570     checkInt(loc, val, 21, rel);
571     write16le(loc,
572               (read16le(loc) & 0xfbc0) |   // opcode cond
573                   ((val >> 10) & 0x0400) | // S
574                   ((val >> 12) & 0x003f)); // imm6
575     write16le(loc + 2,
576               0x8000 |                    // opcode
577                   ((val >> 8) & 0x0800) | // J2
578                   ((val >> 5) & 0x2000) | // J1
579                   ((val >> 1) & 0x07ff)); // imm11
580     break;
581   case R_ARM_THM_CALL: {
582     // R_ARM_THM_CALL is used for BL and BLX instructions, for symbols of type
583     // STT_FUNC we choose whether to write a BL or BLX depending on the
584     // value of bit 0 of Val. With bit 0 == 0 denoting ARM, if the symbol is
585     // not of type STT_FUNC then we must preserve the original instruction.
586     // PLT entries are always ARM state so we know we need to interwork.
587     assert(rel.sym); // R_ARM_THM_CALL is always reached via relocate().
588     bool bit0Thumb = val & 1;
589     bool isBlx = (read16le(loc + 2) & 0x1000) == 0;
590     // lld 10.0 and before always used bit0Thumb when deciding to write a BLX
591     // even when type not STT_FUNC. PLT entries generated by LLD are always ARM.
592     if (!rel.sym->isFunc() && !rel.sym->isInPlt() && isBlx == bit0Thumb)
593       stateChangeWarning(loc, rel.type, *rel.sym);
594     if (rel.sym->isFunc() || rel.sym->isInPlt() ? !bit0Thumb : isBlx) {
595       // We are writing a BLX. Ensure BLX destination is 4-byte aligned. As
596       // the BLX instruction may only be two byte aligned. This must be done
597       // before overflow check.
598       val = alignTo(val, 4);
599       write16le(loc + 2, read16le(loc + 2) & ~0x1000);
600     } else {
601       write16le(loc + 2, (read16le(loc + 2) & ~0x1000) | 1 << 12);
602     }
603     if (!config->armJ1J2BranchEncoding) {
604       // Older Arm architectures do not support R_ARM_THM_JUMP24 and have
605       // different encoding rules and range due to J1 and J2 always being 1.
606       checkInt(loc, val, 23, rel);
607       write16le(loc,
608                 0xf000 |                     // opcode
609                     ((val >> 12) & 0x07ff)); // imm11
610       write16le(loc + 2,
611                 (read16le(loc + 2) & 0xd000) | // opcode
612                     0x2800 |                   // J1 == J2 == 1
613                     ((val >> 1) & 0x07ff));    // imm11
614       break;
615     }
616   }
617     // Fall through as rest of encoding is the same as B.W
618     LLVM_FALLTHROUGH;
619   case R_ARM_THM_JUMP24:
620     // Encoding B  T4, BL T1, BLX T2: Val = S:I1:I2:imm10:imm11:0
621     checkInt(loc, val, 25, rel);
622     write16le(loc,
623               0xf000 |                     // opcode
624                   ((val >> 14) & 0x0400) | // S
625                   ((val >> 12) & 0x03ff)); // imm10
626     write16le(loc + 2,
627               (read16le(loc + 2) & 0xd000) |                  // opcode
628                   (((~(val >> 10)) ^ (val >> 11)) & 0x2000) | // J1
629                   (((~(val >> 11)) ^ (val >> 13)) & 0x0800) | // J2
630                   ((val >> 1) & 0x07ff));                     // imm11
631     break;
632   case R_ARM_MOVW_ABS_NC:
633   case R_ARM_MOVW_PREL_NC:
634   case R_ARM_MOVW_BREL_NC:
635     write32le(loc, (read32le(loc) & ~0x000f0fff) | ((val & 0xf000) << 4) |
636                        (val & 0x0fff));
637     break;
638   case R_ARM_MOVT_ABS:
639   case R_ARM_MOVT_PREL:
640   case R_ARM_MOVT_BREL:
641     write32le(loc, (read32le(loc) & ~0x000f0fff) |
642                        (((val >> 16) & 0xf000) << 4) | ((val >> 16) & 0xfff));
643     break;
644   case R_ARM_THM_MOVT_ABS:
645   case R_ARM_THM_MOVT_PREL:
646   case R_ARM_THM_MOVT_BREL:
647     // Encoding T1: A = imm4:i:imm3:imm8
648     write16le(loc,
649               0xf2c0 |                     // opcode
650                   ((val >> 17) & 0x0400) | // i
651                   ((val >> 28) & 0x000f)); // imm4
652     write16le(loc + 2,
653               (read16le(loc + 2) & 0x8f00) | // opcode
654                   ((val >> 12) & 0x7000) |   // imm3
655                   ((val >> 16) & 0x00ff));   // imm8
656     break;
657   case R_ARM_THM_MOVW_ABS_NC:
658   case R_ARM_THM_MOVW_PREL_NC:
659   case R_ARM_THM_MOVW_BREL_NC:
660     // Encoding T3: A = imm4:i:imm3:imm8
661     write16le(loc,
662               0xf240 |                     // opcode
663                   ((val >> 1) & 0x0400) |  // i
664                   ((val >> 12) & 0x000f)); // imm4
665     write16le(loc + 2,
666               (read16le(loc + 2) & 0x8f00) | // opcode
667                   ((val << 4) & 0x7000) |    // imm3
668                   (val & 0x00ff));           // imm8
669     break;
670   case R_ARM_ALU_PC_G0:
671     encodeAluGroup(loc, rel, val, 0, true);
672     break;
673   case R_ARM_ALU_PC_G0_NC:
674     encodeAluGroup(loc, rel, val, 0, false);
675     break;
676   case R_ARM_ALU_PC_G1:
677     encodeAluGroup(loc, rel, val, 1, true);
678     break;
679   case R_ARM_ALU_PC_G1_NC:
680     encodeAluGroup(loc, rel, val, 1, false);
681     break;
682   case R_ARM_ALU_PC_G2:
683     encodeAluGroup(loc, rel, val, 2, true);
684     break;
685   case R_ARM_LDR_PC_G0:
686     encodeLdrGroup(loc, rel, val, 0);
687     break;
688   case R_ARM_LDR_PC_G1:
689     encodeLdrGroup(loc, rel, val, 1);
690     break;
691   case R_ARM_LDR_PC_G2:
692     encodeLdrGroup(loc, rel, val, 2);
693     break;
694   case R_ARM_LDRS_PC_G0:
695     encodeLdrsGroup(loc, rel, val, 0);
696     break;
697   case R_ARM_LDRS_PC_G1:
698     encodeLdrsGroup(loc, rel, val, 1);
699     break;
700   case R_ARM_LDRS_PC_G2:
701     encodeLdrsGroup(loc, rel, val, 2);
702     break;
703   case R_ARM_THM_ALU_PREL_11_0: {
704     // ADR encoding T2 (sub), T3 (add) i:imm3:imm8
705     int64_t imm = val;
706     uint16_t sub = 0;
707     if (imm < 0) {
708       imm = -imm;
709       sub = 0x00a0;
710     }
711     checkUInt(loc, imm, 12, rel);
712     write16le(loc, (read16le(loc) & 0xfb0f) | sub | (imm & 0x800) >> 1);
713     write16le(loc + 2,
714               (read16le(loc + 2) & 0x8f00) | (imm & 0x700) << 4 | (imm & 0xff));
715     break;
716   }
717   case R_ARM_THM_PC8:
718     // ADR and LDR literal encoding T1 positive offset only imm8:00
719     // R_ARM_THM_PC8 is S + A - Pa, we have ((S + A) | T) - Pa, if S is a
720     // function then addr is 0 (modulo 2) and Pa is 0 (modulo 4) so we can clear
721     // bottom bit to recover S + A - Pa.
722     if (rel.sym->isFunc())
723       val &= ~0x1;
724     checkUInt(loc, val, 10, rel);
725     checkAlignment(loc, val, 4, rel);
726     write16le(loc, (read16le(loc) & 0xff00) | (val & 0x3fc) >> 2);
727     break;
728   case R_ARM_THM_PC12: {
729     // LDR (literal) encoding T2, add = (U == '1') imm12
730     // imm12 is unsigned
731     // R_ARM_THM_PC12 is S + A - Pa, we have ((S + A) | T) - Pa, if S is a
732     // function then addr is 0 (modulo 2) and Pa is 0 (modulo 4) so we can clear
733     // bottom bit to recover S + A - Pa.
734     if (rel.sym->isFunc())
735       val &= ~0x1;
736     int64_t imm12 = val;
737     uint16_t u = 0x0080;
738     if (imm12 < 0) {
739       imm12 = -imm12;
740       u = 0;
741     }
742     checkUInt(loc, imm12, 12, rel);
743     write16le(loc, read16le(loc) | u);
744     write16le(loc + 2, (read16le(loc + 2) & 0xf000) | imm12);
745     break;
746   }
747   default:
748     llvm_unreachable("unknown relocation");
749   }
750 }
751 
752 int64_t ARM::getImplicitAddend(const uint8_t *buf, RelType type) const {
753   switch (type) {
754   default:
755     internalLinkerError(getErrorLocation(buf),
756                         "cannot read addend for relocation " + toString(type));
757     return 0;
758   case R_ARM_ABS32:
759   case R_ARM_BASE_PREL:
760   case R_ARM_GLOB_DAT:
761   case R_ARM_GOTOFF32:
762   case R_ARM_GOT_BREL:
763   case R_ARM_GOT_PREL:
764   case R_ARM_IRELATIVE:
765   case R_ARM_REL32:
766   case R_ARM_RELATIVE:
767   case R_ARM_SBREL32:
768   case R_ARM_TARGET1:
769   case R_ARM_TARGET2:
770   case R_ARM_TLS_DTPMOD32:
771   case R_ARM_TLS_DTPOFF32:
772   case R_ARM_TLS_GD32:
773   case R_ARM_TLS_IE32:
774   case R_ARM_TLS_LDM32:
775   case R_ARM_TLS_LE32:
776   case R_ARM_TLS_LDO32:
777   case R_ARM_TLS_TPOFF32:
778     return SignExtend64<32>(read32le(buf));
779   case R_ARM_PREL31:
780     return SignExtend64<31>(read32le(buf));
781   case R_ARM_CALL:
782   case R_ARM_JUMP24:
783   case R_ARM_PC24:
784   case R_ARM_PLT32:
785     return SignExtend64<26>(read32le(buf) << 2);
786   case R_ARM_THM_JUMP8:
787     return SignExtend64<9>(read16le(buf) << 1);
788   case R_ARM_THM_JUMP11:
789     return SignExtend64<12>(read16le(buf) << 1);
790   case R_ARM_THM_JUMP19: {
791     // Encoding T3: A = S:J2:J1:imm10:imm6:0
792     uint16_t hi = read16le(buf);
793     uint16_t lo = read16le(buf + 2);
794     return SignExtend64<20>(((hi & 0x0400) << 10) | // S
795                             ((lo & 0x0800) << 8) |  // J2
796                             ((lo & 0x2000) << 5) |  // J1
797                             ((hi & 0x003f) << 12) | // imm6
798                             ((lo & 0x07ff) << 1));  // imm11:0
799   }
800   case R_ARM_THM_CALL:
801     if (!config->armJ1J2BranchEncoding) {
802       // Older Arm architectures do not support R_ARM_THM_JUMP24 and have
803       // different encoding rules and range due to J1 and J2 always being 1.
804       uint16_t hi = read16le(buf);
805       uint16_t lo = read16le(buf + 2);
806       return SignExtend64<22>(((hi & 0x7ff) << 12) | // imm11
807                               ((lo & 0x7ff) << 1));  // imm11:0
808       break;
809     }
810     LLVM_FALLTHROUGH;
811   case R_ARM_THM_JUMP24: {
812     // Encoding B T4, BL T1, BLX T2: A = S:I1:I2:imm10:imm11:0
813     // I1 = NOT(J1 EOR S), I2 = NOT(J2 EOR S)
814     uint16_t hi = read16le(buf);
815     uint16_t lo = read16le(buf + 2);
816     return SignExtend64<24>(((hi & 0x0400) << 14) |                    // S
817                             (~((lo ^ (hi << 3)) << 10) & 0x00800000) | // I1
818                             (~((lo ^ (hi << 1)) << 11) & 0x00400000) | // I2
819                             ((hi & 0x003ff) << 12) |                   // imm0
820                             ((lo & 0x007ff) << 1)); // imm11:0
821   }
822   // ELF for the ARM Architecture 4.6.1.1 the implicit addend for MOVW and
823   // MOVT is in the range -32768 <= A < 32768
824   case R_ARM_MOVW_ABS_NC:
825   case R_ARM_MOVT_ABS:
826   case R_ARM_MOVW_PREL_NC:
827   case R_ARM_MOVT_PREL:
828   case R_ARM_MOVW_BREL_NC:
829   case R_ARM_MOVT_BREL: {
830     uint64_t val = read32le(buf) & 0x000f0fff;
831     return SignExtend64<16>(((val & 0x000f0000) >> 4) | (val & 0x00fff));
832   }
833   case R_ARM_THM_MOVW_ABS_NC:
834   case R_ARM_THM_MOVT_ABS:
835   case R_ARM_THM_MOVW_PREL_NC:
836   case R_ARM_THM_MOVT_PREL:
837   case R_ARM_THM_MOVW_BREL_NC:
838   case R_ARM_THM_MOVT_BREL: {
839     // Encoding T3: A = imm4:i:imm3:imm8
840     uint16_t hi = read16le(buf);
841     uint16_t lo = read16le(buf + 2);
842     return SignExtend64<16>(((hi & 0x000f) << 12) | // imm4
843                             ((hi & 0x0400) << 1) |  // i
844                             ((lo & 0x7000) >> 4) |  // imm3
845                             (lo & 0x00ff));         // imm8
846   }
847   case R_ARM_ALU_PC_G0:
848   case R_ARM_ALU_PC_G0_NC:
849   case R_ARM_ALU_PC_G1:
850   case R_ARM_ALU_PC_G1_NC:
851   case R_ARM_ALU_PC_G2: {
852     // 12-bit immediate is a modified immediate made up of a 4-bit even
853     // right rotation and 8-bit constant. After the rotation the value
854     // is zero-extended. When bit 23 is set the instruction is an add, when
855     // bit 22 is set it is a sub.
856     uint32_t instr = read32le(buf);
857     uint32_t val = rotr32(instr & 0xff, ((instr & 0xf00) >> 8) * 2);
858     return (instr & 0x00400000) ? -val : val;
859   }
860   case R_ARM_LDR_PC_G0:
861   case R_ARM_LDR_PC_G1:
862   case R_ARM_LDR_PC_G2: {
863     // ADR (literal) add = bit23, sub = bit22
864     // LDR (literal) u = bit23 unsigned imm12
865     bool u = read32le(buf) & 0x00800000;
866     uint32_t imm12 = read32le(buf) & 0xfff;
867     return u ? imm12 : -imm12;
868   }
869   case R_ARM_LDRS_PC_G0:
870   case R_ARM_LDRS_PC_G1:
871   case R_ARM_LDRS_PC_G2: {
872     // LDRD/LDRH/LDRSB/LDRSH (literal) u = bit23 unsigned imm8
873     uint32_t opcode = read32le(buf);
874     bool u = opcode & 0x00800000;
875     uint32_t imm4l = opcode & 0xf;
876     uint32_t imm4h = (opcode & 0xf00) >> 4;
877     return u ? (imm4h | imm4l) : -(imm4h | imm4l);
878   }
879   case R_ARM_THM_ALU_PREL_11_0: {
880     // Thumb2 ADR, which is an alias for a sub or add instruction with an
881     // unsigned immediate.
882     // ADR encoding T2 (sub), T3 (add) i:imm3:imm8
883     uint16_t hi = read16le(buf);
884     uint16_t lo = read16le(buf + 2);
885     uint64_t imm = (hi & 0x0400) << 1 | // i
886                    (lo & 0x7000) >> 4 | // imm3
887                    (lo & 0x00ff);       // imm8
888     // For sub, addend is negative, add is positive.
889     return (hi & 0x00f0) ? -imm : imm;
890   }
891   case R_ARM_THM_PC8:
892     // ADR and LDR (literal) encoding T1
893     // From ELF for the ARM Architecture the initial signed addend is formed
894     // from an unsigned field using expression (((imm8:00 + 4) & 0x3ff) – 4)
895     // this trick permits the PC bias of -4 to be encoded using imm8 = 0xff
896     return ((((read16le(buf) & 0xff) << 2) + 4) & 0x3ff) - 4;
897   case R_ARM_THM_PC12: {
898     // LDR (literal) encoding T2, add = (U == '1') imm12
899     bool u = read16le(buf) & 0x0080;
900     uint64_t imm12 = read16le(buf + 2) & 0x0fff;
901     return u ? imm12 : -imm12;
902   }
903   case R_ARM_NONE:
904   case R_ARM_V4BX:
905   case R_ARM_JUMP_SLOT:
906     // These relocations are defined as not having an implicit addend.
907     return 0;
908   }
909 }
910 
911 TargetInfo *elf::getARMTargetInfo() {
912   static ARM target;
913   return &target;
914 }
915