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_ABS64 257
  53 #endif
  54
  55 static int fd_map;      /* File descriptor for file being modified. */
  56 static int mmap_failed; /* Boolean flag. */
  57 static void *ehdr_curr; /* current ElfXX_Ehdr *  for resource cleanup */
  58 static char gpfx;       /* prefix for global symbol name (sometimes '_') */
  59 static struct stat sb;  /* Remember .st_size, etc. */
  60 static jmp_buf jmpenv;  /* setjmp/longjmp per-file error escape */
  61 static const char *altmcount;   /* alternate mcount symbol name */
  62 static int warn_on_notrace_sect; /* warn when section has mcount not being recorded */
  63
  64 /* setjmp() return values */
  65 enum {
  66         SJ_SETJMP = 0,  /* hardwired first return */
  67         SJ_FAIL,
  68         SJ_SUCCEED
  69 };
  70
  71 /* Per-file resource cleanup when multiple files. */
  72 static void
  73 cleanup(void)
  74 {
  75         if (!mmap_failed)
  76                 munmap(ehdr_curr, sb.st_size);
  77         else
  78                 free(ehdr_curr);
  79         close(fd_map);
  80 }
  81
  82 static void __attribute__((noreturn))
  83 fail_file(void)
  84 {
  85         cleanup();
  86         longjmp(jmpenv, SJ_FAIL);
  87 }
  88
  89 static void __attribute__((noreturn))
  90 succeed_file(void)
  91 {
  92         cleanup();
  93         longjmp(jmpenv, SJ_SUCCEED);
  94 }
  95
  96 /* ulseek, uread, ...:  Check return value for errors. */
  97
  98 static off_t
  99 ulseek(int const fd, off_t const offset, int const whence)
 100 {
 101         off_t const w = lseek(fd, offset, whence);
 102         if (w == (off_t)-1) {
 103                 perror("lseek");
 104                 fail_file();
 105         }
 106         return w;
 107 }
 108
 109 static size_t
 110 uread(int const fd, void *const buf, size_t const count)
 111 {
 112         size_t const n = read(fd, buf, count);
 113         if (n != count) {
 114                 perror("read");
 115                 fail_file();
 116         }
 117         return n;
 118 }
 119
 120 static size_t
 121 uwrite(int const fd, void const *const buf, size_t const count)
 122 {
 123         size_t const n = write(fd, buf, count);
 124         if (n != count) {
 125                 perror("write");
 126                 fail_file();
 127         }
 128         return n;
 129 }
 130
 131 static void *
 132 umalloc(size_t size)
 133 {
 134         void *const addr = malloc(size);
 135         if (addr == 0) {
 136                 fprintf(stderr, "malloc failed: %zu bytes\n", size);
 137                 fail_file();
 138         }
 139         return addr;
 140 }
 141
 142 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
 143 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
 144 static unsigned char *ideal_nop;
 145
 146 static char rel_type_nop;
 147
 148 static int (*make_nop)(void *map, size_t const offset);
 149
 150 static int make_nop_x86(void *map, size_t const offset)
 151 {
 152         uint32_t *ptr;
 153         unsigned char *op;
 154
 155         /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
 156         ptr = map + offset;
 157         if (*ptr != 0)
 158                 return -1;
 159
 160         op = map + offset - 1;
 161         if (*op != 0xe8)
 162                 return -1;
 163
 164         /* convert to nop */
 165         ulseek(fd_map, offset - 1, SEEK_SET);
 166         uwrite(fd_map, ideal_nop, 5);
 167         return 0;
 168 }
 169
 170 /*
 171  * Get the whole file as a programming convenience in order to avoid
 172  * malloc+lseek+read+free of many pieces.  If successful, then mmap
 173  * avoids copying unused pieces; else just read the whole file.
 174  * Open for both read and write; new info will be appended to the file.
 175  * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
 176  * do not propagate to the file until an explicit overwrite at the last.
 177  * This preserves most aspects of consistency (all except .st_size)
 178  * for simultaneous readers of the file while we are appending to it.
 179  * However, multiple writers still are bad.  We choose not to use
 180  * locking because it is expensive and the use case of kernel build
 181  * makes multiple writers unlikely.
 182  */
 183 static void *mmap_file(char const *fname)
 184 {
 185         void *addr;
 186
 187         fd_map = open(fname, O_RDWR);
 188         if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
 189                 perror(fname);
 190                 fail_file();
 191         }
 192         if (!S_ISREG(sb.st_mode)) {
 193                 fprintf(stderr, "not a regular file: %s\n", fname);
 194                 fail_file();
 195         }
 196         addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
 197                     fd_map, 0);
 198         mmap_failed = 0;
 199         if (addr == MAP_FAILED) {
 200                 mmap_failed = 1;
 201                 addr = umalloc(sb.st_size);
 202                 uread(fd_map, addr, sb.st_size);
 203         }
 204         if (sb.st_nlink != 1) {
 205                 /* file is hard-linked, break the hard link */
 206                 close(fd_map);
 207                 if (unlink(fname) < 0) {
 208                         perror(fname);
 209                         fail_file();
 210                 }
 211                 fd_map = open(fname, O_RDWR | O_CREAT, sb.st_mode);
 212                 if (fd_map < 0) {
 213                         perror(fname);
 214                         fail_file();
 215                 }
 216                 uwrite(fd_map, addr, sb.st_size);
 217         }
 218         return addr;
 219 }
 220
 221 /* w8rev, w8nat, ...: Handle endianness. */
 222
 223 static uint64_t w8rev(uint64_t const x)
 224 {
 225         return   ((0xff & (x >> (0 * 8))) << (7 * 8))
 226                | ((0xff & (x >> (1 * 8))) << (6 * 8))
 227                | ((0xff & (x >> (2 * 8))) << (5 * 8))
 228                | ((0xff & (x >> (3 * 8))) << (4 * 8))
 229                | ((0xff & (x >> (4 * 8))) << (3 * 8))
 230                | ((0xff & (x >> (5 * 8))) << (2 * 8))
 231                | ((0xff & (x >> (6 * 8))) << (1 * 8))
 232                | ((0xff & (x >> (7 * 8))) << (0 * 8));
 233 }
 234
 235 static uint32_t w4rev(uint32_t const x)
 236 {
 237         return   ((0xff & (x >> (0 * 8))) << (3 * 8))
 238                | ((0xff & (x >> (1 * 8))) << (2 * 8))
 239                | ((0xff & (x >> (2 * 8))) << (1 * 8))
 240                | ((0xff & (x >> (3 * 8))) << (0 * 8));
 241 }
 242
 243 static uint32_t w2rev(uint16_t const x)
 244 {
 245         return   ((0xff & (x >> (0 * 8))) << (1 * 8))
 246                | ((0xff & (x >> (1 * 8))) << (0 * 8));
 247 }
 248
 249 static uint64_t w8nat(uint64_t const x)
 250 {
 251         return x;
 252 }
 253
 254 static uint32_t w4nat(uint32_t const x)
 255 {
 256         return x;
 257 }
 258
 259 static uint32_t w2nat(uint16_t const x)
 260 {
 261         return x;
 262 }
 263
 264 static uint64_t (*w8)(uint64_t);
 265 static uint32_t (*w)(uint32_t);
 266 static uint32_t (*w2)(uint16_t);
 267
 268 /* Names of the sections that could contain calls to mcount. */
 269 static int
 270 is_mcounted_section_name(char const *const txtname)
 271 {
 272         return strcmp(".text",           txtname) == 0 ||
 273                 strcmp(".ref.text",      txtname) == 0 ||
 274                 strcmp(".sched.text",    txtname) == 0 ||
 275                 strcmp(".spinlock.text", txtname) == 0 ||
 276                 strcmp(".irqentry.text", txtname) == 0 ||
 277                 strcmp(".kprobes.text", txtname) == 0 ||
 278                 strcmp(".text.unlikely", txtname) == 0;
 279 }
 280
 281 /* 32 bit and 64 bit are very similar */
 282 #include "recordmcount.h"
 283 #define RECORD_MCOUNT_64
 284 #include "recordmcount.h"
 285
 286 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
 287  * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
 288  * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
 289  * to imply the order of the members; the spec does not say so.
 290  *      typedef unsigned char Elf64_Byte;
 291  * fails on MIPS64 because their <elf.h> already has it!
 292  */
 293
 294 typedef uint8_t myElf64_Byte;           /* Type for a 8-bit quantity.  */
 295
 296 union mips_r_info {
 297         Elf64_Xword r_info;
 298         struct {
 299                 Elf64_Word r_sym;               /* Symbol index.  */
 300                 myElf64_Byte r_ssym;            /* Special symbol.  */
 301                 myElf64_Byte r_type3;           /* Third relocation.  */
 302                 myElf64_Byte r_type2;           /* Second relocation.  */
 303                 myElf64_Byte r_type;            /* First relocation.  */
 304         } r_mips;
 305 };
 306
 307 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
 308 {
 309         return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
 310 }
 311
 312 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
 313 {
 314         rp->r_info = ((union mips_r_info){
 315                 .r_mips = { .r_sym = w(sym), .r_type = type }
 316         }).r_info;
 317 }
 318
 319 static void
 320 do_file(char const *const fname)
 321 {
 322         Elf32_Ehdr *const ehdr = mmap_file(fname);
 323         unsigned int reltype = 0;
 324
 325         ehdr_curr = ehdr;
 326         w = w4nat;
 327         w2 = w2nat;
 328         w8 = w8nat;
 329         switch (ehdr->e_ident[EI_DATA]) {
 330                 static unsigned int const endian = 1;
 331         default:
 332                 fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
 333                         ehdr->e_ident[EI_DATA], fname);
 334                 fail_file();
 335                 break;
 336         case ELFDATA2LSB:
 337                 if (*(unsigned char const *)&endian != 1) {
 338                         /* main() is big endian, file.o is little endian. */
 339                         w = w4rev;
 340                         w2 = w2rev;
 341                         w8 = w8rev;
 342                 }
 343                 break;
 344         case ELFDATA2MSB:
 345                 if (*(unsigned char const *)&endian != 0) {
 346                         /* main() is little endian, file.o is big endian. */
 347                         w = w4rev;
 348                         w2 = w2rev;
 349                         w8 = w8rev;
 350                 }
 351                 break;
 352         }  /* end switch */
 353         if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
 354         ||  w2(ehdr->e_type) != ET_REL
 355         ||  ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
 356                 fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
 357                 fail_file();
 358         }
 359
 360         gpfx = 0;
 361         switch (w2(ehdr->e_machine)) {
 362         default:
 363                 fprintf(stderr, "unrecognized e_machine %d %s\n",
 364                         w2(ehdr->e_machine), fname);
 365                 fail_file();
 366                 break;
 367         case EM_386:
 368                 reltype = R_386_32;
 369                 make_nop = make_nop_x86;
 370                 ideal_nop = ideal_nop5_x86_32;
 371                 mcount_adjust_32 = -1;
 372                 break;
 373         case EM_ARM:     reltype = R_ARM_ABS32;
 374                          altmcount = "__gnu_mcount_nc";
 375                          break;
 376         case EM_AARCH64:
 377                          reltype = R_AARCH64_ABS64; gpfx = '_'; break;
 378         case EM_IA_64:   reltype = R_IA64_IMM64;   gpfx = '_'; break;
 379         case EM_METAG:   reltype = R_METAG_ADDR32;
 380                          altmcount = "_mcount_wrapper";
 381                          rel_type_nop = R_METAG_NONE;
 382                          /* We happen to have the same requirement as MIPS */
 383                          is_fake_mcount32 = MIPS32_is_fake_mcount;
 384                          break;
 385         case EM_MIPS:    /* reltype: e_class    */ gpfx = '_'; break;
 386         case EM_PPC:     reltype = R_PPC_ADDR32;   gpfx = '_'; break;
 387         case EM_PPC64:   reltype = R_PPC64_ADDR64; gpfx = '_'; break;
 388         case EM_S390:    /* reltype: e_class    */ gpfx = '_'; break;
 389         case EM_SH:      reltype = R_SH_DIR32;                 break;
 390         case EM_SPARCV9: reltype = R_SPARC_64;     gpfx = '_'; break;
 391         case EM_X86_64:
 392                 make_nop = make_nop_x86;
 393                 ideal_nop = ideal_nop5_x86_64;
 394                 reltype = R_X86_64_64;
 395                 mcount_adjust_64 = -1;
 396                 break;
 397         }  /* end switch */
 398
 399         switch (ehdr->e_ident[EI_CLASS]) {
 400         default:
 401                 fprintf(stderr, "unrecognized ELF class %d %s\n",
 402                         ehdr->e_ident[EI_CLASS], fname);
 403                 fail_file();
 404                 break;
 405         case ELFCLASS32:
 406                 if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
 407                 ||  w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
 408                         fprintf(stderr,
 409                                 "unrecognized ET_REL file: %s\n", fname);
 410                         fail_file();
 411                 }
 412                 if (w2(ehdr->e_machine) == EM_S390) {
 413                         reltype = R_390_32;
 414                         mcount_adjust_32 = -4;
 415                 }
 416                 if (w2(ehdr->e_machine) == EM_MIPS) {
 417                         reltype = R_MIPS_32;
 418                         is_fake_mcount32 = MIPS32_is_fake_mcount;
 419                 }
 420                 do32(ehdr, fname, reltype);
 421                 break;
 422         case ELFCLASS64: {
 423                 Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
 424                 if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
 425                 ||  w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
 426                         fprintf(stderr,
 427                                 "unrecognized ET_REL file: %s\n", fname);
 428                         fail_file();
 429                 }
 430                 if (w2(ghdr->e_machine) == EM_S390) {
 431                         reltype = R_390_64;
 432                         mcount_adjust_64 = -8;
 433                 }
 434                 if (w2(ghdr->e_machine) == EM_MIPS) {
 435                         reltype = R_MIPS_64;
 436                         Elf64_r_sym = MIPS64_r_sym;
 437                         Elf64_r_info = MIPS64_r_info;
 438                         is_fake_mcount64 = MIPS64_is_fake_mcount;
 439                 }
 440                 do64(ghdr, fname, reltype);
 441                 break;
 442         }
 443         }  /* end switch */
 444
 445         cleanup();
 446 }
 447
 448 int
 449 main(int argc, char *argv[])
 450 {
 451         const char ftrace[] = "/ftrace.o";
 452         int ftrace_size = sizeof(ftrace) - 1;
 453         int n_error = 0;  /* gcc-4.3.0 false positive complaint */
 454         int c;
 455         int i;
 456
 457         while ((c = getopt(argc, argv, "w")) >= 0) {
 458                 switch (c) {
 459                 case 'w':
 460                         warn_on_notrace_sect = 1;
 461                         break;
 462                 default:
 463                         fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
 464                         return 0;
 465                 }
 466         }
 467
 468         if ((argc - optind) < 1) {
 469                 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
 470                 return 0;
 471         }
 472
 473         /* Process each file in turn, allowing deep failure. */
 474         for (i = optind; i < argc; i++) {
 475                 char *file = argv[i];
 476                 int const sjval = setjmp(jmpenv);
 477                 int len;
 478
 479                 /*
 480                  * The file kernel/trace/ftrace.o references the mcount
 481                  * function but does not call it. Since ftrace.o should
 482                  * not be traced anyway, we just skip it.
 483                  */
 484                 len = strlen(file);
 485                 if (len >= ftrace_size &&
 486                     strcmp(file + (len - ftrace_size), ftrace) == 0)
 487                         continue;
 488
 489                 switch (sjval) {
 490                 default:
 491                         fprintf(stderr, "internal error: %s\n", file);
 492                         exit(1);
 493                         break;
 494                 case SJ_SETJMP:    /* normal sequence */
 495                         /* Avoid problems if early cleanup() */
 496                         fd_map = -1;
 497                         ehdr_curr = NULL;
 498                         mmap_failed = 1;
 499                         do_file(file);
 500                         break;
 501                 case SJ_FAIL:    /* error in do_file or below */
 502                         ++n_error;
 503                         break;
 504                 case SJ_SUCCEED:    /* premature success */
 505                         /* do nothing */
 506                         break;
 507                 }  /* end switch */
 508         }
 509         return !!n_error;
 510 }
 511
 512