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).
7 * Restructured to fit Linux format, as well as other updates:
8 * Copyright 2010 Steven Rostedt <srostedt@redhat.com>, Red Hat Inc.
12 * Strategy: alter the .o file in-place.
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.]
24 #include <sys/types.h>
36 static int fd_map; /* File descriptor for file being modified. */
37 static int mmap_failed; /* Boolean flag. */
38 static void *ehdr_curr; /* current ElfXX_Ehdr * for resource cleanup */
39 static char gpfx; /* prefix for global symbol name (sometimes '_') */
40 static struct stat sb; /* Remember .st_size, etc. */
41 static jmp_buf jmpenv; /* setjmp/longjmp per-file error escape */
42 static const char *altmcount; /* alternate mcount symbol name */
43 static int warn_on_notrace_sect; /* warn when section has mcount not being recorded */
45 /* setjmp() return values */
47 SJ_SETJMP = 0, /* hardwired first return */
52 /* Per-file resource cleanup when multiple files. */
57 munmap(ehdr_curr, sb.st_size);
63 static void __attribute__((noreturn))
67 longjmp(jmpenv, SJ_FAIL);
70 static void __attribute__((noreturn))
74 longjmp(jmpenv, SJ_SUCCEED);
77 /* ulseek, uread, ...: Check return value for errors. */
80 ulseek(int const fd, off_t const offset, int const whence)
82 off_t const w = lseek(fd, offset, whence);
91 uread(int const fd, void *const buf, size_t const count)
93 size_t const n = read(fd, buf, count);
102 uwrite(int const fd, void const *const buf, size_t const count)
104 size_t const n = write(fd, buf, count);
115 void *const addr = malloc(size);
117 fprintf(stderr, "malloc failed: %zu bytes\n", size);
123 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
124 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
125 static unsigned char *ideal_nop;
127 static char rel_type_nop;
129 static int (*make_nop)(void *map, size_t const offset);
131 static int make_nop_x86(void *map, size_t const offset)
136 /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
141 op = map + offset - 1;
146 ulseek(fd_map, offset - 1, SEEK_SET);
147 uwrite(fd_map, ideal_nop, 5);
152 * Get the whole file as a programming convenience in order to avoid
153 * malloc+lseek+read+free of many pieces. If successful, then mmap
154 * avoids copying unused pieces; else just read the whole file.
155 * Open for both read and write; new info will be appended to the file.
156 * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
157 * do not propagate to the file until an explicit overwrite at the last.
158 * This preserves most aspects of consistency (all except .st_size)
159 * for simultaneous readers of the file while we are appending to it.
160 * However, multiple writers still are bad. We choose not to use
161 * locking because it is expensive and the use case of kernel build
162 * makes multiple writers unlikely.
164 static void *mmap_file(char const *fname)
168 fd_map = open(fname, O_RDWR);
169 if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
173 if (!S_ISREG(sb.st_mode)) {
174 fprintf(stderr, "not a regular file: %s\n", fname);
177 addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
180 if (addr == MAP_FAILED) {
182 addr = umalloc(sb.st_size);
183 uread(fd_map, addr, sb.st_size);
185 if (sb.st_nlink != 1) {
186 /* file is hard-linked, break the hard link */
188 if (unlink(fname) < 0) {
192 fd_map = open(fname, O_RDWR | O_CREAT, sb.st_mode);
197 uwrite(fd_map, addr, sb.st_size);
202 /* w8rev, w8nat, ...: Handle endianness. */
204 static uint64_t w8rev(uint64_t const x)
206 return ((0xff & (x >> (0 * 8))) << (7 * 8))
207 | ((0xff & (x >> (1 * 8))) << (6 * 8))
208 | ((0xff & (x >> (2 * 8))) << (5 * 8))
209 | ((0xff & (x >> (3 * 8))) << (4 * 8))
210 | ((0xff & (x >> (4 * 8))) << (3 * 8))
211 | ((0xff & (x >> (5 * 8))) << (2 * 8))
212 | ((0xff & (x >> (6 * 8))) << (1 * 8))
213 | ((0xff & (x >> (7 * 8))) << (0 * 8));
216 static uint32_t w4rev(uint32_t const x)
218 return ((0xff & (x >> (0 * 8))) << (3 * 8))
219 | ((0xff & (x >> (1 * 8))) << (2 * 8))
220 | ((0xff & (x >> (2 * 8))) << (1 * 8))
221 | ((0xff & (x >> (3 * 8))) << (0 * 8));
224 static uint32_t w2rev(uint16_t const x)
226 return ((0xff & (x >> (0 * 8))) << (1 * 8))
227 | ((0xff & (x >> (1 * 8))) << (0 * 8));
230 static uint64_t w8nat(uint64_t const x)
235 static uint32_t w4nat(uint32_t const x)
240 static uint32_t w2nat(uint16_t const x)
245 static uint64_t (*w8)(uint64_t);
246 static uint32_t (*w)(uint32_t);
247 static uint32_t (*w2)(uint16_t);
249 /* Names of the sections that could contain calls to mcount. */
251 is_mcounted_section_name(char const *const txtname)
253 return strcmp(".text", txtname) == 0 ||
254 strcmp(".ref.text", txtname) == 0 ||
255 strcmp(".sched.text", txtname) == 0 ||
256 strcmp(".spinlock.text", txtname) == 0 ||
257 strcmp(".irqentry.text", txtname) == 0 ||
258 strcmp(".kprobes.text", txtname) == 0 ||
259 strcmp(".text.unlikely", txtname) == 0;
262 /* 32 bit and 64 bit are very similar */
263 #include "recordmcount.h"
264 #define RECORD_MCOUNT_64
265 #include "recordmcount.h"
267 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
268 * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
269 * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
270 * to imply the order of the members; the spec does not say so.
271 * typedef unsigned char Elf64_Byte;
272 * fails on MIPS64 because their <elf.h> already has it!
275 typedef uint8_t myElf64_Byte; /* Type for a 8-bit quantity. */
280 Elf64_Word r_sym; /* Symbol index. */
281 myElf64_Byte r_ssym; /* Special symbol. */
282 myElf64_Byte r_type3; /* Third relocation. */
283 myElf64_Byte r_type2; /* Second relocation. */
284 myElf64_Byte r_type; /* First relocation. */
288 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
290 return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
293 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
295 rp->r_info = ((union mips_r_info){
296 .r_mips = { .r_sym = w(sym), .r_type = type }
301 do_file(char const *const fname)
303 Elf32_Ehdr *const ehdr = mmap_file(fname);
304 unsigned int reltype = 0;
310 switch (ehdr->e_ident[EI_DATA]) {
311 static unsigned int const endian = 1;
313 fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
314 ehdr->e_ident[EI_DATA], fname);
318 if (*(unsigned char const *)&endian != 1) {
319 /* main() is big endian, file.o is little endian. */
326 if (*(unsigned char const *)&endian != 0) {
327 /* main() is little endian, file.o is big endian. */
334 if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
335 || w2(ehdr->e_type) != ET_REL
336 || ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
337 fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
342 switch (w2(ehdr->e_machine)) {
344 fprintf(stderr, "unrecognized e_machine %d %s\n",
345 w2(ehdr->e_machine), fname);
350 make_nop = make_nop_x86;
351 ideal_nop = ideal_nop5_x86_32;
352 mcount_adjust_32 = -1;
354 case EM_ARM: reltype = R_ARM_ABS32;
355 altmcount = "__gnu_mcount_nc";
357 case EM_IA_64: reltype = R_IA64_IMM64; gpfx = '_'; break;
358 case EM_MIPS: /* reltype: e_class */ gpfx = '_'; break;
359 case EM_PPC: reltype = R_PPC_ADDR32; gpfx = '_'; break;
360 case EM_PPC64: reltype = R_PPC64_ADDR64; gpfx = '_'; break;
361 case EM_S390: /* reltype: e_class */ gpfx = '_'; break;
362 case EM_SH: reltype = R_SH_DIR32; break;
363 case EM_SPARCV9: reltype = R_SPARC_64; gpfx = '_'; break;
365 make_nop = make_nop_x86;
366 ideal_nop = ideal_nop5_x86_64;
367 reltype = R_X86_64_64;
368 mcount_adjust_64 = -1;
372 switch (ehdr->e_ident[EI_CLASS]) {
374 fprintf(stderr, "unrecognized ELF class %d %s\n",
375 ehdr->e_ident[EI_CLASS], fname);
379 if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
380 || w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
382 "unrecognized ET_REL file: %s\n", fname);
385 if (w2(ehdr->e_machine) == EM_S390) {
387 mcount_adjust_32 = -4;
389 if (w2(ehdr->e_machine) == EM_MIPS) {
391 is_fake_mcount32 = MIPS32_is_fake_mcount;
393 do32(ehdr, fname, reltype);
396 Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
397 if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
398 || w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
400 "unrecognized ET_REL file: %s\n", fname);
403 if (w2(ghdr->e_machine) == EM_S390) {
405 mcount_adjust_64 = -8;
407 if (w2(ghdr->e_machine) == EM_MIPS) {
409 Elf64_r_sym = MIPS64_r_sym;
410 Elf64_r_info = MIPS64_r_info;
411 is_fake_mcount64 = MIPS64_is_fake_mcount;
413 do64(ghdr, fname, reltype);
422 main(int argc, char *argv[])
424 const char ftrace[] = "/ftrace.o";
425 int ftrace_size = sizeof(ftrace) - 1;
426 int n_error = 0; /* gcc-4.3.0 false positive complaint */
430 while ((c = getopt(argc, argv, "w")) >= 0) {
433 warn_on_notrace_sect = 1;
436 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
441 if ((argc - optind) < 1) {
442 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
446 /* Process each file in turn, allowing deep failure. */
447 for (i = optind; i < argc; i++) {
448 char *file = argv[i];
449 int const sjval = setjmp(jmpenv);
453 * The file kernel/trace/ftrace.o references the mcount
454 * function but does not call it. Since ftrace.o should
455 * not be traced anyway, we just skip it.
458 if (len >= ftrace_size &&
459 strcmp(file + (len - ftrace_size), ftrace) == 0)
464 fprintf(stderr, "internal error: %s\n", file);
467 case SJ_SETJMP: /* normal sequence */
468 /* Avoid problems if early cleanup() */
474 case SJ_FAIL: /* error in do_file or below */
477 case SJ_SUCCEED: /* premature success */