linux/scripts/recordmcount.c
<<
>>
Prefs
   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
  50static int fd_map;      /* File descriptor for file being modified. */
  51static int mmap_failed; /* Boolean flag. */
  52static void *ehdr_curr; /* current ElfXX_Ehdr *  for resource cleanup */
  53static char gpfx;       /* prefix for global symbol name (sometimes '_') */
  54static struct stat sb;  /* Remember .st_size, etc. */
  55static jmp_buf jmpenv;  /* setjmp/longjmp per-file error escape */
  56static const char *altmcount;   /* alternate mcount symbol name */
  57static int warn_on_notrace_sect; /* warn when section has mcount not being recorded */
  58
  59/* setjmp() return values */
  60enum {
  61        SJ_SETJMP = 0,  /* hardwired first return */
  62        SJ_FAIL,
  63        SJ_SUCCEED
  64};
  65
  66/* Per-file resource cleanup when multiple files. */
  67static void
  68cleanup(void)
  69{
  70        if (!mmap_failed)
  71                munmap(ehdr_curr, sb.st_size);
  72        else
  73                free(ehdr_curr);
  74        close(fd_map);
  75}
  76
  77static void __attribute__((noreturn))
  78fail_file(void)
  79{
  80        cleanup();
  81        longjmp(jmpenv, SJ_FAIL);
  82}
  83
  84static void __attribute__((noreturn))
  85succeed_file(void)
  86{
  87        cleanup();
  88        longjmp(jmpenv, SJ_SUCCEED);
  89}
  90
  91/* ulseek, uread, ...:  Check return value for errors. */
  92
  93static off_t
  94ulseek(int const fd, off_t const offset, int const whence)
  95{
  96        off_t const w = lseek(fd, offset, whence);
  97        if (w == (off_t)-1) {
  98                perror("lseek");
  99                fail_file();
 100        }
 101        return w;
 102}
 103
 104static size_t
 105uread(int const fd, void *const buf, size_t const count)
 106{
 107        size_t const n = read(fd, buf, count);
 108        if (n != count) {
 109                perror("read");
 110                fail_file();
 111        }
 112        return n;
 113}
 114
 115static size_t
 116uwrite(int const fd, void const *const buf, size_t const count)
 117{
 118        size_t const n = write(fd, buf, count);
 119        if (n != count) {
 120                perror("write");
 121                fail_file();
 122        }
 123        return n;
 124}
 125
 126static void *
 127umalloc(size_t size)
 128{
 129        void *const addr = malloc(size);
 130        if (addr == 0) {
 131                fprintf(stderr, "malloc failed: %zu bytes\n", size);
 132                fail_file();
 133        }
 134        return addr;
 135}
 136
 137static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
 138static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
 139static unsigned char *ideal_nop;
 140
 141static char rel_type_nop;
 142
 143static int (*make_nop)(void *map, size_t const offset);
 144
 145static int make_nop_x86(void *map, size_t const offset)
 146{
 147        uint32_t *ptr;
 148        unsigned char *op;
 149
 150        /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
 151        ptr = map + offset;
 152        if (*ptr != 0)
 153                return -1;
 154
 155        op = map + offset - 1;
 156        if (*op != 0xe8)
 157                return -1;
 158
 159        /* convert to nop */
 160        ulseek(fd_map, offset - 1, SEEK_SET);
 161        uwrite(fd_map, ideal_nop, 5);
 162        return 0;
 163}
 164
 165/*
 166 * Get the whole file as a programming convenience in order to avoid
 167 * malloc+lseek+read+free of many pieces.  If successful, then mmap
 168 * avoids copying unused pieces; else just read the whole file.
 169 * Open for both read and write; new info will be appended to the file.
 170 * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
 171 * do not propagate to the file until an explicit overwrite at the last.
 172 * This preserves most aspects of consistency (all except .st_size)
 173 * for simultaneous readers of the file while we are appending to it.
 174 * However, multiple writers still are bad.  We choose not to use
 175 * locking because it is expensive and the use case of kernel build
 176 * makes multiple writers unlikely.
 177 */
 178static void *mmap_file(char const *fname)
 179{
 180        void *addr;
 181
 182        fd_map = open(fname, O_RDWR);
 183        if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
 184                perror(fname);
 185                fail_file();
 186        }
 187        if (!S_ISREG(sb.st_mode)) {
 188                fprintf(stderr, "not a regular file: %s\n", fname);
 189                fail_file();
 190        }
 191        addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
 192                    fd_map, 0);
 193        mmap_failed = 0;
 194        if (addr == MAP_FAILED) {
 195                mmap_failed = 1;
 196                addr = umalloc(sb.st_size);
 197                uread(fd_map, addr, sb.st_size);
 198        }
 199        if (sb.st_nlink != 1) {
 200                /* file is hard-linked, break the hard link */
 201                close(fd_map);
 202                if (unlink(fname) < 0) {
 203                        perror(fname);
 204                        fail_file();
 205                }
 206                fd_map = open(fname, O_RDWR | O_CREAT, sb.st_mode);
 207                if (fd_map < 0) {
 208                        perror(fname);
 209                        fail_file();
 210                }
 211                uwrite(fd_map, addr, sb.st_size);
 212        }
 213        return addr;
 214}
 215
 216/* w8rev, w8nat, ...: Handle endianness. */
 217
 218static uint64_t w8rev(uint64_t const x)
 219{
 220        return   ((0xff & (x >> (0 * 8))) << (7 * 8))
 221               | ((0xff & (x >> (1 * 8))) << (6 * 8))
 222               | ((0xff & (x >> (2 * 8))) << (5 * 8))
 223               | ((0xff & (x >> (3 * 8))) << (4 * 8))
 224               | ((0xff & (x >> (4 * 8))) << (3 * 8))
 225               | ((0xff & (x >> (5 * 8))) << (2 * 8))
 226               | ((0xff & (x >> (6 * 8))) << (1 * 8))
 227               | ((0xff & (x >> (7 * 8))) << (0 * 8));
 228}
 229
 230static uint32_t w4rev(uint32_t const x)
 231{
 232        return   ((0xff & (x >> (0 * 8))) << (3 * 8))
 233               | ((0xff & (x >> (1 * 8))) << (2 * 8))
 234               | ((0xff & (x >> (2 * 8))) << (1 * 8))
 235               | ((0xff & (x >> (3 * 8))) << (0 * 8));
 236}
 237
 238static uint32_t w2rev(uint16_t const x)
 239{
 240        return   ((0xff & (x >> (0 * 8))) << (1 * 8))
 241               | ((0xff & (x >> (1 * 8))) << (0 * 8));
 242}
 243
 244static uint64_t w8nat(uint64_t const x)
 245{
 246        return x;
 247}
 248
 249static uint32_t w4nat(uint32_t const x)
 250{
 251        return x;
 252}
 253
 254static uint32_t w2nat(uint16_t const x)
 255{
 256        return x;
 257}
 258
 259static uint64_t (*w8)(uint64_t);
 260static uint32_t (*w)(uint32_t);
 261static uint32_t (*w2)(uint16_t);
 262
 263/* Names of the sections that could contain calls to mcount. */
 264static int
 265is_mcounted_section_name(char const *const txtname)
 266{
 267        return strcmp(".text",           txtname) == 0 ||
 268                strcmp(".ref.text",      txtname) == 0 ||
 269                strcmp(".sched.text",    txtname) == 0 ||
 270                strcmp(".spinlock.text", txtname) == 0 ||
 271                strcmp(".irqentry.text", txtname) == 0 ||
 272                strcmp(".kprobes.text", txtname) == 0 ||
 273                strcmp(".text.unlikely", txtname) == 0;
 274}
 275
 276/* 32 bit and 64 bit are very similar */
 277#include "recordmcount.h"
 278#define RECORD_MCOUNT_64
 279#include "recordmcount.h"
 280
 281/* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
 282 * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
 283 * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
 284 * to imply the order of the members; the spec does not say so.
 285 *      typedef unsigned char Elf64_Byte;
 286 * fails on MIPS64 because their <elf.h> already has it!
 287 */
 288
 289typedef uint8_t myElf64_Byte;           /* Type for a 8-bit quantity.  */
 290
 291union mips_r_info {
 292        Elf64_Xword r_info;
 293        struct {
 294                Elf64_Word r_sym;               /* Symbol index.  */
 295                myElf64_Byte r_ssym;            /* Special symbol.  */
 296                myElf64_Byte r_type3;           /* Third relocation.  */
 297                myElf64_Byte r_type2;           /* Second relocation.  */
 298                myElf64_Byte r_type;            /* First relocation.  */
 299        } r_mips;
 300};
 301
 302static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
 303{
 304        return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
 305}
 306
 307static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
 308{
 309        rp->r_info = ((union mips_r_info){
 310                .r_mips = { .r_sym = w(sym), .r_type = type }
 311        }).r_info;
 312}
 313
 314static void
 315do_file(char const *const fname)
 316{
 317        Elf32_Ehdr *const ehdr = mmap_file(fname);
 318        unsigned int reltype = 0;
 319
 320        ehdr_curr = ehdr;
 321        w = w4nat;
 322        w2 = w2nat;
 323        w8 = w8nat;
 324        switch (ehdr->e_ident[EI_DATA]) {
 325                static unsigned int const endian = 1;
 326        default:
 327                fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
 328                        ehdr->e_ident[EI_DATA], fname);
 329                fail_file();
 330                break;
 331        case ELFDATA2LSB:
 332                if (*(unsigned char const *)&endian != 1) {
 333                        /* main() is big endian, file.o is little endian. */
 334                        w = w4rev;
 335                        w2 = w2rev;
 336                        w8 = w8rev;
 337                }
 338                break;
 339        case ELFDATA2MSB:
 340                if (*(unsigned char const *)&endian != 0) {
 341                        /* main() is little endian, file.o is big endian. */
 342                        w = w4rev;
 343                        w2 = w2rev;
 344                        w8 = w8rev;
 345                }
 346                break;
 347        }  /* end switch */
 348        if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
 349        ||  w2(ehdr->e_type) != ET_REL
 350        ||  ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
 351                fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
 352                fail_file();
 353        }
 354
 355        gpfx = 0;
 356        switch (w2(ehdr->e_machine)) {
 357        default:
 358                fprintf(stderr, "unrecognized e_machine %d %s\n",
 359                        w2(ehdr->e_machine), fname);
 360                fail_file();
 361                break;
 362        case EM_386:
 363                reltype = R_386_32;
 364                make_nop = make_nop_x86;
 365                ideal_nop = ideal_nop5_x86_32;
 366                mcount_adjust_32 = -1;
 367                break;
 368        case EM_ARM:     reltype = R_ARM_ABS32;
 369                         altmcount = "__gnu_mcount_nc";
 370                         break;
 371        case EM_IA_64:   reltype = R_IA64_IMM64;   gpfx = '_'; break;
 372        case EM_METAG:   reltype = R_METAG_ADDR32;
 373                         altmcount = "_mcount_wrapper";
 374                         rel_type_nop = R_METAG_NONE;
 375                         /* We happen to have the same requirement as MIPS */
 376                         is_fake_mcount32 = MIPS32_is_fake_mcount;
 377                         break;
 378        case EM_MIPS:    /* reltype: e_class    */ gpfx = '_'; break;
 379        case EM_PPC:     reltype = R_PPC_ADDR32;   gpfx = '_'; break;
 380        case EM_PPC64:   reltype = R_PPC64_ADDR64; gpfx = '_'; break;
 381        case EM_S390:    /* reltype: e_class    */ gpfx = '_'; break;
 382        case EM_SH:      reltype = R_SH_DIR32;                 break;
 383        case EM_SPARCV9: reltype = R_SPARC_64;     gpfx = '_'; break;
 384        case EM_X86_64:
 385                make_nop = make_nop_x86;
 386                ideal_nop = ideal_nop5_x86_64;
 387                reltype = R_X86_64_64;
 388                mcount_adjust_64 = -1;
 389                break;
 390        }  /* end switch */
 391
 392        switch (ehdr->e_ident[EI_CLASS]) {
 393        default:
 394                fprintf(stderr, "unrecognized ELF class %d %s\n",
 395                        ehdr->e_ident[EI_CLASS], fname);
 396                fail_file();
 397                break;
 398        case ELFCLASS32:
 399                if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
 400                ||  w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
 401                        fprintf(stderr,
 402                                "unrecognized ET_REL file: %s\n", fname);
 403                        fail_file();
 404                }
 405                if (w2(ehdr->e_machine) == EM_S390) {
 406                        reltype = R_390_32;
 407                        mcount_adjust_32 = -4;
 408                }
 409                if (w2(ehdr->e_machine) == EM_MIPS) {
 410                        reltype = R_MIPS_32;
 411                        is_fake_mcount32 = MIPS32_is_fake_mcount;
 412                }
 413                do32(ehdr, fname, reltype);
 414                break;
 415        case ELFCLASS64: {
 416                Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
 417                if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
 418                ||  w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
 419                        fprintf(stderr,
 420                                "unrecognized ET_REL file: %s\n", fname);
 421                        fail_file();
 422                }
 423                if (w2(ghdr->e_machine) == EM_S390) {
 424                        reltype = R_390_64;
 425                        mcount_adjust_64 = -8;
 426                }
 427                if (w2(ghdr->e_machine) == EM_MIPS) {
 428                        reltype = R_MIPS_64;
 429                        Elf64_r_sym = MIPS64_r_sym;
 430                        Elf64_r_info = MIPS64_r_info;
 431                        is_fake_mcount64 = MIPS64_is_fake_mcount;
 432                }
 433                do64(ghdr, fname, reltype);
 434                break;
 435        }
 436        }  /* end switch */
 437
 438        cleanup();
 439}
 440
 441int
 442main(int argc, char *argv[])
 443{
 444        const char ftrace[] = "/ftrace.o";
 445        int ftrace_size = sizeof(ftrace) - 1;
 446        int n_error = 0;  /* gcc-4.3.0 false positive complaint */
 447        int c;
 448        int i;
 449
 450        while ((c = getopt(argc, argv, "w")) >= 0) {
 451                switch (c) {
 452                case 'w':
 453                        warn_on_notrace_sect = 1;
 454                        break;
 455                default:
 456                        fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
 457                        return 0;
 458                }
 459        }
 460
 461        if ((argc - optind) < 1) {
 462                fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
 463                return 0;
 464        }
 465
 466        /* Process each file in turn, allowing deep failure. */
 467        for (i = optind; i < argc; i++) {
 468                char *file = argv[i];
 469                int const sjval = setjmp(jmpenv);
 470                int len;
 471
 472                /*
 473                 * The file kernel/trace/ftrace.o references the mcount
 474                 * function but does not call it. Since ftrace.o should
 475                 * not be traced anyway, we just skip it.
 476                 */
 477                len = strlen(file);
 478                if (len >= ftrace_size &&
 479                    strcmp(file + (len - ftrace_size), ftrace) == 0)
 480                        continue;
 481
 482                switch (sjval) {
 483                default:
 484                        fprintf(stderr, "internal error: %s\n", file);
 485                        exit(1);
 486                        break;
 487                case SJ_SETJMP:    /* normal sequence */
 488                        /* Avoid problems if early cleanup() */
 489                        fd_map = -1;
 490                        ehdr_curr = NULL;
 491                        mmap_failed = 1;
 492                        do_file(file);
 493                        break;
 494                case SJ_FAIL:    /* error in do_file or below */
 495                        ++n_error;
 496                        break;
 497                case SJ_SUCCEED:    /* premature success */
 498                        /* do nothing */
 499                        break;
 500                }  /* end switch */
 501        }
 502        return !!n_error;
 503}
 504
 505
 506