scripts/recordmcount.c

   1 /*
   2  * recordmcount.c: construct a table of the locations of calls to 'mcount'
   3  * so that ftrace can find them quickly.
   4  * Copyright 2009 John F. Reiser <jreiser@BitWagon.com>.  All rights reserved.
   5  * Licensed under the GNU General Public License, version 2 (GPLv2).
   6  *
   7  * Restructured to fit Linux format, as well as other updates:
   8  *  Copyright 2010 Steven Rostedt <srostedt@redhat.com>, Red Hat Inc.
   9  */
  10
  11 /*
  12  * Strategy: alter the .o file in-place.
  13  *
  14  * Append a new STRTAB that has the new section names, followed by a new array
  15  * ElfXX_Shdr[] that has the new section headers, followed by the section
  16  * contents for __mcount_loc and its relocations.  The old shstrtab strings,
  17  * and the old ElfXX_Shdr[] array, remain as "garbage" (commonly, a couple
  18  * kilobytes.)  Subsequent processing by /bin/ld (or the kernel module loader)
  19  * will ignore the garbage regions, because they are not designated by the
  20  * new .e_shoff nor the new ElfXX_Shdr[].  [In order to remove the garbage,
  21  * then use "ld -r" to create a new file that omits the garbage.]
  22  */
  23
  24 #include <sys/types.h>
  25 #include <sys/mman.h>
  26 #include <sys/stat.h>
  27 #include <getopt.h>
  28 #include <elf.h>
  29 #include <fcntl.h>
  30 #include <setjmp.h>
  31 #include <stdio.h>
  32 #include <stdlib.h>
  33 #include <string.h>
  34 #include <unistd.h>
  35
  36 /*
  37  * glibc synced up and added the metag number but didn't add the relocations.
  38  * Work around this in a crude manner for now.
  39  */
  40 #ifndef EM_METAG
  41 #define EM_METAG      174
  42 #endif
  43 #ifndef R_METAG_ADDR32
  44 #define R_METAG_ADDR32                   2
  45 #endif
  46 #ifndef R_METAG_NONE
  47 #define R_METAG_NONE                     3
  48 #endif
  49
  50 #ifndef EM_AARCH64
  51 #define EM_AARCH64      183
  52 #define R_AARCH64_NONE          0
  53 #define R_AARCH64_ABS64 257
  54 #endif
  55
  56 static int fd_map;      /* File descriptor for file being modified. */
  57 static int mmap_failed; /* Boolean flag. */
  58 static void *ehdr_curr; /* current ElfXX_Ehdr *  for resource cleanup */
  59 static char gpfx;       /* prefix for global symbol name (sometimes '_') */
  60 static struct stat sb;  /* Remember .st_size, etc. */
  61 static jmp_buf jmpenv;  /* setjmp/longjmp per-file error escape */
  62 static const char *altmcount;   /* alternate mcount symbol name */
  63 static int warn_on_notrace_sect; /* warn when section has mcount not being recorded */
  64
  65 /* setjmp() return values */
  66 enum {
  67         SJ_SETJMP = 0,  /* hardwired first return */
  68         SJ_FAIL,
  69         SJ_SUCCEED
  70 };
  71
  72 /* Per-file resource cleanup when multiple files. */
  73 static void
  74 cleanup(void)
  75 {
  76         if (!mmap_failed)
  77                 munmap(ehdr_curr, sb.st_size);
  78         else
  79                 free(ehdr_curr);
  80         close(fd_map);
  81 }
  82
  83 static void __attribute__((noreturn))
  84 fail_file(void)
  85 {
  86         cleanup();
  87         longjmp(jmpenv, SJ_FAIL);
  88 }
  89
  90 static void __attribute__((noreturn))
  91 succeed_file(void)
  92 {
  93         cleanup();
  94         longjmp(jmpenv, SJ_SUCCEED);
  95 }
  96
  97 /* ulseek, uread, ...:  Check return value for errors. */
  98
  99 static off_t
 100 ulseek(int const fd, off_t const offset, int const whence)
 101 {
 102         off_t const w = lseek(fd, offset, whence);
 103         if (w == (off_t)-1) {
 104                 perror("lseek");
 105                 fail_file();
 106         }
 107         return w;
 108 }
 109
 110 static size_t
 111 uread(int const fd, void *const buf, size_t const count)
 112 {
 113         size_t const n = read(fd, buf, count);
 114         if (n != count) {
 115                 perror("read");
 116                 fail_file();
 117         }
 118         return n;
 119 }
 120
 121 static size_t
 122 uwrite(int const fd, void const *const buf, size_t const count)
 123 {
 124         size_t const n = write(fd, buf, count);
 125         if (n != count) {
 126                 perror("write");
 127                 fail_file();
 128         }
 129         return n;
 130 }
 131
 132 static void *
 133 umalloc(size_t size)
 134 {
 135         void *const addr = malloc(size);
 136         if (addr == 0) {
 137                 fprintf(stderr, "malloc failed: %zu bytes\n", size);
 138                 fail_file();
 139         }
 140         return addr;
 141 }
 142
 143 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
 144 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
 145 static unsigned char *ideal_nop;
 146
 147 static char rel_type_nop;
 148
 149 static int (*make_nop)(void *map, size_t const offset);
 150
 151 static int make_nop_x86(void *map, size_t const offset)
 152 {
 153         uint32_t *ptr;
 154         unsigned char *op;
 155
 156         /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
 157         ptr = map + offset;
 158         if (*ptr != 0)
 159                 return -1;
 160
 161         op = map + offset - 1;
 162         if (*op != 0xe8)
 163                 return -1;
 164
 165         /* convert to nop */
 166         ulseek(fd_map, offset - 1, SEEK_SET);
 167         uwrite(fd_map, ideal_nop, 5);
 168         return 0;
 169 }
 170
 171 static unsigned char ideal_nop4_arm64[4] = {0x1f, 0x20, 0x03, 0xd5};
 172 static int make_nop_arm64(void *map, size_t const offset)
 173 {
 174         uint32_t *ptr;
 175
 176         ptr = map + offset;
 177         /* bl <_mcount> is 0x94000000 before relocation */
 178         if (*ptr != 0x94000000)
 179                 return -1;
 180
 181         /* Convert to nop */
 182         ulseek(fd_map, offset, SEEK_SET);
 183         uwrite(fd_map, ideal_nop, 4);
 184         return 0;
 185 }
 186
 187 /*
 188  * Get the whole file as a programming convenience in order to avoid
 189  * malloc+lseek+read+free of many pieces.  If successful, then mmap
 190  * avoids copying unused pieces; else just read the whole file.
 191  * Open for both read and write; new info will be appended to the file.
 192  * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
 193  * do not propagate to the file until an explicit overwrite at the last.
 194  * This preserves most aspects of consistency (all except .st_size)
 195  * for simultaneous readers of the file while we are appending to it.
 196  * However, multiple writers still are bad.  We choose not to use
 197  * locking because it is expensive and the use case of kernel build
 198  * makes multiple writers unlikely.
 199  */
 200 static void *mmap_file(char const *fname)
 201 {
 202         void *addr;
 203
 204         fd_map = open(fname, O_RDWR);
 205         if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
 206                 perror(fname);
 207                 fail_file();
 208         }
 209         if (!S_ISREG(sb.st_mode)) {
 210                 fprintf(stderr, "not a regular file: %s\n", fname);
 211                 fail_file();
 212         }
 213         addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
 214                     fd_map, 0);
 215         mmap_failed = 0;
 216         if (addr == MAP_FAILED) {
 217                 mmap_failed = 1;
 218                 addr = umalloc(sb.st_size);
 219                 uread(fd_map, addr, sb.st_size);
 220         }
 221         return addr;
 222 }
 223
 224 /* w8rev, w8nat, ...: Handle endianness. */
 225
 226 static uint64_t w8rev(uint64_t const x)
 227 {
 228         return   ((0xff & (x >> (0 * 8))) << (7 * 8))
 229                | ((0xff & (x >> (1 * 8))) << (6 * 8))
 230                | ((0xff & (x >> (2 * 8))) << (5 * 8))
 231                | ((0xff & (x >> (3 * 8))) << (4 * 8))
 232                | ((0xff & (x >> (4 * 8))) << (3 * 8))
 233                | ((0xff & (x >> (5 * 8))) << (2 * 8))
 234                | ((0xff & (x >> (6 * 8))) << (1 * 8))
 235                | ((0xff & (x >> (7 * 8))) << (0 * 8));
 236 }
 237
 238 static uint32_t w4rev(uint32_t const x)
 239 {
 240         return   ((0xff & (x >> (0 * 8))) << (3 * 8))
 241                | ((0xff & (x >> (1 * 8))) << (2 * 8))
 242                | ((0xff & (x >> (2 * 8))) << (1 * 8))
 243                | ((0xff & (x >> (3 * 8))) << (0 * 8));
 244 }
 245
 246 static uint32_t w2rev(uint16_t const x)
 247 {
 248         return   ((0xff & (x >> (0 * 8))) << (1 * 8))
 249                | ((0xff & (x >> (1 * 8))) << (0 * 8));
 250 }
 251
 252 static uint64_t w8nat(uint64_t const x)
 253 {
 254         return x;
 255 }
 256
 257 static uint32_t w4nat(uint32_t const x)
 258 {
 259         return x;
 260 }
 261
 262 static uint32_t w2nat(uint16_t const x)
 263 {
 264         return x;
 265 }
 266
 267 static uint64_t (*w8)(uint64_t);
 268 static uint32_t (*w)(uint32_t);
 269 static uint32_t (*w2)(uint16_t);
 270
 271 /* Names of the sections that could contain calls to mcount. */
 272 static int
 273 is_mcounted_section_name(char const *const txtname)
 274 {
 275         return strcmp(".text",           txtname) == 0 ||
 276                 strcmp(".ref.text",      txtname) == 0 ||
 277                 strcmp(".sched.text",    txtname) == 0 ||
 278                 strcmp(".spinlock.text", txtname) == 0 ||
 279                 strcmp(".irqentry.text", txtname) == 0 ||
 280                 strcmp(".kprobes.text", txtname) == 0 ||
 281                 strcmp(".text.unlikely", txtname) == 0;
 282 }
 283
 284 /* 32 bit and 64 bit are very similar */
 285 #include "recordmcount.h"
 286 #define RECORD_MCOUNT_64
 287 #include "recordmcount.h"
 288
 289 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
 290  * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
 291  * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
 292  * to imply the order of the members; the spec does not say so.
 293  *      typedef unsigned char Elf64_Byte;
 294  * fails on MIPS64 because their <elf.h> already has it!
 295  */
 296
 297 typedef uint8_t myElf64_Byte;           /* Type for a 8-bit quantity.  */
 298
 299 union mips_r_info {
 300         Elf64_Xword r_info;
 301         struct {
 302                 Elf64_Word r_sym;               /* Symbol index.  */
 303                 myElf64_Byte r_ssym;            /* Special symbol.  */
 304                 myElf64_Byte r_type3;           /* Third relocation.  */
 305                 myElf64_Byte r_type2;           /* Second relocation.  */
 306                 myElf64_Byte r_type;            /* First relocation.  */
 307         } r_mips;
 308 };
 309
 310 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
 311 {
 312         return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
 313 }
 314
 315 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
 316 {
 317         rp->r_info = ((union mips_r_info){
 318                 .r_mips = { .r_sym = w(sym), .r_type = type }
 319         }).r_info;
 320 }
 321
 322 static void
 323 do_file(char const *const fname)
 324 {
 325         Elf32_Ehdr *const ehdr = mmap_file(fname);
 326         unsigned int reltype = 0;
 327
 328         ehdr_curr = ehdr;
 329         w = w4nat;
 330         w2 = w2nat;
 331         w8 = w8nat;
 332         switch (ehdr->e_ident[EI_DATA]) {
 333                 static unsigned int const endian = 1;
 334         default:
 335                 fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
 336                         ehdr->e_ident[EI_DATA], fname);
 337                 fail_file();
 338                 break;
 339         case ELFDATA2LSB:
 340                 if (*(unsigned char const *)&endian != 1) {
 341                         /* main() is big endian, file.o is little endian. */
 342                         w = w4rev;
 343                         w2 = w2rev;
 344                         w8 = w8rev;
 345                 }
 346                 break;
 347         case ELFDATA2MSB:
 348                 if (*(unsigned char const *)&endian != 0) {
 349                         /* main() is little endian, file.o is big endian. */
 350                         w = w4rev;
 351                         w2 = w2rev;
 352                         w8 = w8rev;
 353                 }
 354                 break;
 355         }  /* end switch */
 356         if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
 357         ||  w2(ehdr->e_type) != ET_REL
 358         ||  ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
 359                 fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
 360                 fail_file();
 361         }
 362
 363         gpfx = 0;
 364         switch (w2(ehdr->e_machine)) {
 365         default:
 366                 fprintf(stderr, "unrecognized e_machine %d %s\n",
 367                         w2(ehdr->e_machine), fname);
 368                 fail_file();
 369                 break;
 370         case EM_386:
 371                 reltype = R_386_32;
 372                 make_nop = make_nop_x86;
 373                 ideal_nop = ideal_nop5_x86_32;
 374                 mcount_adjust_32 = -1;
 375                 break;
 376         case EM_ARM:     reltype = R_ARM_ABS32;
 377                          altmcount = "__gnu_mcount_nc";
 378                          break;
 379         case EM_AARCH64:
 380                         reltype = R_AARCH64_ABS64;
 381                         make_nop = make_nop_arm64;
 382                         rel_type_nop = R_AARCH64_NONE;
 383                         ideal_nop = ideal_nop4_arm64;
 384                         gpfx = '_';
 385                         break;
 386         case EM_IA_64:   reltype = R_IA64_IMM64;   gpfx = '_'; break;
 387         case EM_METAG:   reltype = R_METAG_ADDR32;
 388                          altmcount = "_mcount_wrapper";
 389                          rel_type_nop = R_METAG_NONE;
 390                          /* We happen to have the same requirement as MIPS */
 391                          is_fake_mcount32 = MIPS32_is_fake_mcount;
 392                          break;
 393         case EM_MIPS:    /* reltype: e_class    */ gpfx = '_'; break;
 394         case EM_PPC:     reltype = R_PPC_ADDR32;   gpfx = '_'; break;
 395         case EM_PPC64:   reltype = R_PPC64_ADDR64; gpfx = '_'; break;
 396         case EM_S390:    /* reltype: e_class    */ gpfx = '_'; break;
 397         case EM_SH:      reltype = R_SH_DIR32;                 break;
 398         case EM_SPARCV9: reltype = R_SPARC_64;     gpfx = '_'; break;
 399         case EM_X86_64:
 400                 make_nop = make_nop_x86;
 401                 ideal_nop = ideal_nop5_x86_64;
 402                 reltype = R_X86_64_64;
 403                 mcount_adjust_64 = -1;
 404                 break;
 405         }  /* end switch */
 406
 407         switch (ehdr->e_ident[EI_CLASS]) {
 408         default:
 409                 fprintf(stderr, "unrecognized ELF class %d %s\n",
 410                         ehdr->e_ident[EI_CLASS], fname);
 411                 fail_file();
 412                 break;
 413         case ELFCLASS32:
 414                 if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
 415                 ||  w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
 416                         fprintf(stderr,
 417                                 "unrecognized ET_REL file: %s\n", fname);
 418                         fail_file();
 419                 }
 420                 if (w2(ehdr->e_machine) == EM_MIPS) {
 421                         reltype = R_MIPS_32;
 422                         is_fake_mcount32 = MIPS32_is_fake_mcount;
 423                 }
 424                 do32(ehdr, fname, reltype);
 425                 break;
 426         case ELFCLASS64: {
 427                 Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
 428                 if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
 429                 ||  w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
 430                         fprintf(stderr,
 431                                 "unrecognized ET_REL file: %s\n", fname);
 432                         fail_file();
 433                 }
 434                 if (w2(ghdr->e_machine) == EM_S390) {
 435                         reltype = R_390_64;
 436                         mcount_adjust_64 = -14;
 437                 }
 438                 if (w2(ghdr->e_machine) == EM_MIPS) {
 439                         reltype = R_MIPS_64;
 440                         Elf64_r_sym = MIPS64_r_sym;
 441                         Elf64_r_info = MIPS64_r_info;
 442                         is_fake_mcount64 = MIPS64_is_fake_mcount;
 443                 }
 444                 do64(ghdr, fname, reltype);
 445                 break;
 446         }
 447         }  /* end switch */
 448
 449         cleanup();
 450 }
 451
 452 int
 453 main(int argc, char *argv[])
 454 {
 455         const char ftrace[] = "/ftrace.o";
 456         int ftrace_size = sizeof(ftrace) - 1;
 457         int n_error = 0;  /* gcc-4.3.0 false positive complaint */
 458         int c;
 459         int i;
 460
 461         while ((c = getopt(argc, argv, "w")) >= 0) {
 462                 switch (c) {
 463                 case 'w':
 464                         warn_on_notrace_sect = 1;
 465                         break;
 466                 default:
 467                         fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
 468                         return 0;
 469                 }
 470         }
 471
 472         if ((argc - optind) < 1) {
 473                 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
 474                 return 0;
 475         }
 476
 477         /* Process each file in turn, allowing deep failure. */
 478         for (i = optind; i < argc; i++) {
 479                 char *file = argv[i];
 480                 int const sjval = setjmp(jmpenv);
 481                 int len;
 482
 483                 /*
 484                  * The file kernel/trace/ftrace.o references the mcount
 485                  * function but does not call it. Since ftrace.o should
 486                  * not be traced anyway, we just skip it.
 487                  */
 488                 len = strlen(file);
 489                 if (len >= ftrace_size &&
 490                     strcmp(file + (len - ftrace_size), ftrace) == 0)
 491                         continue;
 492
 493                 switch (sjval) {
 494                 default:
 495                         fprintf(stderr, "internal error: %s\n", file);
 496                         exit(1);
 497                         break;
 498                 case SJ_SETJMP:    /* normal sequence */
 499                         /* Avoid problems if early cleanup() */
 500                         fd_map = -1;
 501                         ehdr_curr = NULL;
 502                         mmap_failed = 1;
 503                         do_file(file);
 504                         break;
 505                 case SJ_FAIL:    /* error in do_file or below */
 506                         ++n_error;
 507                         break;
 508                 case SJ_SUCCEED:    /* premature success */
 509                         /* do nothing */
 510                         break;
 511                 }  /* end switch */
 512         }
 513         return !!n_error;
 514 }