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 <elf.h>
  28 #include <fcntl.h>
  29 #include <setjmp.h>
  30 #include <stdio.h>
  31 #include <stdlib.h>
  32 #include <string.h>
  33 #include <unistd.h>
  34
  35 static int fd_map;      /* File descriptor for file being modified. */
  36 static int mmap_failed; /* Boolean flag. */
  37 static void *ehdr_curr; /* current ElfXX_Ehdr *  for resource cleanup */
  38 static char gpfx;       /* prefix for global symbol name (sometimes '_') */
  39 static struct stat sb;  /* Remember .st_size, etc. */
  40 static jmp_buf jmpenv;  /* setjmp/longjmp per-file error escape */
  41 static const char *altmcount;   /* alternate mcount symbol name */
  42
  43 /* setjmp() return values */
  44 enum {
  45         SJ_SETJMP = 0,  /* hardwired first return */
  46         SJ_FAIL,
  47         SJ_SUCCEED
  48 };
  49
  50 /* Per-file resource cleanup when multiple files. */
  51 static void
  52 cleanup(void)
  53 {
  54         if (!mmap_failed)
  55                 munmap(ehdr_curr, sb.st_size);
  56         else
  57                 free(ehdr_curr);
  58         close(fd_map);
  59 }
  60
  61 static void __attribute__((noreturn))
  62 fail_file(void)
  63 {
  64         cleanup();
  65         longjmp(jmpenv, SJ_FAIL);
  66 }
  67
  68 static void __attribute__((noreturn))
  69 succeed_file(void)
  70 {
  71         cleanup();
  72         longjmp(jmpenv, SJ_SUCCEED);
  73 }
  74
  75 /* ulseek, uread, ...:  Check return value for errors. */
  76
  77 static off_t
  78 ulseek(int const fd, off_t const offset, int const whence)
  79 {
  80         off_t const w = lseek(fd, offset, whence);
  81         if (w == (off_t)-1) {
  82                 perror("lseek");
  83                 fail_file();
  84         }
  85         return w;
  86 }
  87
  88 static size_t
  89 uread(int const fd, void *const buf, size_t const count)
  90 {
  91         size_t const n = read(fd, buf, count);
  92         if (n != count) {
  93                 perror("read");
  94                 fail_file();
  95         }
  96         return n;
  97 }
  98
  99 static size_t
 100 uwrite(int const fd, void const *const buf, size_t const count)
 101 {
 102         size_t const n = write(fd, buf, count);
 103         if (n != count) {
 104                 perror("write");
 105                 fail_file();
 106         }
 107         return n;
 108 }
 109
 110 static void *
 111 umalloc(size_t size)
 112 {
 113         void *const addr = malloc(size);
 114         if (addr == 0) {
 115                 fprintf(stderr, "malloc failed: %zu bytes\n", size);
 116                 fail_file();
 117         }
 118         return addr;
 119 }
 120
 121 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
 122 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
 123 static unsigned char *ideal_nop;
 124
 125 static char rel_type_nop;
 126
 127 static int (*make_nop)(void *map, size_t const offset);
 128
 129 static int make_nop_x86(void *map, size_t const offset)
 130 {
 131         uint32_t *ptr;
 132         unsigned char *op;
 133
 134         /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
 135         ptr = map + offset;
 136         if (*ptr != 0)
 137                 return -1;
 138
 139         op = map + offset - 1;
 140         if (*op != 0xe8)
 141                 return -1;
 142
 143         /* convert to nop */
 144         ulseek(fd_map, offset - 1, SEEK_SET);
 145         uwrite(fd_map, ideal_nop, 5);
 146         return 0;
 147 }
 148
 149 /*
 150  * Get the whole file as a programming convenience in order to avoid
 151  * malloc+lseek+read+free of many pieces.  If successful, then mmap
 152  * avoids copying unused pieces; else just read the whole file.
 153  * Open for both read and write; new info will be appended to the file.
 154  * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
 155  * do not propagate to the file until an explicit overwrite at the last.
 156  * This preserves most aspects of consistency (all except .st_size)
 157  * for simultaneous readers of the file while we are appending to it.
 158  * However, multiple writers still are bad.  We choose not to use
 159  * locking because it is expensive and the use case of kernel build
 160  * makes multiple writers unlikely.
 161  */
 162 static void *mmap_file(char const *fname)
 163 {
 164         void *addr;
 165
 166         fd_map = open(fname, O_RDWR);
 167         if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
 168                 perror(fname);
 169                 fail_file();
 170         }
 171         if (!S_ISREG(sb.st_mode)) {
 172                 fprintf(stderr, "not a regular file: %s\n", fname);
 173                 fail_file();
 174         }
 175         addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
 176                     fd_map, 0);
 177         mmap_failed = 0;
 178         if (addr == MAP_FAILED) {
 179                 mmap_failed = 1;
 180                 addr = umalloc(sb.st_size);
 181                 uread(fd_map, addr, sb.st_size);
 182         }
 183         return addr;
 184 }
 185
 186 /* w8rev, w8nat, ...: Handle endianness. */
 187
 188 static uint64_t w8rev(uint64_t const x)
 189 {
 190         return   ((0xff & (x >> (0 * 8))) << (7 * 8))
 191                | ((0xff & (x >> (1 * 8))) << (6 * 8))
 192                | ((0xff & (x >> (2 * 8))) << (5 * 8))
 193                | ((0xff & (x >> (3 * 8))) << (4 * 8))
 194                | ((0xff & (x >> (4 * 8))) << (3 * 8))
 195                | ((0xff & (x >> (5 * 8))) << (2 * 8))
 196                | ((0xff & (x >> (6 * 8))) << (1 * 8))
 197                | ((0xff & (x >> (7 * 8))) << (0 * 8));
 198 }
 199
 200 static uint32_t w4rev(uint32_t const x)
 201 {
 202         return   ((0xff & (x >> (0 * 8))) << (3 * 8))
 203                | ((0xff & (x >> (1 * 8))) << (2 * 8))
 204                | ((0xff & (x >> (2 * 8))) << (1 * 8))
 205                | ((0xff & (x >> (3 * 8))) << (0 * 8));
 206 }
 207
 208 static uint32_t w2rev(uint16_t const x)
 209 {
 210         return   ((0xff & (x >> (0 * 8))) << (1 * 8))
 211                | ((0xff & (x >> (1 * 8))) << (0 * 8));
 212 }
 213
 214 static uint64_t w8nat(uint64_t const x)
 215 {
 216         return x;
 217 }
 218
 219 static uint32_t w4nat(uint32_t const x)
 220 {
 221         return x;
 222 }
 223
 224 static uint32_t w2nat(uint16_t const x)
 225 {
 226         return x;
 227 }
 228
 229 static uint64_t (*w8)(uint64_t);
 230 static uint32_t (*w)(uint32_t);
 231 static uint32_t (*w2)(uint16_t);
 232
 233 /* Names of the sections that could contain calls to mcount. */
 234 static int
 235 is_mcounted_section_name(char const *const txtname)
 236 {
 237         return strcmp(".text",           txtname) == 0 ||
 238                 strcmp(".ref.text",      txtname) == 0 ||
 239                 strcmp(".sched.text",    txtname) == 0 ||
 240                 strcmp(".spinlock.text", txtname) == 0 ||
 241                 strcmp(".irqentry.text", txtname) == 0 ||
 242                 strcmp(".kprobes.text", txtname) == 0 ||
 243                 strcmp(".text.unlikely", txtname) == 0;
 244 }
 245
 246 /* 32 bit and 64 bit are very similar */
 247 #include "recordmcount.h"
 248 #define RECORD_MCOUNT_64
 249 #include "recordmcount.h"
 250
 251 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
 252  * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
 253  * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
 254  * to imply the order of the members; the spec does not say so.
 255  *      typedef unsigned char Elf64_Byte;
 256  * fails on MIPS64 because their <elf.h> already has it!
 257  */
 258
 259 typedef uint8_t myElf64_Byte;           /* Type for a 8-bit quantity.  */
 260
 261 union mips_r_info {
 262         Elf64_Xword r_info;
 263         struct {
 264                 Elf64_Word r_sym;               /* Symbol index.  */
 265                 myElf64_Byte r_ssym;            /* Special symbol.  */
 266                 myElf64_Byte r_type3;           /* Third relocation.  */
 267                 myElf64_Byte r_type2;           /* Second relocation.  */
 268                 myElf64_Byte r_type;            /* First relocation.  */
 269         } r_mips;
 270 };
 271
 272 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
 273 {
 274         return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
 275 }
 276
 277 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
 278 {
 279         rp->r_info = ((union mips_r_info){
 280                 .r_mips = { .r_sym = w(sym), .r_type = type }
 281         }).r_info;
 282 }
 283
 284 static void
 285 do_file(char const *const fname)
 286 {
 287         Elf32_Ehdr *const ehdr = mmap_file(fname);
 288         unsigned int reltype = 0;
 289
 290         ehdr_curr = ehdr;
 291         w = w4nat;
 292         w2 = w2nat;
 293         w8 = w8nat;
 294         switch (ehdr->e_ident[EI_DATA]) {
 295                 static unsigned int const endian = 1;
 296         default:
 297                 fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
 298                         ehdr->e_ident[EI_DATA], fname);
 299                 fail_file();
 300                 break;
 301         case ELFDATA2LSB:
 302                 if (*(unsigned char const *)&endian != 1) {
 303                         /* main() is big endian, file.o is little endian. */
 304                         w = w4rev;
 305                         w2 = w2rev;
 306                         w8 = w8rev;
 307                 }
 308                 break;
 309         case ELFDATA2MSB:
 310                 if (*(unsigned char const *)&endian != 0) {
 311                         /* main() is little endian, file.o is big endian. */
 312                         w = w4rev;
 313                         w2 = w2rev;
 314                         w8 = w8rev;
 315                 }
 316                 break;
 317         }  /* end switch */
 318         if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
 319         ||  w2(ehdr->e_type) != ET_REL
 320         ||  ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
 321                 fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
 322                 fail_file();
 323         }
 324
 325         gpfx = 0;
 326         switch (w2(ehdr->e_machine)) {
 327         default:
 328                 fprintf(stderr, "unrecognized e_machine %d %s\n",
 329                         w2(ehdr->e_machine), fname);
 330                 fail_file();
 331                 break;
 332         case EM_386:
 333                 reltype = R_386_32;
 334                 make_nop = make_nop_x86;
 335                 ideal_nop = ideal_nop5_x86_32;
 336                 break;
 337         case EM_ARM:     reltype = R_ARM_ABS32;
 338                          altmcount = "__gnu_mcount_nc";
 339                          break;
 340         case EM_IA_64:   reltype = R_IA64_IMM64;   gpfx = '_'; break;
 341         case EM_MIPS:    /* reltype: e_class    */ gpfx = '_'; break;
 342         case EM_PPC:     reltype = R_PPC_ADDR32;   gpfx = '_'; break;
 343         case EM_PPC64:   reltype = R_PPC64_ADDR64; gpfx = '_'; break;
 344         case EM_S390:    /* reltype: e_class    */ gpfx = '_'; break;
 345         case EM_SH:      reltype = R_SH_DIR32;                 break;
 346         case EM_SPARCV9: reltype = R_SPARC_64;     gpfx = '_'; break;
 347         case EM_X86_64:
 348                 make_nop = make_nop_x86;
 349                 ideal_nop = ideal_nop5_x86_64;
 350                 reltype = R_X86_64_64;
 351                 break;
 352         }  /* end switch */
 353
 354         switch (ehdr->e_ident[EI_CLASS]) {
 355         default:
 356                 fprintf(stderr, "unrecognized ELF class %d %s\n",
 357                         ehdr->e_ident[EI_CLASS], fname);
 358                 fail_file();
 359                 break;
 360         case ELFCLASS32:
 361                 if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
 362                 ||  w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
 363                         fprintf(stderr,
 364                                 "unrecognized ET_REL file: %s\n", fname);
 365                         fail_file();
 366                 }
 367                 if (w2(ehdr->e_machine) == EM_S390)
 368                         reltype = R_390_32;
 369                 if (w2(ehdr->e_machine) == EM_MIPS) {
 370                         reltype = R_MIPS_32;
 371                         is_fake_mcount32 = MIPS32_is_fake_mcount;
 372                 }
 373                 do32(ehdr, fname, reltype);
 374                 break;
 375         case ELFCLASS64: {
 376                 Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
 377                 if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
 378                 ||  w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
 379                         fprintf(stderr,
 380                                 "unrecognized ET_REL file: %s\n", fname);
 381                         fail_file();
 382                 }
 383                 if (w2(ghdr->e_machine) == EM_S390)
 384                         reltype = R_390_64;
 385                 if (w2(ghdr->e_machine) == EM_MIPS) {
 386                         reltype = R_MIPS_64;
 387                         Elf64_r_sym = MIPS64_r_sym;
 388                         Elf64_r_info = MIPS64_r_info;
 389                         is_fake_mcount64 = MIPS64_is_fake_mcount;
 390                 }
 391                 do64(ghdr, fname, reltype);
 392                 break;
 393         }
 394         }  /* end switch */
 395
 396         cleanup();
 397 }
 398
 399 int
 400 main(int argc, char const *argv[])
 401 {
 402         const char ftrace[] = "/ftrace.o";
 403         int ftrace_size = sizeof(ftrace) - 1;
 404         int n_error = 0;  /* gcc-4.3.0 false positive complaint */
 405
 406         if (argc <= 1) {
 407                 fprintf(stderr, "usage: recordmcount file.o...\n");
 408                 return 0;
 409         }
 410
 411         /* Process each file in turn, allowing deep failure. */
 412         for (--argc, ++argv; argc > 0; --argc, ++argv) {
 413                 int const sjval = setjmp(jmpenv);
 414                 int len;
 415
 416                 /*
 417                  * The file kernel/trace/ftrace.o references the mcount
 418                  * function but does not call it. Since ftrace.o should
 419                  * not be traced anyway, we just skip it.
 420                  */
 421                 len = strlen(argv[0]);
 422                 if (len >= ftrace_size &&
 423                     strcmp(argv[0] + (len - ftrace_size), ftrace) == 0)
 424                         continue;
 425
 426                 switch (sjval) {
 427                 default:
 428                         fprintf(stderr, "internal error: %s\n", argv[0]);
 429                         exit(1);
 430                         break;
 431                 case SJ_SETJMP:    /* normal sequence */
 432                         /* Avoid problems if early cleanup() */
 433                         fd_map = -1;
 434                         ehdr_curr = NULL;
 435                         mmap_failed = 1;
 436                         do_file(argv[0]);
 437                         break;
 438                 case SJ_FAIL:    /* error in do_file or below */
 439                         ++n_error;
 440                         break;
 441                 case SJ_SUCCEED:    /* premature success */
 442                         /* do nothing */
 443                         break;
 444                 }  /* end switch */
 445         }
 446         return !!n_error;
 447 }
 448
 449