deps/lightning/lib/jit_disasm.c

   1 /*
   2  * Copyright (C) 2012-2019  Free Software Foundation, Inc.
   3  *
   4  * This file is part of GNU lightning.
   5  *
   6  * GNU lightning is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU Lesser General Public License as published
   8  * by the Free Software Foundation; either version 3, or (at your option)
   9  * any later version.
  10  *
  11  * GNU lightning is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  13  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
  14  * License for more details.
  15  *
  16  * Authors:
  17  *      Paulo Cesar Pereira de Andrade
  18  */
  19
  20 #include <lightning.h>
  21 #include <lightning/jit_private.h>
  22 #if DISASSEMBLER
  23 #  include <dis-asm.h>
  24 #endif
  25
  26 /*
  27  * Prototypes
  28  */
  29 #if DISASSEMBLER
  30 static int
  31 disasm_compare_symbols(const void *ap, const void *bp);
  32
  33 static void
  34 disasm_print_address(bfd_vma addr, struct disassemble_info *info);
  35
  36 #define disassemble(u, v)       _disassemble(_jit, u, v)
  37 static void
  38 _disassemble(jit_state_t *_jit, jit_pointer_t code, jit_int32_t length);
  39 #endif
  40
  41 /*
  42  * Initialization
  43  */
  44 #if DISASSEMBLER
  45 static bfd                       *disasm_bfd;
  46 static disassemble_info           disasm_info;
  47 static disassembler_ftype         disasm_print;
  48 static asymbol                  **disasm_symbols;
  49 static asymbol                   *disasm_synthetic;
  50 static long                       disasm_num_symbols;
  51 static long                       disasm_num_synthetic;
  52 static jit_state_t               *disasm_jit;
  53 #define disasm_stream             stdout
  54 #endif
  55
  56 /*
  57  * Implementation
  58  */
  59 void
  60 jit_init_debug(const char *progname)
  61 {
  62 #if DISASSEMBLER
  63     bfd_init();
  64
  65     if (progname)
  66         disasm_bfd = bfd_openr(progname, NULL);
  67     if (disasm_bfd == NULL) {
  68 #if defined(__linux__)
  69         disasm_bfd = bfd_openr("/proc/self/exe", NULL);
  70         if (disasm_bfd == NULL)
  71 #endif
  72             return;
  73     }
  74     bfd_check_format(disasm_bfd, bfd_object);
  75     bfd_check_format(disasm_bfd, bfd_archive);
  76     INIT_DISASSEMBLE_INFO(disasm_info, disasm_stream, fprintf);
  77 #  if defined(__i386__) || defined(__x86_64__)
  78     disasm_info.arch = bfd_arch_i386;
  79 #    if defined(__x86_64__)
  80 #      if __WORDSIZE == 32
  81     disasm_info.mach = bfd_mach_x64_32;
  82 #      else
  83     disasm_info.mach = bfd_mach_x86_64;
  84 #      endif
  85 #    else
  86     disasm_info.mach = bfd_mach_i386_i386;
  87 #    endif
  88 #  endif
  89 #  if defined(__powerpc__)
  90     disasm_info.arch = bfd_arch_powerpc;
  91     disasm_info.mach = bfd_mach_ppc64;
  92 #    if HAVE_DISASSEMBLE_INIT_FOR_TARGET
  93     disassemble_init_for_target(&disasm_info);
  94 #    elif HAVE_DISASSEMBLE_INIT_POWERPC
  95     disassemble_init_powerpc(&disasm_info);
  96 #    endif
  97 #    if defined(__powerpc64__)
  98     disasm_info.disassembler_options = "64";
  99 #    endif
 100 #    if HAVE_DISASSEMBLE_INIT_FOR_TARGET
 101     disassemble_init_for_target(&disasm_info);
 102 #    elif HAVE_DISASSEMBLE_INIT_POWERPC
 103     disassemble_init_powerpc(&disasm_info);
 104 #    endif
 105 #  endif
 106 #  if defined(__sparc__)
 107     disasm_info.endian = disasm_info.display_endian = BFD_ENDIAN_BIG;
 108 #  endif
 109 #  if defined(__s390__) || defined(__s390x__)
 110     disasm_info.arch = bfd_arch_s390;
 111 #    if __WORDSIZE == 32
 112     disasm_info.mach = bfd_mach_s390_31;
 113 #    else
 114     disasm_info.mach = bfd_mach_s390_64;
 115 #    endif
 116     disasm_info.endian = disasm_info.display_endian = BFD_ENDIAN_BIG;
 117     disasm_info.disassembler_options = "zarch";
 118 #  endif
 119 #  if defined(__alpha__)
 120     disasm_info.arch = bfd_arch_alpha;
 121     disasm_info.mach = bfd_mach_alpha_ev6;
 122 #  endif
 123 #  if defined(__hppa__)
 124     disasm_info.arch = bfd_arch_hppa;
 125     disasm_info.mach = bfd_mach_hppa10;
 126 #  endif
 127 #  if defined(__riscv)
 128     disasm_info.arch = bfd_arch_riscv;
 129 #  if __WORDSIZE == 32
 130     disasm_info.mach = bfd_mach_riscv32;
 131 #  else
 132     disasm_info.mach = bfd_mach_riscv64;
 133 #  endif
 134 #  endif
 135     disasm_info.print_address_func = disasm_print_address;
 136
 137 # if BINUTILS_2_29
 138     disasm_print = disassembler(disasm_info.arch, __BYTE_ORDER == __BIG_ENDIAN,
 139                                 disasm_info.mach, disasm_bfd);
 140 #  else
 141     disasm_print = disassembler(disasm_bfd);
 142 #  endif
 143     assert(disasm_print);
 144
 145     if (bfd_get_file_flags(disasm_bfd) & HAS_SYMS) {
 146         asymbol         **in;
 147         asymbol         **out;
 148         asymbol          *symbol;
 149         long              offset;
 150         long              sym_count;
 151         long              dyn_count;
 152         long              sym_storage;
 153         long              dyn_storage;
 154
 155         if ((sym_storage = bfd_get_symtab_upper_bound(disasm_bfd)) >= 0) {
 156
 157             if (bfd_get_file_flags(disasm_bfd) & DYNAMIC) {
 158                 dyn_storage = bfd_get_dynamic_symtab_upper_bound(disasm_bfd);
 159 #  if defined(__alpha__)
 160                 /* XXX */
 161                 if (dyn_storage < 0)
 162                     dyn_storage = 0;
 163 #  else
 164                 assert(dyn_storage >= 0);
 165 #  endif
 166             }
 167             else
 168                 dyn_storage = 0;
 169
 170             jit_alloc((jit_pointer_t *)&disasm_symbols,
 171                       (sym_storage + dyn_storage) * sizeof(asymbol *));
 172             sym_count = bfd_canonicalize_symtab(disasm_bfd, disasm_symbols);
 173             assert(sym_count >= 0);
 174             if (dyn_storage) {
 175                 dyn_count = bfd_canonicalize_dynamic_symtab(disasm_bfd,
 176                                                             disasm_symbols +
 177                                                             sym_count);
 178                 assert(dyn_count >= 0);
 179             }
 180             else
 181                 dyn_count = 0;
 182             disasm_num_symbols = sym_count + dyn_count;
 183
 184             disasm_num_synthetic = bfd_get_synthetic_symtab(disasm_bfd,
 185                                                             sym_count,
 186                                                             disasm_symbols,
 187                                                             dyn_count,
 188                                                             disasm_symbols +
 189                                                             sym_count,
 190                                                             &disasm_synthetic);
 191             if (disasm_num_synthetic > 0) {
 192                 jit_realloc((jit_pointer_t *)&disasm_symbols,
 193                             (sym_storage + dyn_storage) * sizeof(asymbol *),
 194                             (sym_storage + dyn_storage + disasm_num_synthetic) *
 195                             sizeof(asymbol *));
 196                 for (offset = 0; offset < disasm_num_synthetic; offset++)
 197                     disasm_symbols[disasm_num_symbols++] =
 198                         disasm_synthetic + offset;
 199             }
 200
 201             /* remove symbols not useful for disassemble */
 202             in = out = disasm_symbols;
 203             for (offset = 0; offset < disasm_num_symbols; offset++) {
 204                 symbol = *in++;
 205                 if (symbol->name &&
 206                     symbol->name[0] != '\0' &&
 207                     !(symbol->flags & (BSF_DEBUGGING | BSF_SECTION_SYM)) &&
 208                     !bfd_is_und_section(symbol->section) &&
 209                     !bfd_is_com_section(symbol->section))
 210                     *out++ = symbol;
 211             }
 212             disasm_num_symbols = out - disasm_symbols;
 213             qsort(disasm_symbols, disasm_num_symbols,
 214                   sizeof(asymbol *), disasm_compare_symbols);
 215         }
 216     }
 217 #endif
 218 }
 219
 220 void
 221 jit_finish_debug(void)
 222 {
 223 #if DISASSEMBLER
 224     if (disasm_synthetic)
 225         jit_free((jit_pointer_t *)&disasm_synthetic);
 226     if (disasm_symbols)
 227         jit_free((jit_pointer_t *)&disasm_symbols);
 228     if (disasm_bfd)
 229         bfd_close (disasm_bfd);
 230 #endif
 231 }
 232
 233 void
 234 _jit_disassemble(jit_state_t *_jit)
 235 {
 236 #if DISASSEMBLER
 237     if (disasm_bfd) {
 238 #  if defined(__arm__)
 239         /* FIXME add mapping for prolog switching to arm and possible jump
 240          * before first prolog also in arm mode */
 241         disasm_info.disassembler_options = jit_cpu.thumb ? "force-thumb" : "";
 242 #  endif
 243
 244         disassemble(_jit->code.ptr, _jit->pc.uc - _jit->code.ptr);
 245     }
 246 #endif
 247 }
 248
 249 #if DISASSEMBLER
 250 /* Based on objdump source */
 251 static int
 252 disasm_compare_symbols(const void *ap, const void *bp)
 253 {
 254     const asymbol       *a = *(const asymbol **)ap;
 255     const asymbol       *b = *(const asymbol **)bp;
 256
 257     if (bfd_asymbol_value(a) > bfd_asymbol_value(b))
 258         return (1);
 259     if (bfd_asymbol_value(a) < bfd_asymbol_value(b))
 260         return (-1);
 261     return (0);
 262 }
 263
 264 #if __WORDSIZE == 32
 265 #  define address_buffer_length         16
 266 #  define address_buffer_format         "%llx"
 267 #else
 268 #  define address_buffer_length         32
 269 #  define address_buffer_format         "%lx"
 270 #endif
 271 static void
 272 disasm_print_address(bfd_vma addr, struct disassemble_info *info)
 273 {
 274     char                *name;
 275     char                *file;
 276     int                  line;
 277     char                 buffer[address_buffer_length];
 278
 279     sprintf(buffer, address_buffer_format, (long long)addr);
 280     (*info->fprintf_func)(info->stream, "0x%s", buffer);
 281
 282 #  define _jit                          disasm_jit
 283 #  undef jit_pointer_p
 284 #  define jit_pointer_p(u)                                      \
 285         ((u) >= _jit->code.ptr && (u) < _jit->pc.uc)
 286     if (jit_pointer_p((jit_uint8_t *)(jit_word_t)addr)) {
 287         if (jit_get_note((jit_uint8_t *)(jit_word_t)addr, &name, &file, &line))
 288             (*info->fprintf_func)(info->stream, " %s:%s:%d",
 289                                   name ? name : "",
 290                                   file ? file : "",
 291                                   line);
 292     }
 293 #  undef jit_pointer_p
 294 #  undef _jit
 295     else if (disasm_num_symbols) {
 296         long             low;
 297         long             high;
 298         long             offset;
 299         asymbol         *symbol;
 300
 301         low = 0;
 302         high = disasm_num_symbols;
 303         do {
 304             offset = (low + high) >> 1;
 305             symbol = disasm_symbols[offset];
 306             if (bfd_asymbol_value(symbol) > addr)
 307                 high = offset - 1;
 308             else if (bfd_asymbol_value(symbol) < addr)
 309                 low = offset + 1;
 310             else
 311                 break;
 312         } while (low < high);
 313
 314         if (offset >= 0 && offset < disasm_num_symbols) {
 315             if (bfd_asymbol_value(symbol) < addr) {
 316                 while (++offset < disasm_num_symbols) {
 317                     symbol = disasm_symbols[offset];
 318                     if (bfd_asymbol_value(symbol) >= addr)
 319                         break;
 320                 }
 321             }
 322             else if (bfd_asymbol_value(symbol) > addr) {
 323                 while (offset--) {
 324                     if (bfd_asymbol_value(disasm_symbols[offset]) < addr)
 325                         break;
 326                     symbol = disasm_symbols[offset];
 327                 }
 328             }
 329             if (bfd_asymbol_value(symbol) == addr)
 330                 (*info->fprintf_func)(info->stream, " # %s", symbol->name);
 331         }
 332     }
 333 }
 334
 335 static void
 336 _disassemble(jit_state_t *_jit, jit_pointer_t code, jit_int32_t length)
 337 {
 338     int                  bytes;
 339     char                *name, *old_name;
 340     char                *file, *old_file;
 341     int                  line,  old_line;
 342 #if __arm__
 343     jit_int32_t          offset;
 344     jit_bool_t           data_info;
 345     jit_int32_t          data_offset;
 346 #endif
 347     bfd_vma              pc = (jit_uword_t)code;
 348     bfd_vma              end = (jit_uword_t)code + length;
 349     char                 buffer[address_buffer_length];
 350 #if DEVEL_DISASSEMBLER
 351     jit_node_t          *node;
 352     jit_uword_t          prevw;
 353 #endif
 354
 355 #if __arm__
 356     data_info = _jitc && _jitc->data_info.ptr;
 357     data_offset = 0;
 358 #endif
 359     disasm_info.buffer = code;
 360     disasm_info.buffer_vma = (jit_uword_t)code;
 361     disasm_info.buffer_length = length;
 362     old_file = old_name = NULL;
 363     old_line = 0;
 364     disasm_jit = _jit;
 365 #if DEVEL_DISASSEMBLER
 366     node = _jitc->head;
 367     prevw = pc;
 368 #endif
 369     while (pc < end) {
 370 #if DEVEL_DISASSEMBLER
 371         while (node && (jit_uword_t)(prevw + node->offset) < (jit_uword_t)pc) {
 372             prevw += node->offset;
 373             node = node->next;
 374         }
 375         while (node && (jit_uword_t)(prevw + node->offset) == (jit_uword_t)pc) {
 376             jit_print_node(node);
 377             fputc('\n', stdout);
 378             prevw += node->offset;
 379             node = node->next;
 380         }
 381 #endif
 382 #if __arm__
 383     again:
 384         if (data_info) {
 385             while (_jitc->data_info.ptr[data_offset].code < pc) {
 386                 if (++data_offset >= _jitc->data_info.length) {
 387                     data_info = 0;
 388                     goto again;
 389                 }
 390             }
 391             if (pc == _jitc->data_info.ptr[data_offset].code) {
 392                 offset = _jitc->data_info.ptr[data_offset].length;
 393                 for (; offset >= 4; offset -= 4, pc += 4) {
 394                     bytes = sprintf(buffer, address_buffer_format, pc);
 395                     (*disasm_info.fprintf_func)(disasm_stream,
 396                                                 "%*c0x%s\t.data\t0x%08x\n",
 397                                                 16 - bytes, ' ', buffer,
 398                                                 *(jit_uint32_t *)
 399                                                 (jit_uint32_t)pc);
 400                 }
 401                 /* reset disassemble information instead of attempting
 402                  * to hack the arm specific backend data structures to
 403                  * tell it to forward the required number of bytes. */
 404                 disasm_info.buffer = (jit_pointer_t)(jit_uint32_t)pc;
 405                 disasm_info.buffer_vma = (jit_uword_t)pc;
 406                 if ((disasm_info.buffer_length = end - pc) <= 0)
 407                     break;
 408             }
 409         }
 410 #endif
 411         if (jit_get_note((jit_uint8_t *)(jit_word_t)pc, &name, &file, &line) &&
 412             (name != old_name || file != old_file || line != old_line)) {
 413             (*disasm_info.fprintf_func)(disasm_stream, "# %s:%s:%d\n",
 414                                         name ? name : "",
 415                                         file ? file : "",
 416                                         line);
 417             old_name = name;
 418             old_file = file;
 419             old_line = line;
 420         }
 421
 422         bytes = sprintf(buffer, address_buffer_format, (long long)pc);
 423         (*disasm_info.fprintf_func)(disasm_stream, "%*c0x%s\t",
 424                                     16 - bytes, ' ', buffer);
 425         pc += (*disasm_print)(pc, &disasm_info);
 426         putc('\n', disasm_stream);
 427     }
 428 }
 429 #endif