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