| 1 | /* Machine-dependent ELF dynamic relocation functions. PowerPC version. |
| 2 | Copyright (C) 1995-2024 Free Software Foundation, Inc. |
| 3 | This file is part of the GNU C Library. |
| 4 | |
| 5 | The GNU C Library is free software; you can redistribute it and/or |
| 6 | modify it under the terms of the GNU Lesser General Public |
| 7 | License as published by the Free Software Foundation; either |
| 8 | version 2.1 of the License, or (at your option) any later version. |
| 9 | |
| 10 | The GNU C Library is distributed in the hope that it will be useful, |
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | Lesser General Public License for more details. |
| 14 | |
| 15 | You should have received a copy of the GNU Lesser General Public |
| 16 | License along with the GNU C Library; if not, see |
| 17 | <https://www.gnu.org/licenses/>. */ |
| 18 | |
| 19 | #include <unistd.h> |
| 20 | #include <string.h> |
| 21 | #include <sys/param.h> |
| 22 | #include <link.h> |
| 23 | #include <ldsodefs.h> |
| 24 | #include <elf/dynamic-link.h> |
| 25 | #include <dl-machine.h> |
| 26 | #include <_itoa.h> |
| 27 | |
| 28 | /* Stuff for the PLT. */ |
| 29 | #define PLT_INITIAL_ENTRY_WORDS 18 |
| 30 | #define PLT_LONGBRANCH_ENTRY_WORDS 0 |
| 31 | #define PLT_TRAMPOLINE_ENTRY_WORDS 6 |
| 32 | #define PLT_DOUBLE_SIZE (1<<13) |
| 33 | #define PLT_ENTRY_START_WORDS(entry_number) \ |
| 34 | (PLT_INITIAL_ENTRY_WORDS + (entry_number)*2 \ |
| 35 | + ((entry_number) > PLT_DOUBLE_SIZE \ |
| 36 | ? ((entry_number) - PLT_DOUBLE_SIZE)*2 \ |
| 37 | : 0)) |
| 38 | #define PLT_DATA_START_WORDS(num_entries) PLT_ENTRY_START_WORDS(num_entries) |
| 39 | |
| 40 | /* Macros to build PowerPC opcode words. */ |
| 41 | #define OPCODE_ADDI(rd,ra,simm) \ |
| 42 | (0x38000000 | (rd) << 21 | (ra) << 16 | ((simm) & 0xffff)) |
| 43 | #define OPCODE_ADDIS(rd,ra,simm) \ |
| 44 | (0x3c000000 | (rd) << 21 | (ra) << 16 | ((simm) & 0xffff)) |
| 45 | #define OPCODE_ADD(rd,ra,rb) \ |
| 46 | (0x7c000214 | (rd) << 21 | (ra) << 16 | (rb) << 11) |
| 47 | #define OPCODE_B(target) (0x48000000 | ((target) & 0x03fffffc)) |
| 48 | #define OPCODE_BA(target) (0x48000002 | ((target) & 0x03fffffc)) |
| 49 | #define OPCODE_BCTR() 0x4e800420 |
| 50 | #define OPCODE_LWZ(rd,d,ra) \ |
| 51 | (0x80000000 | (rd) << 21 | (ra) << 16 | ((d) & 0xffff)) |
| 52 | #define OPCODE_LWZU(rd,d,ra) \ |
| 53 | (0x84000000 | (rd) << 21 | (ra) << 16 | ((d) & 0xffff)) |
| 54 | #define OPCODE_MTCTR(rd) (0x7C0903A6 | (rd) << 21) |
| 55 | #define OPCODE_RLWINM(ra,rs,sh,mb,me) \ |
| 56 | (0x54000000 | (rs) << 21 | (ra) << 16 | (sh) << 11 | (mb) << 6 | (me) << 1) |
| 57 | |
| 58 | #define OPCODE_LI(rd,simm) OPCODE_ADDI(rd,0,simm) |
| 59 | #define OPCODE_ADDIS_HI(rd,ra,value) \ |
| 60 | OPCODE_ADDIS(rd,ra,((value) + 0x8000) >> 16) |
| 61 | #define OPCODE_LIS_HI(rd,value) OPCODE_ADDIS_HI(rd,0,value) |
| 62 | #define OPCODE_SLWI(ra,rs,sh) OPCODE_RLWINM(ra,rs,sh,0,31-sh) |
| 63 | |
| 64 | |
| 65 | #define PPC_DCBST(where) asm volatile ("dcbst 0,%0" : : "r"(where) : "memory") |
| 66 | #define PPC_SYNC asm volatile ("sync" : : : "memory") |
| 67 | #define PPC_ISYNC asm volatile ("sync; isync" : : : "memory") |
| 68 | #define PPC_ICBI(where) asm volatile ("icbi 0,%0" : : "r"(where) : "memory") |
| 69 | #define PPC_DIE asm volatile ("tweq 0,0") |
| 70 | |
| 71 | /* Use this when you've modified some code, but it won't be in the |
| 72 | instruction fetch queue (or when it doesn't matter if it is). */ |
| 73 | #define MODIFIED_CODE_NOQUEUE(where) \ |
| 74 | do { PPC_DCBST(where); PPC_SYNC; PPC_ICBI(where); } while (0) |
| 75 | /* Use this when it might be in the instruction queue. */ |
| 76 | #define MODIFIED_CODE(where) \ |
| 77 | do { PPC_DCBST(where); PPC_SYNC; PPC_ICBI(where); PPC_ISYNC; } while (0) |
| 78 | |
| 79 | |
| 80 | /* The idea here is that to conform to the ABI, we are supposed to try |
| 81 | to load dynamic objects between 0x10000 (we actually use 0x40000 as |
| 82 | the lower bound, to increase the chance of a memory reference from |
| 83 | a null pointer giving a segfault) and the program's load address; |
| 84 | this may allow us to use a branch instruction in the PLT rather |
| 85 | than a computed jump. The address is only used as a preference for |
| 86 | mmap, so if we get it wrong the worst that happens is that it gets |
| 87 | mapped somewhere else. */ |
| 88 | |
| 89 | ElfW(Addr) |
| 90 | __elf_preferred_address (struct link_map *loader, size_t maplength, |
| 91 | ElfW(Addr) mapstartpref) |
| 92 | { |
| 93 | ElfW(Addr) low, high; |
| 94 | struct link_map *l; |
| 95 | Lmid_t nsid; |
| 96 | |
| 97 | /* If the object has a preference, load it there! */ |
| 98 | if (mapstartpref != 0) |
| 99 | return mapstartpref; |
| 100 | |
| 101 | /* Otherwise, quickly look for a suitable gap between 0x3FFFF and |
| 102 | 0x70000000. 0x3FFFF is so that references off NULL pointers will |
| 103 | cause a segfault, 0x70000000 is just paranoia (it should always |
| 104 | be superseded by the program's load address). */ |
| 105 | low = 0x0003FFFF; |
| 106 | high = 0x70000000; |
| 107 | for (nsid = 0; nsid < DL_NNS; ++nsid) |
| 108 | for (l = GL(dl_ns)[nsid]._ns_loaded; l; l = l->l_next) |
| 109 | { |
| 110 | ElfW(Addr) mapstart, mapend; |
| 111 | mapstart = l->l_map_start & ~(GLRO(dl_pagesize) - 1); |
| 112 | mapend = l->l_map_end | (GLRO(dl_pagesize) - 1); |
| 113 | assert (mapend > mapstart); |
| 114 | |
| 115 | /* Prefer gaps below the main executable, note that l == |
| 116 | _dl_loaded does not work for static binaries loading |
| 117 | e.g. libnss_*.so. */ |
| 118 | if ((mapend >= high || l->l_type == lt_executable) |
| 119 | && high >= mapstart) |
| 120 | high = mapstart; |
| 121 | else if (mapend >= low && low >= mapstart) |
| 122 | low = mapend; |
| 123 | else if (high >= mapend && mapstart >= low) |
| 124 | { |
| 125 | if (high - mapend >= mapstart - low) |
| 126 | low = mapend; |
| 127 | else |
| 128 | high = mapstart; |
| 129 | } |
| 130 | } |
| 131 | |
| 132 | high -= 0x10000; /* Allow some room between objects. */ |
| 133 | maplength = (maplength | (GLRO(dl_pagesize) - 1)) + 1; |
| 134 | if (high <= low || high - low < maplength ) |
| 135 | return 0; |
| 136 | return high - maplength; /* Both high and maplength are page-aligned. */ |
| 137 | } |
| 138 | |
| 139 | /* Set up the loaded object described by L so its unrelocated PLT |
| 140 | entries will jump to the on-demand fixup code in dl-runtime.c. |
| 141 | Also install a small trampoline to be used by entries that have |
| 142 | been relocated to an address too far away for a single branch. */ |
| 143 | |
| 144 | /* There are many kinds of PLT entries: |
| 145 | |
| 146 | (1) A direct jump to the actual routine, either a relative or |
| 147 | absolute branch. These are set up in __elf_machine_fixup_plt. |
| 148 | |
| 149 | (2) Short lazy entries. These cover the first 8192 slots in |
| 150 | the PLT, and look like (where 'index' goes from 0 to 8191): |
| 151 | |
| 152 | li %r11, index*4 |
| 153 | b &plt[PLT_TRAMPOLINE_ENTRY_WORDS+1] |
| 154 | |
| 155 | (3) Short indirect jumps. These replace (2) when a direct jump |
| 156 | wouldn't reach. They look the same except that the branch |
| 157 | is 'b &plt[PLT_LONGBRANCH_ENTRY_WORDS]'. |
| 158 | |
| 159 | (4) Long lazy entries. These cover the slots when a short entry |
| 160 | won't fit ('index*4' overflows its field), and look like: |
| 161 | |
| 162 | lis %r11, %hi(index*4 + &plt[PLT_DATA_START_WORDS]) |
| 163 | lwzu %r12, %r11, %lo(index*4 + &plt[PLT_DATA_START_WORDS]) |
| 164 | b &plt[PLT_TRAMPOLINE_ENTRY_WORDS] |
| 165 | bctr |
| 166 | |
| 167 | (5) Long indirect jumps. These replace (4) when a direct jump |
| 168 | wouldn't reach. They look like: |
| 169 | |
| 170 | lis %r11, %hi(index*4 + &plt[PLT_DATA_START_WORDS]) |
| 171 | lwz %r12, %r11, %lo(index*4 + &plt[PLT_DATA_START_WORDS]) |
| 172 | mtctr %r12 |
| 173 | bctr |
| 174 | |
| 175 | (6) Long direct jumps. These are used when thread-safety is not |
| 176 | required. They look like: |
| 177 | |
| 178 | lis %r12, %hi(finaladdr) |
| 179 | addi %r12, %r12, %lo(finaladdr) |
| 180 | mtctr %r12 |
| 181 | bctr |
| 182 | |
| 183 | |
| 184 | The lazy entries, (2) and (4), are set up here in |
| 185 | __elf_machine_runtime_setup. (1), (3), and (5) are set up in |
| 186 | __elf_machine_fixup_plt. (1), (3), and (6) can also be constructed |
| 187 | in __process_machine_rela. |
| 188 | |
| 189 | The reason for the somewhat strange construction of the long |
| 190 | entries, (4) and (5), is that we need to ensure thread-safety. For |
| 191 | (1) and (3), this is obvious because only one instruction is |
| 192 | changed and the PPC architecture guarantees that aligned stores are |
| 193 | atomic. For (5), this is more tricky. When changing (4) to (5), |
| 194 | the `b' instruction is first changed to `mtctr'; this is safe |
| 195 | and is why the `lwzu' instruction is not just a simple `addi'. |
| 196 | Once this is done, and is visible to all processors, the `lwzu' can |
| 197 | safely be changed to a `lwz'. */ |
| 198 | int |
| 199 | __elf_machine_runtime_setup (struct link_map *map, int lazy, int profile) |
| 200 | { |
| 201 | if (map->l_info[DT_JMPREL]) |
| 202 | { |
| 203 | Elf32_Word i; |
| 204 | Elf32_Word *plt = (Elf32_Word *) D_PTR (map, l_info[DT_PLTGOT]); |
| 205 | Elf32_Word num_plt_entries = (map->l_info[DT_PLTRELSZ]->d_un.d_val |
| 206 | / sizeof (Elf32_Rela)); |
| 207 | Elf32_Word rel_offset_words = PLT_DATA_START_WORDS (num_plt_entries); |
| 208 | Elf32_Word data_words = (Elf32_Word) (plt + rel_offset_words); |
| 209 | Elf32_Word size_modified; |
| 210 | |
| 211 | extern void _dl_runtime_resolve (void); |
| 212 | extern void _dl_prof_resolve (void); |
| 213 | |
| 214 | /* Convert the index in r11 into an actual address, and get the |
| 215 | word at that address. */ |
| 216 | plt[PLT_LONGBRANCH_ENTRY_WORDS] = OPCODE_ADDIS_HI (11, 11, data_words); |
| 217 | plt[PLT_LONGBRANCH_ENTRY_WORDS + 1] = OPCODE_LWZ (11, data_words, 11); |
| 218 | |
| 219 | /* Call the procedure at that address. */ |
| 220 | plt[PLT_LONGBRANCH_ENTRY_WORDS + 2] = OPCODE_MTCTR (11); |
| 221 | plt[PLT_LONGBRANCH_ENTRY_WORDS + 3] = OPCODE_BCTR (); |
| 222 | |
| 223 | if (lazy) |
| 224 | { |
| 225 | Elf32_Word *tramp = plt + PLT_TRAMPOLINE_ENTRY_WORDS; |
| 226 | Elf32_Word dlrr; |
| 227 | Elf32_Word offset; |
| 228 | |
| 229 | #if !defined PROF && defined SHARED |
| 230 | dlrr = (Elf32_Word) (profile |
| 231 | ? _dl_prof_resolve |
| 232 | : _dl_runtime_resolve); |
| 233 | if (profile && GLRO(dl_profile) != NULL |
| 234 | && _dl_name_match_p (GLRO(dl_profile), map)) |
| 235 | /* This is the object we are looking for. Say that we really |
| 236 | want profiling and the timers are started. */ |
| 237 | GL(dl_profile_map) = map; |
| 238 | #else |
| 239 | dlrr = (Elf32_Word) _dl_runtime_resolve; |
| 240 | #endif |
| 241 | |
| 242 | /* For the long entries, subtract off data_words. */ |
| 243 | tramp[0] = OPCODE_ADDIS_HI (11, 11, -data_words); |
| 244 | tramp[1] = OPCODE_ADDI (11, 11, -data_words); |
| 245 | |
| 246 | /* Multiply index of entry by 3 (in r11). */ |
| 247 | tramp[2] = OPCODE_SLWI (12, 11, 1); |
| 248 | tramp[3] = OPCODE_ADD (11, 12, 11); |
| 249 | if (dlrr <= 0x01fffffc || dlrr >= 0xfe000000) |
| 250 | { |
| 251 | /* Load address of link map in r12. */ |
| 252 | tramp[4] = OPCODE_LI (12, (Elf32_Word) map); |
| 253 | tramp[5] = OPCODE_ADDIS_HI (12, 12, (Elf32_Word) map); |
| 254 | |
| 255 | /* Call _dl_runtime_resolve. */ |
| 256 | tramp[6] = OPCODE_BA (dlrr); |
| 257 | } |
| 258 | else |
| 259 | { |
| 260 | /* Get address of _dl_runtime_resolve in CTR. */ |
| 261 | tramp[4] = OPCODE_LI (12, dlrr); |
| 262 | tramp[5] = OPCODE_ADDIS_HI (12, 12, dlrr); |
| 263 | tramp[6] = OPCODE_MTCTR (12); |
| 264 | |
| 265 | /* Load address of link map in r12. */ |
| 266 | tramp[7] = OPCODE_LI (12, (Elf32_Word) map); |
| 267 | tramp[8] = OPCODE_ADDIS_HI (12, 12, (Elf32_Word) map); |
| 268 | |
| 269 | /* Call _dl_runtime_resolve. */ |
| 270 | tramp[9] = OPCODE_BCTR (); |
| 271 | } |
| 272 | |
| 273 | /* Set up the lazy PLT entries. */ |
| 274 | offset = PLT_INITIAL_ENTRY_WORDS; |
| 275 | i = 0; |
| 276 | while (i < num_plt_entries && i < PLT_DOUBLE_SIZE) |
| 277 | { |
| 278 | plt[offset ] = OPCODE_LI (11, i * 4); |
| 279 | plt[offset+1] = OPCODE_B ((PLT_TRAMPOLINE_ENTRY_WORDS + 2 |
| 280 | - (offset+1)) |
| 281 | * 4); |
| 282 | i++; |
| 283 | offset += 2; |
| 284 | } |
| 285 | while (i < num_plt_entries) |
| 286 | { |
| 287 | plt[offset ] = OPCODE_LIS_HI (11, i * 4 + data_words); |
| 288 | plt[offset+1] = OPCODE_LWZU (12, i * 4 + data_words, 11); |
| 289 | plt[offset+2] = OPCODE_B ((PLT_TRAMPOLINE_ENTRY_WORDS |
| 290 | - (offset+2)) |
| 291 | * 4); |
| 292 | plt[offset+3] = OPCODE_BCTR (); |
| 293 | i++; |
| 294 | offset += 4; |
| 295 | } |
| 296 | } |
| 297 | |
| 298 | /* Now, we've modified code. We need to write the changes from |
| 299 | the data cache to a second-level unified cache, then make |
| 300 | sure that stale data in the instruction cache is removed. |
| 301 | (In a multiprocessor system, the effect is more complex.) |
| 302 | Most of the PLT shouldn't be in the instruction cache, but |
| 303 | there may be a little overlap at the start and the end. |
| 304 | |
| 305 | Assumes that dcbst and icbi apply to lines of 16 bytes or |
| 306 | more. Current known line sizes are 16, 32, and 128 bytes. |
| 307 | The following gets the cache line size, when available. */ |
| 308 | |
| 309 | /* Default minimum 4 words per cache line. */ |
| 310 | int line_size_words = 4; |
| 311 | |
| 312 | if (lazy && GLRO(dl_cache_line_size) != 0) |
| 313 | /* Convert bytes to words. */ |
| 314 | line_size_words = GLRO(dl_cache_line_size) / 4; |
| 315 | |
| 316 | size_modified = lazy ? rel_offset_words : 6; |
| 317 | for (i = 0; i < size_modified; i += line_size_words) |
| 318 | PPC_DCBST (plt + i); |
| 319 | PPC_DCBST (plt + size_modified - 1); |
| 320 | PPC_SYNC; |
| 321 | |
| 322 | for (i = 0; i < size_modified; i += line_size_words) |
| 323 | PPC_ICBI (plt + i); |
| 324 | PPC_ICBI (plt + size_modified - 1); |
| 325 | PPC_ISYNC; |
| 326 | } |
| 327 | |
| 328 | return lazy; |
| 329 | } |
| 330 | |
| 331 | Elf32_Addr |
| 332 | __elf_machine_fixup_plt (struct link_map *map, |
| 333 | Elf32_Addr *reloc_addr, Elf32_Addr finaladdr) |
| 334 | { |
| 335 | Elf32_Sword delta = finaladdr - (Elf32_Word) reloc_addr; |
| 336 | if (delta << 6 >> 6 == delta) |
| 337 | *reloc_addr = OPCODE_B (delta); |
| 338 | else if (finaladdr <= 0x01fffffc || finaladdr >= 0xfe000000) |
| 339 | *reloc_addr = OPCODE_BA (finaladdr); |
| 340 | else |
| 341 | { |
| 342 | Elf32_Word *plt, *data_words; |
| 343 | Elf32_Word index, offset, num_plt_entries; |
| 344 | |
| 345 | num_plt_entries = (map->l_info[DT_PLTRELSZ]->d_un.d_val |
| 346 | / sizeof (Elf32_Rela)); |
| 347 | plt = (Elf32_Word *) D_PTR (map, l_info[DT_PLTGOT]); |
| 348 | offset = reloc_addr - plt; |
| 349 | index = (offset - PLT_INITIAL_ENTRY_WORDS)/2; |
| 350 | data_words = plt + PLT_DATA_START_WORDS (num_plt_entries); |
| 351 | |
| 352 | reloc_addr += 1; |
| 353 | |
| 354 | if (index < PLT_DOUBLE_SIZE) |
| 355 | { |
| 356 | data_words[index] = finaladdr; |
| 357 | PPC_SYNC; |
| 358 | *reloc_addr = OPCODE_B ((PLT_LONGBRANCH_ENTRY_WORDS - (offset+1)) |
| 359 | * 4); |
| 360 | } |
| 361 | else |
| 362 | { |
| 363 | index -= (index - PLT_DOUBLE_SIZE)/2; |
| 364 | |
| 365 | data_words[index] = finaladdr; |
| 366 | PPC_SYNC; |
| 367 | |
| 368 | reloc_addr[1] = OPCODE_MTCTR (12); |
| 369 | MODIFIED_CODE_NOQUEUE (reloc_addr + 1); |
| 370 | PPC_SYNC; |
| 371 | |
| 372 | reloc_addr[0] = OPCODE_LWZ (12, |
| 373 | (Elf32_Word) (data_words + index), 11); |
| 374 | } |
| 375 | } |
| 376 | MODIFIED_CODE (reloc_addr); |
| 377 | return finaladdr; |
| 378 | } |
| 379 | |
| 380 | void |
| 381 | _dl_reloc_overflow (struct link_map *map, |
| 382 | const char *name, |
| 383 | Elf32_Addr *const reloc_addr, |
| 384 | const Elf32_Sym *refsym) |
| 385 | { |
| 386 | char buffer[128]; |
| 387 | char *t; |
| 388 | t = stpcpy (buffer, name); |
| 389 | t = stpcpy (t, " relocation at 0x00000000" ); |
| 390 | _itoa_word (value: (unsigned) reloc_addr, buflim: t, base: 16, upper_case: 0); |
| 391 | if (refsym) |
| 392 | { |
| 393 | const char *strtab; |
| 394 | |
| 395 | strtab = (const void *) D_PTR (map, l_info[DT_STRTAB]); |
| 396 | t = stpcpy (t, " for symbol `" ); |
| 397 | t = stpcpy (t, strtab + refsym->st_name); |
| 398 | t = stpcpy (t, "'" ); |
| 399 | } |
| 400 | t = stpcpy (t, " out of range" ); |
| 401 | _dl_signal_error (errcode: 0, object: map->l_name, NULL, errstring: buffer); |
| 402 | } |
| 403 | |
| 404 | void |
| 405 | __process_machine_rela (struct link_map *map, |
| 406 | const Elf32_Rela *reloc, |
| 407 | struct link_map *sym_map, |
| 408 | const Elf32_Sym *sym, |
| 409 | const Elf32_Sym *refsym, |
| 410 | Elf32_Addr *const reloc_addr, |
| 411 | Elf32_Addr const finaladdr, |
| 412 | int rinfo, bool skip_ifunc) |
| 413 | { |
| 414 | union unaligned |
| 415 | { |
| 416 | uint16_t u2; |
| 417 | uint32_t u4; |
| 418 | } __attribute__((__packed__)); |
| 419 | |
| 420 | switch (rinfo) |
| 421 | { |
| 422 | case R_PPC_NONE: |
| 423 | return; |
| 424 | |
| 425 | case R_PPC_ADDR32: |
| 426 | case R_PPC_GLOB_DAT: |
| 427 | case R_PPC_RELATIVE: |
| 428 | *reloc_addr = finaladdr; |
| 429 | return; |
| 430 | |
| 431 | case R_PPC_IRELATIVE: |
| 432 | if (__glibc_likely (!skip_ifunc)) |
| 433 | *reloc_addr = ((Elf32_Addr (*) (void)) finaladdr) (); |
| 434 | return; |
| 435 | |
| 436 | case R_PPC_UADDR32: |
| 437 | ((union unaligned *) reloc_addr)->u4 = finaladdr; |
| 438 | break; |
| 439 | |
| 440 | case R_PPC_ADDR24: |
| 441 | if (__glibc_unlikely (finaladdr > 0x01fffffc && finaladdr < 0xfe000000)) |
| 442 | _dl_reloc_overflow (map, name: "R_PPC_ADDR24" , reloc_addr, refsym); |
| 443 | *reloc_addr = (*reloc_addr & 0xfc000003) | (finaladdr & 0x3fffffc); |
| 444 | break; |
| 445 | |
| 446 | case R_PPC_ADDR16: |
| 447 | if (__glibc_unlikely (finaladdr > 0x7fff && finaladdr < 0xffff8000)) |
| 448 | _dl_reloc_overflow (map, name: "R_PPC_ADDR16" , reloc_addr, refsym); |
| 449 | *(Elf32_Half*) reloc_addr = finaladdr; |
| 450 | break; |
| 451 | |
| 452 | case R_PPC_UADDR16: |
| 453 | if (__glibc_unlikely (finaladdr > 0x7fff && finaladdr < 0xffff8000)) |
| 454 | _dl_reloc_overflow (map, name: "R_PPC_UADDR16" , reloc_addr, refsym); |
| 455 | ((union unaligned *) reloc_addr)->u2 = finaladdr; |
| 456 | break; |
| 457 | |
| 458 | case R_PPC_ADDR16_LO: |
| 459 | *(Elf32_Half*) reloc_addr = finaladdr; |
| 460 | break; |
| 461 | |
| 462 | case R_PPC_ADDR16_HI: |
| 463 | *(Elf32_Half*) reloc_addr = finaladdr >> 16; |
| 464 | break; |
| 465 | |
| 466 | case R_PPC_ADDR16_HA: |
| 467 | *(Elf32_Half*) reloc_addr = (finaladdr + 0x8000) >> 16; |
| 468 | break; |
| 469 | |
| 470 | case R_PPC_ADDR14: |
| 471 | case R_PPC_ADDR14_BRTAKEN: |
| 472 | case R_PPC_ADDR14_BRNTAKEN: |
| 473 | if (__glibc_unlikely (finaladdr > 0x7fff && finaladdr < 0xffff8000)) |
| 474 | _dl_reloc_overflow (map, name: "R_PPC_ADDR14" , reloc_addr, refsym); |
| 475 | *reloc_addr = (*reloc_addr & 0xffff0003) | (finaladdr & 0xfffc); |
| 476 | if (rinfo != R_PPC_ADDR14) |
| 477 | *reloc_addr = ((*reloc_addr & 0xffdfffff) |
| 478 | | ((rinfo == R_PPC_ADDR14_BRTAKEN) |
| 479 | ^ (finaladdr >> 31)) << 21); |
| 480 | break; |
| 481 | |
| 482 | case R_PPC_REL24: |
| 483 | { |
| 484 | Elf32_Sword delta = finaladdr - (Elf32_Word) reloc_addr; |
| 485 | if (delta << 6 >> 6 != delta) |
| 486 | _dl_reloc_overflow (map, name: "R_PPC_REL24" , reloc_addr, refsym); |
| 487 | *reloc_addr = (*reloc_addr & 0xfc000003) | (delta & 0x3fffffc); |
| 488 | } |
| 489 | break; |
| 490 | |
| 491 | case R_PPC_COPY: |
| 492 | if (sym == NULL) |
| 493 | /* This can happen in trace mode when an object could not be |
| 494 | found. */ |
| 495 | return; |
| 496 | if (sym->st_size > refsym->st_size |
| 497 | || (GLRO(dl_verbose) && sym->st_size < refsym->st_size)) |
| 498 | { |
| 499 | const char *strtab; |
| 500 | |
| 501 | strtab = (const void *) D_PTR (map, l_info[DT_STRTAB]); |
| 502 | _dl_error_printf (fmt: "\ |
| 503 | %s: Symbol `%s' has different size in shared object, consider re-linking\n" , |
| 504 | RTLD_PROGNAME, strtab + refsym->st_name); |
| 505 | } |
| 506 | memcpy (reloc_addr, (char *) finaladdr, MIN (sym->st_size, |
| 507 | refsym->st_size)); |
| 508 | return; |
| 509 | |
| 510 | case R_PPC_REL32: |
| 511 | *reloc_addr = finaladdr - (Elf32_Word) reloc_addr; |
| 512 | return; |
| 513 | |
| 514 | case R_PPC_JMP_SLOT: |
| 515 | /* It used to be that elf_machine_fixup_plt was used here, |
| 516 | but that doesn't work when ld.so relocates itself |
| 517 | for the second time. On the bright side, there's |
| 518 | no need to worry about thread-safety here. */ |
| 519 | { |
| 520 | Elf32_Sword delta = finaladdr - (Elf32_Word) reloc_addr; |
| 521 | if (delta << 6 >> 6 == delta) |
| 522 | *reloc_addr = OPCODE_B (delta); |
| 523 | else if (finaladdr <= 0x01fffffc || finaladdr >= 0xfe000000) |
| 524 | *reloc_addr = OPCODE_BA (finaladdr); |
| 525 | else |
| 526 | { |
| 527 | Elf32_Word *plt, *data_words; |
| 528 | Elf32_Word index, offset, num_plt_entries; |
| 529 | |
| 530 | plt = (Elf32_Word *) D_PTR (map, l_info[DT_PLTGOT]); |
| 531 | offset = reloc_addr - plt; |
| 532 | |
| 533 | if (offset < PLT_DOUBLE_SIZE*2 + PLT_INITIAL_ENTRY_WORDS) |
| 534 | { |
| 535 | index = (offset - PLT_INITIAL_ENTRY_WORDS)/2; |
| 536 | num_plt_entries = (map->l_info[DT_PLTRELSZ]->d_un.d_val |
| 537 | / sizeof (Elf32_Rela)); |
| 538 | data_words = plt + PLT_DATA_START_WORDS (num_plt_entries); |
| 539 | data_words[index] = finaladdr; |
| 540 | reloc_addr[0] = OPCODE_LI (11, index * 4); |
| 541 | reloc_addr[1] = OPCODE_B ((PLT_LONGBRANCH_ENTRY_WORDS |
| 542 | - (offset+1)) |
| 543 | * 4); |
| 544 | MODIFIED_CODE_NOQUEUE (reloc_addr + 1); |
| 545 | } |
| 546 | else |
| 547 | { |
| 548 | reloc_addr[0] = OPCODE_LIS_HI (12, finaladdr); |
| 549 | reloc_addr[1] = OPCODE_ADDI (12, 12, finaladdr); |
| 550 | reloc_addr[2] = OPCODE_MTCTR (12); |
| 551 | reloc_addr[3] = OPCODE_BCTR (); |
| 552 | MODIFIED_CODE_NOQUEUE (reloc_addr + 3); |
| 553 | } |
| 554 | } |
| 555 | } |
| 556 | break; |
| 557 | |
| 558 | #define DO_TLS_RELOC(suffix) \ |
| 559 | case R_PPC_DTPREL##suffix: \ |
| 560 | /* During relocation all TLS symbols are defined and used. \ |
| 561 | Therefore the offset is already correct. */ \ |
| 562 | if (sym_map != NULL) \ |
| 563 | do_reloc##suffix ("R_PPC_DTPREL"#suffix, \ |
| 564 | TLS_DTPREL_VALUE (sym, reloc)); \ |
| 565 | break; \ |
| 566 | case R_PPC_TPREL##suffix: \ |
| 567 | if (sym_map != NULL) \ |
| 568 | { \ |
| 569 | CHECK_STATIC_TLS (map, sym_map); \ |
| 570 | do_reloc##suffix ("R_PPC_TPREL"#suffix, \ |
| 571 | TLS_TPREL_VALUE (sym_map, sym, reloc)); \ |
| 572 | } \ |
| 573 | break; |
| 574 | |
| 575 | inline void do_reloc16 (const char *r_name, Elf32_Addr value) |
| 576 | { |
| 577 | if (__glibc_unlikely (value > 0x7fff && value < 0xffff8000)) |
| 578 | _dl_reloc_overflow (map, r_name, reloc_addr, refsym); |
| 579 | *(Elf32_Half *) reloc_addr = value; |
| 580 | } |
| 581 | inline void do_reloc16_LO (const char *r_name, Elf32_Addr value) |
| 582 | { |
| 583 | *(Elf32_Half *) reloc_addr = value; |
| 584 | } |
| 585 | inline void do_reloc16_HI (const char *r_name, Elf32_Addr value) |
| 586 | { |
| 587 | *(Elf32_Half *) reloc_addr = value >> 16; |
| 588 | } |
| 589 | inline void do_reloc16_HA (const char *r_name, Elf32_Addr value) |
| 590 | { |
| 591 | *(Elf32_Half *) reloc_addr = (value + 0x8000) >> 16; |
| 592 | } |
| 593 | DO_TLS_RELOC (16) |
| 594 | DO_TLS_RELOC (16_LO) |
| 595 | DO_TLS_RELOC (16_HI) |
| 596 | DO_TLS_RELOC (16_HA) |
| 597 | |
| 598 | default: |
| 599 | _dl_reloc_bad_type (map, type: rinfo, plt: 0); |
| 600 | return; |
| 601 | } |
| 602 | |
| 603 | MODIFIED_CODE_NOQUEUE (reloc_addr); |
| 604 | } |
| 605 | |