1 /* $NetBSD: ppc_reloc.c,v 1.10 2001/09/10 06:09:41 mycroft Exp $ */ 2 3 /*- 4 * Copyright (C) 1998 Tsubai Masanari 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The name of the author may not be used to endorse or promote products 16 * derived from this software without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 * 29 * $FreeBSD$ 30 */ 31 32 #include <sys/param.h> 33 #include <sys/mman.h> 34 35 #include <errno.h> 36 #include <stdio.h> 37 #include <stdlib.h> 38 #include <string.h> 39 #include <unistd.h> 40 #include <machine/cpu.h> 41 #include <machine/cpufunc.h> 42 #include <machine/md_var.h> 43 44 #include "debug.h" 45 #include "rtld.h" 46 47 #define _ppc_ha(x) ((((u_int32_t)(x) & 0x8000) ? \ 48 ((u_int32_t)(x) + 0x10000) : (u_int32_t)(x)) >> 16) 49 #define _ppc_la(x) ((u_int32_t)(x) & 0xffff) 50 51 #define min(a,b) (((a) < (b)) ? (a) : (b)) 52 #define max(a,b) (((a) > (b)) ? (a) : (b)) 53 54 #define PLT_EXTENDED_BEGIN (1 << 13) 55 #define JMPTAB_BASE(N) (18 + N*2 + ((N > PLT_EXTENDED_BEGIN) ? \ 56 (N - PLT_EXTENDED_BEGIN)*2 : 0)) 57 58 /* 59 * Process the R_PPC_COPY relocations 60 */ 61 int 62 do_copy_relocations(Obj_Entry *dstobj) 63 { 64 const Elf_Rela *relalim; 65 const Elf_Rela *rela; 66 67 /* 68 * COPY relocs are invalid outside of the main program 69 */ 70 assert(dstobj->mainprog); 71 72 relalim = (const Elf_Rela *) ((caddr_t) dstobj->rela + 73 dstobj->relasize); 74 for (rela = dstobj->rela; rela < relalim; rela++) { 75 void *dstaddr; 76 const Elf_Sym *dstsym; 77 const char *name; 78 size_t size; 79 const void *srcaddr; 80 const Elf_Sym *srcsym = NULL; 81 const Obj_Entry *srcobj, *defobj; 82 SymLook req; 83 int res; 84 85 if (ELF_R_TYPE(rela->r_info) != R_PPC_COPY) { 86 continue; 87 } 88 89 dstaddr = (void *) (dstobj->relocbase + rela->r_offset); 90 dstsym = dstobj->symtab + ELF_R_SYM(rela->r_info); 91 name = dstobj->strtab + dstsym->st_name; 92 size = dstsym->st_size; 93 symlook_init(&req, name); 94 req.ventry = fetch_ventry(dstobj, ELF_R_SYM(rela->r_info)); 95 96 for (srcobj = dstobj->next; srcobj != NULL; 97 srcobj = srcobj->next) { 98 res = symlook_obj(&req, srcobj); 99 if (res == 0) { 100 srcsym = req.sym_out; 101 defobj = req.defobj_out; 102 break; 103 } 104 } 105 106 if (srcobj == NULL) { 107 _rtld_error("Undefined symbol \"%s\" " 108 " referenced from COPY" 109 " relocation in %s", name, dstobj->path); 110 return (-1); 111 } 112 113 srcaddr = (const void *) (defobj->relocbase+srcsym->st_value); 114 memcpy(dstaddr, srcaddr, size); 115 dbg("copy_reloc: src=%p,dst=%p,size=%d\n",srcaddr,dstaddr,size); 116 } 117 118 return (0); 119 } 120 121 122 /* 123 * Perform early relocation of the run-time linker image 124 */ 125 void 126 reloc_non_plt_self(Elf_Dyn *dynp, Elf_Addr relocbase) 127 { 128 const Elf_Rela *rela = 0, *relalim; 129 Elf_Addr relasz = 0; 130 Elf_Addr *where; 131 132 /* 133 * Extract the rela/relasz values from the dynamic section 134 */ 135 for (; dynp->d_tag != DT_NULL; dynp++) { 136 switch (dynp->d_tag) { 137 case DT_RELA: 138 rela = (const Elf_Rela *)(relocbase+dynp->d_un.d_ptr); 139 break; 140 case DT_RELASZ: 141 relasz = dynp->d_un.d_val; 142 break; 143 } 144 } 145 146 /* 147 * Relocate these values 148 */ 149 relalim = (const Elf_Rela *)((caddr_t)rela + relasz); 150 for (; rela < relalim; rela++) { 151 where = (Elf_Addr *)(relocbase + rela->r_offset); 152 *where = (Elf_Addr)(relocbase + rela->r_addend); 153 } 154 } 155 156 157 /* 158 * Relocate a non-PLT object with addend. 159 */ 160 static int 161 reloc_nonplt_object(Obj_Entry *obj_rtld, Obj_Entry *obj, const Elf_Rela *rela, 162 SymCache *cache, RtldLockState *lockstate) 163 { 164 Elf_Addr *where = (Elf_Addr *)(obj->relocbase + rela->r_offset); 165 const Elf_Sym *def; 166 const Obj_Entry *defobj; 167 Elf_Addr tmp; 168 169 switch (ELF_R_TYPE(rela->r_info)) { 170 171 case R_PPC_NONE: 172 break; 173 174 case R_PPC_ADDR32: /* word32 S + A */ 175 case R_PPC_GLOB_DAT: /* word32 S + A */ 176 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, 177 false, cache, lockstate); 178 if (def == NULL) { 179 return (-1); 180 } 181 182 tmp = (Elf_Addr)(defobj->relocbase + def->st_value + 183 rela->r_addend); 184 185 /* Don't issue write if unnecessary; avoid COW page fault */ 186 if (*where != tmp) { 187 *where = tmp; 188 } 189 break; 190 191 case R_PPC_RELATIVE: /* word32 B + A */ 192 tmp = (Elf_Addr)(obj->relocbase + rela->r_addend); 193 194 /* As above, don't issue write unnecessarily */ 195 if (*where != tmp) { 196 *where = tmp; 197 } 198 break; 199 200 case R_PPC_COPY: 201 /* 202 * These are deferred until all other relocations 203 * have been done. All we do here is make sure 204 * that the COPY relocation is not in a shared 205 * library. They are allowed only in executable 206 * files. 207 */ 208 if (!obj->mainprog) { 209 _rtld_error("%s: Unexpected R_COPY " 210 " relocation in shared library", 211 obj->path); 212 return (-1); 213 } 214 break; 215 216 case R_PPC_JMP_SLOT: 217 /* 218 * These will be handled by the plt/jmpslot routines 219 */ 220 break; 221 222 case R_PPC_DTPMOD32: 223 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, 224 false, cache, lockstate); 225 226 if (def == NULL) 227 return (-1); 228 229 *where = (Elf_Addr) defobj->tlsindex; 230 231 break; 232 233 case R_PPC_TPREL32: 234 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, 235 false, cache, lockstate); 236 237 if (def == NULL) 238 return (-1); 239 240 /* 241 * We lazily allocate offsets for static TLS as we 242 * see the first relocation that references the 243 * TLS block. This allows us to support (small 244 * amounts of) static TLS in dynamically loaded 245 * modules. If we run out of space, we generate an 246 * error. 247 */ 248 if (!defobj->tls_done) { 249 if (!allocate_tls_offset((Obj_Entry*) defobj)) { 250 _rtld_error("%s: No space available for static " 251 "Thread Local Storage", obj->path); 252 return (-1); 253 } 254 } 255 256 *(Elf_Addr **)where = *where * sizeof(Elf_Addr) 257 + (Elf_Addr *)(def->st_value + rela->r_addend 258 + defobj->tlsoffset - TLS_TP_OFFSET); 259 260 break; 261 262 case R_PPC_DTPREL32: 263 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, 264 false, cache, lockstate); 265 266 if (def == NULL) 267 return (-1); 268 269 *where += (Elf_Addr)(def->st_value + rela->r_addend 270 - TLS_DTV_OFFSET); 271 272 break; 273 274 default: 275 _rtld_error("%s: Unsupported relocation type %d" 276 " in non-PLT relocations\n", obj->path, 277 ELF_R_TYPE(rela->r_info)); 278 return (-1); 279 } 280 return (0); 281 } 282 283 284 /* 285 * Process non-PLT relocations 286 */ 287 int 288 reloc_non_plt(Obj_Entry *obj, Obj_Entry *obj_rtld, RtldLockState *lockstate) 289 { 290 const Elf_Rela *relalim; 291 const Elf_Rela *rela; 292 SymCache *cache; 293 int r = -1; 294 295 /* 296 * The dynamic loader may be called from a thread, we have 297 * limited amounts of stack available so we cannot use alloca(). 298 */ 299 if (obj != obj_rtld) { 300 cache = calloc(obj->nchains, sizeof(SymCache)); 301 /* No need to check for NULL here */ 302 } else 303 cache = NULL; 304 305 /* 306 * From the SVR4 PPC ABI: 307 * "The PowerPC family uses only the Elf32_Rela relocation 308 * entries with explicit addends." 309 */ 310 relalim = (const Elf_Rela *)((caddr_t)obj->rela + obj->relasize); 311 for (rela = obj->rela; rela < relalim; rela++) { 312 if (reloc_nonplt_object(obj_rtld, obj, rela, cache, lockstate) 313 < 0) 314 goto done; 315 } 316 r = 0; 317 done: 318 if (cache != NULL) 319 free(cache); 320 return (r); 321 } 322 323 /* 324 * Initialise a PLT slot to the resolving trampoline 325 */ 326 static int 327 reloc_plt_object(Obj_Entry *obj, const Elf_Rela *rela) 328 { 329 Elf_Word *where = (Elf_Word *)(obj->relocbase + rela->r_offset); 330 Elf_Addr *pltresolve, *pltlongresolve, *jmptab; 331 Elf_Addr distance; 332 int N = obj->pltrelasize / sizeof(Elf_Rela); 333 int reloff; 334 335 reloff = rela - obj->pltrela; 336 337 if (reloff < 0) 338 return (-1); 339 340 pltlongresolve = obj->pltgot + 5; 341 pltresolve = pltlongresolve + 5; 342 343 distance = (Elf_Addr)pltresolve - (Elf_Addr)(where + 1); 344 345 dbg(" reloc_plt_object: where=%p,pltres=%p,reloff=%x,distance=%x", 346 (void *)where, (void *)pltresolve, reloff, distance); 347 348 if (reloff < PLT_EXTENDED_BEGIN) { 349 /* li r11,reloff */ 350 /* b pltresolve */ 351 where[0] = 0x39600000 | reloff; 352 where[1] = 0x48000000 | (distance & 0x03fffffc); 353 } else { 354 jmptab = obj->pltgot + JMPTAB_BASE(N); 355 jmptab[reloff] = (u_int)pltlongresolve; 356 357 /* lis r11,jmptab[reloff]@ha */ 358 /* lwzu r12,jmptab[reloff]@l(r11) */ 359 /* mtctr r12 */ 360 /* bctr */ 361 where[0] = 0x3d600000 | _ppc_ha(&jmptab[reloff]); 362 where[1] = 0x858b0000 | _ppc_la(&jmptab[reloff]); 363 where[2] = 0x7d8903a6; 364 where[3] = 0x4e800420; 365 } 366 367 368 /* 369 * The icache will be sync'd in init_pltgot, which is called 370 * after all the slots have been updated 371 */ 372 373 return (0); 374 } 375 376 377 /* 378 * Process the PLT relocations. 379 */ 380 int 381 reloc_plt(Obj_Entry *obj) 382 { 383 const Elf_Rela *relalim; 384 const Elf_Rela *rela; 385 386 if (obj->pltrelasize != 0) { 387 388 relalim = (const Elf_Rela *)((char *)obj->pltrela + 389 obj->pltrelasize); 390 for (rela = obj->pltrela; rela < relalim; rela++) { 391 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT); 392 393 if (reloc_plt_object(obj, rela) < 0) { 394 return (-1); 395 } 396 } 397 } 398 399 return (0); 400 } 401 402 403 /* 404 * LD_BIND_NOW was set - force relocation for all jump slots 405 */ 406 int 407 reloc_jmpslots(Obj_Entry *obj, RtldLockState *lockstate) 408 { 409 const Obj_Entry *defobj; 410 const Elf_Rela *relalim; 411 const Elf_Rela *rela; 412 const Elf_Sym *def; 413 Elf_Addr *where; 414 Elf_Addr target; 415 416 relalim = (const Elf_Rela *)((char *)obj->pltrela + obj->pltrelasize); 417 for (rela = obj->pltrela; rela < relalim; rela++) { 418 assert(ELF_R_TYPE(rela->r_info) == R_PPC_JMP_SLOT); 419 where = (Elf_Addr *)(obj->relocbase + rela->r_offset); 420 def = find_symdef(ELF_R_SYM(rela->r_info), obj, &defobj, 421 true, NULL, lockstate); 422 if (def == NULL) { 423 dbg("reloc_jmpslots: sym not found"); 424 return (-1); 425 } 426 427 target = (Elf_Addr)(defobj->relocbase + def->st_value); 428 429 #if 0 430 /* PG XXX */ 431 dbg("\"%s\" in \"%s\" --> %p in \"%s\"", 432 defobj->strtab + def->st_name, basename(obj->path), 433 (void *)target, basename(defobj->path)); 434 #endif 435 436 reloc_jmpslot(where, target, defobj, obj, 437 (const Elf_Rel *) rela); 438 } 439 440 obj->jmpslots_done = true; 441 442 return (0); 443 } 444 445 446 /* 447 * Update the value of a PLT jump slot. Branch directly to the target if 448 * it is within +/- 32Mb, otherwise go indirectly via the pltcall 449 * trampoline call and jump table. 450 */ 451 Elf_Addr 452 reloc_jmpslot(Elf_Addr *wherep, Elf_Addr target, const Obj_Entry *defobj, 453 const Obj_Entry *obj, const Elf_Rel *rel) 454 { 455 Elf_Addr offset; 456 const Elf_Rela *rela = (const Elf_Rela *) rel; 457 458 dbg(" reloc_jmpslot: where=%p, target=%p", 459 (void *)wherep, (void *)target); 460 461 /* 462 * At the PLT entry pointed at by `wherep', construct 463 * a direct transfer to the now fully resolved function 464 * address. 465 */ 466 offset = target - (Elf_Addr)wherep; 467 468 if (abs(offset) < 32*1024*1024) { /* inside 32MB? */ 469 /* b value # branch directly */ 470 *wherep = 0x48000000 | (offset & 0x03fffffc); 471 __syncicache(wherep, 4); 472 } else { 473 Elf_Addr *pltcall, *jmptab; 474 int distance; 475 int N = obj->pltrelasize / sizeof(Elf_Rela); 476 int reloff = rela - obj->pltrela; 477 478 if (reloff < 0) 479 return (-1); 480 481 pltcall = obj->pltgot; 482 483 dbg(" reloc_jmpslot: indir, reloff=%x, N=%x\n", 484 reloff, N); 485 486 jmptab = obj->pltgot + JMPTAB_BASE(N); 487 jmptab[reloff] = target; 488 powerpc_mb(); /* Order jmptab update before next changes */ 489 490 if (reloff < PLT_EXTENDED_BEGIN) { 491 /* for extended PLT entries, we keep the old code */ 492 493 distance = (Elf_Addr)pltcall - (Elf_Addr)(wherep + 1); 494 495 /* li r11,reloff */ 496 /* b pltcall # use indirect pltcall routine */ 497 498 /* first instruction same as before */ 499 wherep[1] = 0x48000000 | (distance & 0x03fffffc); 500 __syncicache(wherep, 8); 501 } 502 } 503 504 return (target); 505 } 506 507 int 508 reloc_iresolve(Obj_Entry *obj, struct Struct_RtldLockState *lockstate) 509 { 510 511 /* XXX not implemented */ 512 return (0); 513 } 514 515 int 516 reloc_gnu_ifunc(Obj_Entry *obj, struct Struct_RtldLockState *lockstate) 517 { 518 519 /* XXX not implemented */ 520 return (0); 521 } 522 523 /* 524 * Setup the plt glue routines. 525 */ 526 #define PLTCALL_SIZE 20 527 #define PLTLONGRESOLVE_SIZE 20 528 #define PLTRESOLVE_SIZE 24 529 530 void 531 init_pltgot(Obj_Entry *obj) 532 { 533 Elf_Word *pltcall, *pltresolve, *pltlongresolve; 534 Elf_Word *jmptab; 535 int N = obj->pltrelasize / sizeof(Elf_Rela); 536 537 pltcall = obj->pltgot; 538 539 if (pltcall == NULL) { 540 return; 541 } 542 543 /* 544 * From the SVR4 PPC ABI: 545 * 546 * 'The first 18 words (72 bytes) of the PLT are reserved for 547 * use by the dynamic linker. 548 * ... 549 * 'If the executable or shared object requires N procedure 550 * linkage table entries, the link editor shall reserve 3*N 551 * words (12*N bytes) following the 18 reserved words. The 552 * first 2*N of these words are the procedure linkage table 553 * entries themselves. The static linker directs calls to bytes 554 * (72 + (i-1)*8), for i between 1 and N inclusive. The remaining 555 * N words (4*N bytes) are reserved for use by the dynamic linker.' 556 */ 557 558 /* 559 * Copy the absolute-call assembler stub into the first part of 560 * the reserved PLT area. 561 */ 562 memcpy(pltcall, _rtld_powerpc_pltcall, PLTCALL_SIZE); 563 564 /* 565 * Determine the address of the jumptable, which is the dyn-linker 566 * reserved area after the call cells. Write the absolute address 567 * of the jumptable into the absolute-call assembler code so it 568 * can determine this address. 569 */ 570 jmptab = obj->pltgot + JMPTAB_BASE(N); 571 pltcall[1] |= _ppc_ha(jmptab); /* addis 11,11,jmptab@ha */ 572 pltcall[2] |= _ppc_la(jmptab); /* lwz 11,jmptab@l(11) */ 573 574 /* 575 * Skip down 20 bytes into the initial reserved area and copy 576 * in the standard resolving assembler call. Into this assembler, 577 * insert the absolute address of the _rtld_bind_start routine 578 * and the address of the relocation object. 579 * 580 * We place pltlongresolve first, so it can fix up its arguments 581 * and then fall through to the regular PLT resolver. 582 */ 583 pltlongresolve = obj->pltgot + 5; 584 585 memcpy(pltlongresolve, _rtld_powerpc_pltlongresolve, 586 PLTLONGRESOLVE_SIZE); 587 pltlongresolve[0] |= _ppc_ha(jmptab); /* lis 12,jmptab@ha */ 588 pltlongresolve[1] |= _ppc_la(jmptab); /* addi 12,12,jmptab@l */ 589 590 pltresolve = pltlongresolve + PLTLONGRESOLVE_SIZE/sizeof(uint32_t); 591 memcpy(pltresolve, _rtld_powerpc_pltresolve, PLTRESOLVE_SIZE); 592 pltresolve[0] |= _ppc_ha(_rtld_bind_start); 593 pltresolve[1] |= _ppc_la(_rtld_bind_start); 594 pltresolve[3] |= _ppc_ha(obj); 595 pltresolve[4] |= _ppc_la(obj); 596 597 /* 598 * Sync the icache for the byte range represented by the 599 * trampoline routines and call slots. 600 */ 601 __syncicache(obj->pltgot, JMPTAB_BASE(N)*4); 602 } 603 604 void 605 allocate_initial_tls(Obj_Entry *list) 606 { 607 register Elf_Addr **tp __asm__("r2"); 608 Elf_Addr **_tp; 609 610 /* 611 * Fix the size of the static TLS block by using the maximum 612 * offset allocated so far and adding a bit for dynamic modules to 613 * use. 614 */ 615 616 tls_static_space = tls_last_offset + tls_last_size + RTLD_STATIC_TLS_EXTRA; 617 618 _tp = (Elf_Addr **) ((char *) allocate_tls(list, NULL, TLS_TCB_SIZE, 8) 619 + TLS_TP_OFFSET + TLS_TCB_SIZE); 620 621 /* 622 * XXX gcc seems to ignore 'tp = _tp;' 623 */ 624 625 __asm __volatile("mr %0,%1" : "=r"(tp) : "r"(_tp)); 626 } 627 628 void* 629 __tls_get_addr(tls_index* ti) 630 { 631 register Elf_Addr **tp __asm__("r2"); 632 char *p; 633 634 p = tls_get_addr_common((Elf_Addr**)((Elf_Addr)tp - TLS_TP_OFFSET 635 - TLS_TCB_SIZE), ti->ti_module, ti->ti_offset); 636 637 return (p + TLS_DTV_OFFSET); 638 } 639