sysdeps/alpha/dl-machine.h

   1 /* Machine-dependent ELF dynamic relocation inline functions.  Alpha version.
   2    Copyright (C) 1996-2002, 2003, 2004 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4    Contributed by Richard Henderson <rth@tamu.edu>.
   5
   6    The GNU C Library is free software; you can redistribute it and/or
   7    modify it under the terms of the GNU Lesser General Public
   8    License as published by the Free Software Foundation; either
   9    version 2.1 of the License, or (at your option) any later version.
  10
  11    The GNU C Library is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14    Lesser General Public License for more details.
  15
  16    You should have received a copy of the GNU Lesser General Public
  17    License along with the GNU C Library; if not, write to the Free
  18    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  19    02111-1307 USA.  */
  20
  21 /* This was written in the absence of an ABI -- don't expect
  22    it to remain unchanged.  */
  23
  24 #ifndef dl_machine_h
  25 #define dl_machine_h 1
  26
  27 #define ELF_MACHINE_NAME "alpha"
  28
  29 #include <string.h>
  30
  31
  32 /* Mask identifying addresses reserved for the user program,
  33    where the dynamic linker should not map anything.  */
  34 #define ELF_MACHINE_USER_ADDRESS_MASK   0x120000000UL
  35
  36 /* Return nonzero iff ELF header is compatible with the running host.  */
  37 static inline int
  38 elf_machine_matches_host (const Elf64_Ehdr *ehdr)
  39 {
  40   return ehdr->e_machine == EM_ALPHA;
  41 }
  42
  43 /* Return the link-time address of _DYNAMIC.  The multiple-got-capable
  44    linker no longer allocates the first .got entry for this.  But not to
  45    worry, no special tricks are needed.  */
  46 static inline Elf64_Addr
  47 elf_machine_dynamic (void)
  48 {
  49 #ifndef NO_AXP_MULTI_GOT_LD
  50   return (Elf64_Addr) &_DYNAMIC;
  51 #else
  52   register Elf64_Addr *gp __asm__ ("$29");
  53   return gp[-4096];
  54 #endif
  55 }
  56
  57 /* Return the run-time load address of the shared object.  */
  58 static inline Elf64_Addr
  59 elf_machine_load_address (void)
  60 {
  61   /* NOTE: While it is generally unfriendly to put data in the text
  62      segment, it is only slightly less so when the "data" is an
  63      instruction.  While we don't have to worry about GLD just yet, an
  64      optimizing linker might decide that our "data" is an unreachable
  65      instruction and throw it away -- with the right switches, DEC's
  66      linker will do this.  What ought to happen is we should add
  67      something to GAS to allow us access to the new GPREL_HI32/LO32
  68      relocation types stolen from OSF/1 3.0.  */
  69   /* This code relies on the fact that BRADDR relocations do not
  70      appear in dynamic relocation tables.  Not that that would be very
  71      useful anyway -- br/bsr has a 4MB range and the shared libraries
  72      are usually many many terabytes away.  */
  73
  74   Elf64_Addr dot;
  75   long int zero_disp;
  76
  77   asm("br %0, 1f\n"
  78       "0:\n\t"
  79       "br $0, 2f\n"
  80       "1:\n\t"
  81       ".section\t.data\n"
  82       "2:\n\t"
  83       ".quad 0b\n\t"
  84       ".previous"
  85       : "=r"(dot));
  86
  87   zero_disp = *(int *) dot;
  88   zero_disp = (zero_disp << 43) >> 41;
  89
  90   return dot - *(Elf64_Addr *) (dot + 4 + zero_disp);
  91 }
  92
  93 /* Set up the loaded object described by L so its unrelocated PLT
  94    entries will jump to the on-demand fixup code in dl-runtime.c.  */
  95
  96 static inline int
  97 elf_machine_runtime_setup (struct link_map *l, int lazy, int profile)
  98 {
  99   Elf64_Addr plt;
 100   extern void _dl_runtime_resolve (void);
 101   extern void _dl_runtime_profile (void);
 102
 103   if (l->l_info[DT_JMPREL] && lazy)
 104     {
 105       /* The GOT entries for the functions in the PLT have not been
 106          filled in yet.  Their initial contents are directed to the
 107          PLT which arranges for the dynamic linker to be called.  */
 108       plt = D_PTR (l, l_info[DT_PLTGOT]);
 109
 110       /* This function will be called to perform the relocation.  */
 111       if (__builtin_expect (profile, 0))
 112         {
 113           *(Elf64_Addr *)(plt + 16) = (Elf64_Addr) &_dl_runtime_profile;
 114
 115           if (GLRO(dl_profile) != NULL
 116               && _dl_name_match_p (GLRO(dl_profile), l))
 117             {
 118               /* This is the object we are looking for.  Say that we really
 119                  want profiling and the timers are started.  */
 120               GL(dl_profile_map) = l;
 121             }
 122         }
 123       else
 124         *(Elf64_Addr *)(plt + 16) = (Elf64_Addr) &_dl_runtime_resolve;
 125
 126       /* Identify this shared object */
 127       *(Elf64_Addr *)(plt + 24) = (Elf64_Addr) l;
 128
 129       /* If the first instruction of the plt entry is not
 130          "br $28, plt0", we have to reinitialize .plt for lazy relocation.  */
 131       if (*(unsigned int *)(plt + 32) != 0xc39ffff7)
 132         {
 133           unsigned int val = 0xc39ffff7;
 134           unsigned int *slot, *end;
 135           const Elf64_Rela *rela = (const Elf64_Rela *)
 136                                    D_PTR (l, l_info[DT_JMPREL]);
 137           Elf64_Addr l_addr = l->l_addr;
 138
 139           /* br t12,.+4; ldq t12,12(t12); nop; jmp t12,(t12),.+4 */
 140           *(unsigned long *)plt = 0xa77b000cc3600000;
 141           *(unsigned long *)(plt + 8) = 0x6b7b000047ff041f;
 142           slot = (unsigned int *)(plt + 32);
 143           end = (unsigned int *)(plt + 32
 144                                  + l->l_info[DT_PLTRELSZ]->d_un.d_val / 2);
 145           while (slot < end)
 146             {
 147               /* br at,.plt+0 */
 148               *slot = val;
 149               *(Elf64_Addr *) rela->r_offset = (Elf64_Addr) slot - l_addr;
 150               val -= 3;
 151               slot += 3;
 152               ++rela;
 153             }
 154         }
 155     }
 156
 157   return lazy;
 158 }
 159
 160 /* Initial entry point code for the dynamic linker.
 161    The C function `_dl_start' is the real entry point;
 162    its return value is the user program's entry point.  */
 163
 164 #define RTLD_START asm ("\
 165         .section .text                                          \n\
 166         .set at                                                 \n\
 167         .globl _start                                           \n\
 168         .ent _start                                             \n\
 169 _start:                                                         \n\
 170         .frame $31,0,$31,0                                      \n\
 171         br      $gp, 0f                                         \n\
 172 0:      ldgp    $gp, 0($gp)                                     \n\
 173         .prologue 0                                             \n\
 174         /* Pass pointer to argument block to _dl_start.  */     \n\
 175         mov     $sp, $16                                        \n\
 176         bsr     $26, _dl_start          !samegp                 \n\
 177         .end _start                                             \n\
 178         /* FALLTHRU */                                          \n\
 179         .globl _dl_start_user                                   \n\
 180         .ent _dl_start_user                                     \n\
 181 _dl_start_user:                                                 \n\
 182         .frame $31,0,$31,0                                      \n\
 183         .prologue 0                                             \n\
 184         /* Save the user entry point address in s0.  */         \n\
 185         mov     $0, $9                                          \n\
 186         /* See if we were run as a command with the executable  \n\
 187            file name as an extra leading argument.  */          \n\
 188         ldah    $1, _dl_skip_args($gp)  !gprelhigh              \n\
 189         ldl     $1, _dl_skip_args($1)   !gprellow               \n\
 190         bne     $1, $fixup_stack                                \n\
 191 $fixup_stack_ret:                                               \n\
 192         /* The special initializer gets called with the stack   \n\
 193            just as the application's entry point will see it;   \n\
 194            it can switch stacks if it moves these contents      \n\
 195            over.  */                                            \n\
 196 " RTLD_START_SPECIAL_INIT "                                     \n\
 197         /* Call _dl_init(_dl_loaded, argc, argv, envp) to run   \n\
 198            initializers.  */                                    \n\
 199         ldah    $16, _rtld_local($gp)   !gprelhigh              \n\
 200         ldq     $16, _rtld_local($16)   !gprellow               \n\
 201         ldq     $17, 0($sp)                                     \n\
 202         lda     $18, 8($sp)                                     \n\
 203         s8addq  $17, 8, $19                                     \n\
 204         addq    $19, $18, $19                                   \n\
 205         bsr     $26, _dl_init_internal  !samegp                 \n\
 206         /* Pass our finalizer function to the user in $0. */    \n\
 207         ldah    $0, _dl_fini($gp)       !gprelhigh              \n\
 208         lda     $0, _dl_fini($0)        !gprellow               \n\
 209         /* Jump to the user's entry point.  */                  \n\
 210         mov     $9, $27                                         \n\
 211         jmp     ($9)                                            \n\
 212 $fixup_stack:                                                   \n\
 213         /* Adjust the stack pointer to skip _dl_skip_args words.\n\
 214            This involves copying everything down, since the     \n\
 215            stack pointer must always be 16-byte aligned.  */    \n\
 216         ldah    $7, _dl_argv_internal($gp) !gprelhigh           \n\
 217         ldq     $2, 0($sp)                                      \n\
 218         ldq     $5, _dl_argv_internal($7) !gprellow             \n\
 219         subq    $31, $1, $6                                     \n\
 220         subq    $2, $1, $2                                      \n\
 221         s8addq  $6, $5, $5                                      \n\
 222         mov     $sp, $4                                         \n\
 223         s8addq  $1, $sp, $3                                     \n\
 224         stq     $2, 0($sp)                                      \n\
 225         stq     $5, _dl_argv_internal($7) !gprellow             \n\
 226         /* Copy down argv.  */                                  \n\
 227 0:      ldq     $5, 8($3)                                       \n\
 228         addq    $4, 8, $4                                       \n\
 229         addq    $3, 8, $3                                       \n\
 230         stq     $5, 0($4)                                       \n\
 231         bne     $5, 0b                                          \n\
 232         /* Copy down envp.  */                                  \n\
 233 1:      ldq     $5, 8($3)                                       \n\
 234         addq    $4, 8, $4                                       \n\
 235         addq    $3, 8, $3                                       \n\
 236         stq     $5, 0($4)                                       \n\
 237         bne     $5, 1b                                          \n\
 238         /* Copy down auxiliary table.  */                       \n\
 239 2:      ldq     $5, 8($3)                                       \n\
 240         ldq     $6, 16($3)                                      \n\
 241         addq    $4, 16, $4                                      \n\
 242         addq    $3, 16, $3                                      \n\
 243         stq     $5, -8($4)                                      \n\
 244         stq     $6, 0($4)                                       \n\
 245         bne     $5, 2b                                          \n\
 246         br      $fixup_stack_ret                                \n\
 247         .end _dl_start_user                                     \n\
 248         .set noat                                               \n\
 249 .previous");
 250
 251 #ifndef RTLD_START_SPECIAL_INIT
 252 #define RTLD_START_SPECIAL_INIT /* nothing */
 253 #endif
 254
 255 /* ELF_RTYPE_CLASS_PLT iff TYPE describes relocation of a PLT entry
 256    or TLS variables, so undefined references should not be allowed
 257    to define the value.
 258
 259    ELF_RTYPE_CLASS_NOCOPY iff TYPE should not be allowed to resolve
 260    to one of the main executable's symbols, as for a COPY reloc.
 261    This is unused on Alpha.  */
 262
 263 #if defined USE_TLS && (!defined RTLD_BOOTSTRAP || USE___THREAD)
 264 #define elf_machine_type_class(type)    \
 265   (((type) == R_ALPHA_JMP_SLOT          \
 266     || (type) == R_ALPHA_DTPMOD64       \
 267     || (type) == R_ALPHA_DTPREL64       \
 268     || (type) == R_ALPHA_TPREL64) * ELF_RTYPE_CLASS_PLT)
 269 #else
 270 #define elf_machine_type_class(type)    \
 271   (((type) == R_ALPHA_JMP_SLOT) * ELF_RTYPE_CLASS_PLT)
 272 #endif
 273
 274 /* A reloc type used for ld.so cmdline arg lookups to reject PLT entries.  */
 275 #define ELF_MACHINE_JMP_SLOT     R_ALPHA_JMP_SLOT
 276
 277 /* The alpha never uses Elf64_Rel relocations.  */
 278 #define ELF_MACHINE_NO_REL 1
 279
 280 /* Fix up the instructions of a PLT entry to invoke the function
 281    rather than the dynamic linker.  */
 282 static inline Elf64_Addr
 283 elf_machine_fixup_plt (struct link_map *l, lookup_t t,
 284                        const Elf64_Rela *reloc,
 285                        Elf64_Addr *got_addr, Elf64_Addr value)
 286 {
 287   const Elf64_Rela *rela_plt;
 288   Elf64_Word *plte;
 289   long int edisp;
 290
 291   /* Store the value we are going to load.  */
 292   *got_addr = value;
 293
 294   /* Recover the PLT entry address by calculating reloc's index into the
 295      .rela.plt, and finding that entry in the .plt.  */
 296   rela_plt = (void *) D_PTR (l, l_info[DT_JMPREL]);
 297   plte = (void *) (D_PTR (l, l_info[DT_PLTGOT]) + 32);
 298   plte += 3 * (reloc - rela_plt);
 299
 300   /* Find the displacement from the plt entry to the function.  */
 301   edisp = (long int) (value - (Elf64_Addr)&plte[3]) / 4;
 302
 303   if (edisp >= -0x100000 && edisp < 0x100000)
 304     {
 305       /* If we are in range, use br to perfect branch prediction and
 306          elide the dependency on the address load.  This case happens,
 307          e.g., when a shared library call is resolved to the same library.  */
 308
 309       int hi, lo;
 310       hi = value - (Elf64_Addr)&plte[0];
 311       lo = (short int) hi;
 312       hi = (hi - lo) >> 16;
 313
 314       /* Emit "lda $27,lo($27)" */
 315       plte[1] = 0x237b0000 | (lo & 0xffff);
 316
 317       /* Emit "br $31,function" */
 318       plte[2] = 0xc3e00000 | (edisp & 0x1fffff);
 319
 320       /* Think about thread-safety -- the previous instructions must be
 321          committed to memory before the first is overwritten.  */
 322       __asm__ __volatile__("wmb" : : : "memory");
 323
 324       /* Emit "ldah $27,hi($27)" */
 325       plte[0] = 0x277b0000 | (hi & 0xffff);
 326     }
 327   else
 328     {
 329       /* Don't bother with the hint since we already know the hint is
 330          wrong.  Eliding it prevents the wrong page from getting pulled
 331          into the cache.  */
 332
 333       int hi, lo;
 334       hi = (Elf64_Addr)got_addr - (Elf64_Addr)&plte[0];
 335       lo = (short)hi;
 336       hi = (hi - lo) >> 16;
 337
 338       /* Emit "ldq $27,lo($27)" */
 339       plte[1] = 0xa77b0000 | (lo & 0xffff);
 340
 341       /* Emit "jmp $31,($27)" */
 342       plte[2] = 0x6bfb0000;
 343
 344       /* Think about thread-safety -- the previous instructions must be
 345          committed to memory before the first is overwritten.  */
 346       __asm__ __volatile__("wmb" : : : "memory");
 347
 348       /* Emit "ldah $27,hi($27)" */
 349       plte[0] = 0x277b0000 | (hi & 0xffff);
 350     }
 351
 352   /* At this point, if we've been doing runtime resolution, Icache is dirty.
 353      This will be taken care of in _dl_runtime_resolve.  If instead we are
 354      doing this as part of non-lazy startup relocation, that bit of code
 355      hasn't made it into Icache yet, so there's nothing to clean up.  */
 356
 357   return value;
 358 }
 359
 360 /* Return the final value of a plt relocation.  */
 361 static inline Elf64_Addr
 362 elf_machine_plt_value (struct link_map *map, const Elf64_Rela *reloc,
 363                        Elf64_Addr value)
 364 {
 365   return value + reloc->r_addend;
 366 }
 367
 368 /* Names of the architecture-specific auditing callback functions.  */
 369 #define ARCH_LA_PLTENTER        alpha_gnu_pltenter
 370 #define ARCH_LA_PLTEXIT         alpha_gnu_pltexit
 371
 372 #endif /* !dl_machine_h */
 373
 374 #ifdef RESOLVE_MAP
 375
 376 /* Perform the relocation specified by RELOC and SYM (which is fully resolved).
 377    MAP is the object containing the reloc.  */
 378 auto inline void
 379 elf_machine_rela (struct link_map *map,
 380                   const Elf64_Rela *reloc,
 381                   const Elf64_Sym *sym,
 382                   const struct r_found_version *version,
 383                   void *const reloc_addr_arg)
 384 {
 385   Elf64_Addr *const reloc_addr = reloc_addr_arg;
 386   unsigned long int const r_type = ELF64_R_TYPE (reloc->r_info);
 387
 388 #if !defined RTLD_BOOTSTRAP && !defined HAVE_Z_COMBRELOC && !defined SHARED
 389   /* This is defined in rtld.c, but nowhere in the static libc.a; make the
 390      reference weak so static programs can still link.  This declaration
 391      cannot be done when compiling rtld.c (i.e.  #ifdef RTLD_BOOTSTRAP)
 392      because rtld.c contains the common defn for _dl_rtld_map, which is
 393      incompatible with a weak decl in the same file.  */
 394   weak_extern (_dl_rtld_map);
 395 #endif
 396
 397   /* We cannot use a switch here because we cannot locate the switch
 398      jump table until we've self-relocated.  */
 399
 400 #if !defined RTLD_BOOTSTRAP || !defined HAVE_Z_COMBRELOC
 401   if (__builtin_expect (r_type == R_ALPHA_RELATIVE, 0))
 402     {
 403 # if !defined RTLD_BOOTSTRAP && !defined HAVE_Z_COMBRELOC
 404       /* Already done in dynamic linker.  */
 405       if (map != &GL(dl_rtld_map))
 406 # endif
 407         {
 408           /* XXX Make some timings.  Maybe it's preferable to test for
 409              unaligned access and only do it the complex way if necessary.  */
 410           Elf64_Addr reloc_addr_val;
 411
 412           /* Load value without causing unaligned trap. */
 413           memcpy (&reloc_addr_val, reloc_addr_arg, 8);
 414           reloc_addr_val += map->l_addr;
 415
 416           /* Store value without causing unaligned trap. */
 417           memcpy (reloc_addr_arg, &reloc_addr_val, 8);
 418         }
 419     }
 420   else
 421 #endif
 422     if (__builtin_expect (r_type == R_ALPHA_NONE, 0))
 423       return;
 424   else
 425     {
 426       struct link_map *sym_map = RESOLVE_MAP (&sym, version, r_type);
 427       Elf64_Addr sym_value;
 428       Elf64_Addr sym_raw_value;
 429
 430       sym_raw_value = sym_value = reloc->r_addend;
 431       if (sym_map)
 432         {
 433           sym_raw_value += sym->st_value;
 434           sym_value = sym_raw_value + sym_map->l_addr;
 435         }
 436
 437       if (r_type == R_ALPHA_GLOB_DAT)
 438         *reloc_addr = sym_value;
 439 #ifdef RESOLVE_CONFLICT_FIND_MAP
 440       /* In .gnu.conflict section, R_ALPHA_JMP_SLOT relocations have
 441          R_ALPHA_JMP_SLOT in lower 8 bits and the remaining 24 bits
 442          are .rela.plt index.  */
 443       else if ((r_type & 0xff) == R_ALPHA_JMP_SLOT)
 444         {
 445           /* elf_machine_fixup_plt needs the map reloc_addr points into,
 446              while in _dl_resolve_conflicts map is _dl_loaded.  */
 447           RESOLVE_CONFLICT_FIND_MAP (map, reloc_addr);
 448           reloc = ((const Elf64_Rela *) D_PTR (map, l_info[DT_JMPREL]))
 449                   + (r_type >> 8);
 450           elf_machine_fixup_plt (map, 0, reloc, reloc_addr, sym_value);
 451         }
 452 #else
 453       else if (r_type == R_ALPHA_JMP_SLOT)
 454         elf_machine_fixup_plt (map, 0, reloc, reloc_addr, sym_value);
 455 #endif
 456 #ifndef RTLD_BOOTSTRAP
 457       else if (r_type == R_ALPHA_REFQUAD)
 458         {
 459           /* Store value without causing unaligned trap.  */
 460           memcpy (reloc_addr_arg, &sym_value, 8);
 461         }
 462 #endif
 463 #if defined USE_TLS && (!defined RTLD_BOOTSTRAP || USE___THREAD)
 464       else if (r_type == R_ALPHA_DTPMOD64)
 465         {
 466 #ifdef RTLD_BOOTSTRAP
 467           /* During startup the dynamic linker is always index 1.  */
 468           *reloc_addr = 1;
 469 #else
 470           /* Get the information from the link map returned by the
 471              resolv function.  */
 472           if (sym_map != NULL)
 473             *reloc_addr = sym_map->l_tls_modid;
 474 #endif
 475         }
 476       else if (r_type == R_ALPHA_DTPREL64)
 477         {
 478 #ifndef RTLD_BOOTSTRAP
 479           /* During relocation all TLS symbols are defined and used.
 480              Therefore the offset is already correct.  */
 481           *reloc_addr = sym_raw_value;
 482 #endif
 483         }
 484       else if (r_type == R_ALPHA_TPREL64)
 485         {
 486 #ifdef RTLD_BOOTSTRAP
 487           *reloc_addr = sym_raw_value + map->l_tls_offset;
 488 #else
 489           if (sym_map)
 490             {
 491               CHECK_STATIC_TLS (map, sym_map);
 492               *reloc_addr = sym_raw_value + sym_map->l_tls_offset;
 493             }
 494 #endif
 495         }
 496 #endif /* USE_TLS */
 497       else
 498         _dl_reloc_bad_type (map, r_type, 0);
 499     }
 500 }
 501
 502 /* Let do-rel.h know that on Alpha if l_addr is 0, all RELATIVE relocs
 503    can be skipped.  */
 504 #define ELF_MACHINE_REL_RELATIVE 1
 505
 506 auto inline void
 507 elf_machine_rela_relative (Elf64_Addr l_addr, const Elf64_Rela *reloc,
 508                            void *const reloc_addr_arg)
 509 {
 510   /* XXX Make some timings.  Maybe it's preferable to test for
 511      unaligned access and only do it the complex way if necessary.  */
 512   Elf64_Addr reloc_addr_val;
 513
 514   /* Load value without causing unaligned trap. */
 515   memcpy (&reloc_addr_val, reloc_addr_arg, 8);
 516   reloc_addr_val += l_addr;
 517
 518   /* Store value without causing unaligned trap. */
 519   memcpy (reloc_addr_arg, &reloc_addr_val, 8);
 520 }
 521
 522 auto inline void
 523 elf_machine_lazy_rel (struct link_map *map,
 524                       Elf64_Addr l_addr, const Elf64_Rela *reloc)
 525 {
 526   Elf64_Addr * const reloc_addr = (void *)(l_addr + reloc->r_offset);
 527   unsigned long int const r_type = ELF64_R_TYPE (reloc->r_info);
 528
 529   if (r_type == R_ALPHA_JMP_SLOT)
 530     {
 531       /* Perform a RELATIVE reloc on the .got entry that transfers
 532          to the .plt.  */
 533       *reloc_addr += l_addr;
 534     }
 535   else if (r_type == R_ALPHA_NONE)
 536     return;
 537   else
 538     _dl_reloc_bad_type (map, r_type, 1);
 539 }
 540
 541 #endif /* RESOLVE_MAP */