mm, compaction: make full priority ignore pageblock suitability
[linux-stable.git] / scripts / recordmcount.c
bloba68f03133df9bd30543c7d7190f6e2b8d285c628
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).
7 * Restructured to fit Linux format, as well as other updates:
8 * Copyright 2010 Steven Rostedt <srostedt@redhat.com>, Red Hat Inc.
9 */
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>
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>
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.
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
50 #ifndef EM_AARCH64
51 #define EM_AARCH64 183
52 #define R_AARCH64_NONE 0
53 #define R_AARCH64_ABS64 257
54 #endif
56 static int fd_map; /* File descriptor for file being modified. */
57 static int mmap_failed; /* Boolean flag. */
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 static void *file_map; /* pointer of the mapped file */
64 static void *file_end; /* pointer to the end of the mapped file */
65 static int file_updated; /* flag to state file was changed */
66 static void *file_ptr; /* current file pointer location */
67 static void *file_append; /* added to the end of the file */
68 static size_t file_append_size; /* how much is added to end of file */
70 /* setjmp() return values */
71 enum {
72 SJ_SETJMP = 0, /* hardwired first return */
73 SJ_FAIL,
74 SJ_SUCCEED
77 /* Per-file resource cleanup when multiple files. */
78 static void
79 cleanup(void)
81 if (!mmap_failed)
82 munmap(file_map, sb.st_size);
83 else
84 free(file_map);
85 file_map = NULL;
86 free(file_append);
87 file_append = NULL;
88 file_append_size = 0;
89 file_updated = 0;
92 static void __attribute__((noreturn))
93 fail_file(void)
95 cleanup();
96 longjmp(jmpenv, SJ_FAIL);
99 static void __attribute__((noreturn))
100 succeed_file(void)
102 cleanup();
103 longjmp(jmpenv, SJ_SUCCEED);
106 /* ulseek, uread, ...: Check return value for errors. */
108 static off_t
109 ulseek(int const fd, off_t const offset, int const whence)
111 switch (whence) {
112 case SEEK_SET:
113 file_ptr = file_map + offset;
114 break;
115 case SEEK_CUR:
116 file_ptr += offset;
117 break;
118 case SEEK_END:
119 file_ptr = file_map + (sb.st_size - offset);
120 break;
122 if (file_ptr < file_map) {
123 fprintf(stderr, "lseek: seek before file\n");
124 fail_file();
126 return file_ptr - file_map;
129 static size_t
130 uread(int const fd, void *const buf, size_t const count)
132 size_t const n = read(fd, buf, count);
133 if (n != count) {
134 perror("read");
135 fail_file();
137 return n;
140 static size_t
141 uwrite(int const fd, void const *const buf, size_t const count)
143 size_t cnt = count;
144 off_t idx = 0;
146 file_updated = 1;
148 if (file_ptr + count >= file_end) {
149 off_t aoffset = (file_ptr + count) - file_end;
151 if (aoffset > file_append_size) {
152 file_append = realloc(file_append, aoffset);
153 file_append_size = aoffset;
155 if (!file_append) {
156 perror("write");
157 fail_file();
159 if (file_ptr < file_end) {
160 cnt = file_end - file_ptr;
161 } else {
162 cnt = 0;
163 idx = aoffset - count;
167 if (cnt)
168 memcpy(file_ptr, buf, cnt);
170 if (cnt < count)
171 memcpy(file_append + idx, buf + cnt, count - cnt);
173 file_ptr += count;
174 return count;
177 static void *
178 umalloc(size_t size)
180 void *const addr = malloc(size);
181 if (addr == 0) {
182 fprintf(stderr, "malloc failed: %zu bytes\n", size);
183 fail_file();
185 return addr;
188 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
189 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
190 static unsigned char *ideal_nop;
192 static char rel_type_nop;
194 static int (*make_nop)(void *map, size_t const offset);
196 static int make_nop_x86(void *map, size_t const offset)
198 uint32_t *ptr;
199 unsigned char *op;
201 /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
202 ptr = map + offset;
203 if (*ptr != 0)
204 return -1;
206 op = map + offset - 1;
207 if (*op != 0xe8)
208 return -1;
210 /* convert to nop */
211 ulseek(fd_map, offset - 1, SEEK_SET);
212 uwrite(fd_map, ideal_nop, 5);
213 return 0;
216 static unsigned char ideal_nop4_arm64[4] = {0x1f, 0x20, 0x03, 0xd5};
217 static int make_nop_arm64(void *map, size_t const offset)
219 uint32_t *ptr;
221 ptr = map + offset;
222 /* bl <_mcount> is 0x94000000 before relocation */
223 if (*ptr != 0x94000000)
224 return -1;
226 /* Convert to nop */
227 ulseek(fd_map, offset, SEEK_SET);
228 uwrite(fd_map, ideal_nop, 4);
229 return 0;
233 * Get the whole file as a programming convenience in order to avoid
234 * malloc+lseek+read+free of many pieces. If successful, then mmap
235 * avoids copying unused pieces; else just read the whole file.
236 * Open for both read and write; new info will be appended to the file.
237 * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
238 * do not propagate to the file until an explicit overwrite at the last.
239 * This preserves most aspects of consistency (all except .st_size)
240 * for simultaneous readers of the file while we are appending to it.
241 * However, multiple writers still are bad. We choose not to use
242 * locking because it is expensive and the use case of kernel build
243 * makes multiple writers unlikely.
245 static void *mmap_file(char const *fname)
247 fd_map = open(fname, O_RDONLY);
248 if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
249 perror(fname);
250 fail_file();
252 if (!S_ISREG(sb.st_mode)) {
253 fprintf(stderr, "not a regular file: %s\n", fname);
254 fail_file();
256 file_map = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
257 fd_map, 0);
258 mmap_failed = 0;
259 if (file_map == MAP_FAILED) {
260 mmap_failed = 1;
261 file_map = umalloc(sb.st_size);
262 uread(fd_map, file_map, sb.st_size);
264 close(fd_map);
266 file_end = file_map + sb.st_size;
268 return file_map;
271 static void write_file(const char *fname)
273 char tmp_file[strlen(fname) + 4];
274 size_t n;
276 if (!file_updated)
277 return;
279 sprintf(tmp_file, "%s.rc", fname);
282 * After reading the entire file into memory, delete it
283 * and write it back, to prevent weird side effects of modifying
284 * an object file in place.
286 fd_map = open(tmp_file, O_WRONLY | O_TRUNC | O_CREAT, sb.st_mode);
287 if (fd_map < 0) {
288 perror(fname);
289 fail_file();
291 n = write(fd_map, file_map, sb.st_size);
292 if (n != sb.st_size) {
293 perror("write");
294 fail_file();
296 if (file_append_size) {
297 n = write(fd_map, file_append, file_append_size);
298 if (n != file_append_size) {
299 perror("write");
300 fail_file();
303 close(fd_map);
304 if (rename(tmp_file, fname) < 0) {
305 perror(fname);
306 fail_file();
310 /* w8rev, w8nat, ...: Handle endianness. */
312 static uint64_t w8rev(uint64_t const x)
314 return ((0xff & (x >> (0 * 8))) << (7 * 8))
315 | ((0xff & (x >> (1 * 8))) << (6 * 8))
316 | ((0xff & (x >> (2 * 8))) << (5 * 8))
317 | ((0xff & (x >> (3 * 8))) << (4 * 8))
318 | ((0xff & (x >> (4 * 8))) << (3 * 8))
319 | ((0xff & (x >> (5 * 8))) << (2 * 8))
320 | ((0xff & (x >> (6 * 8))) << (1 * 8))
321 | ((0xff & (x >> (7 * 8))) << (0 * 8));
324 static uint32_t w4rev(uint32_t const x)
326 return ((0xff & (x >> (0 * 8))) << (3 * 8))
327 | ((0xff & (x >> (1 * 8))) << (2 * 8))
328 | ((0xff & (x >> (2 * 8))) << (1 * 8))
329 | ((0xff & (x >> (3 * 8))) << (0 * 8));
332 static uint32_t w2rev(uint16_t const x)
334 return ((0xff & (x >> (0 * 8))) << (1 * 8))
335 | ((0xff & (x >> (1 * 8))) << (0 * 8));
338 static uint64_t w8nat(uint64_t const x)
340 return x;
343 static uint32_t w4nat(uint32_t const x)
345 return x;
348 static uint32_t w2nat(uint16_t const x)
350 return x;
353 static uint64_t (*w8)(uint64_t);
354 static uint32_t (*w)(uint32_t);
355 static uint32_t (*w2)(uint16_t);
357 /* Names of the sections that could contain calls to mcount. */
358 static int
359 is_mcounted_section_name(char const *const txtname)
361 return strcmp(".text", txtname) == 0 ||
362 strcmp(".ref.text", txtname) == 0 ||
363 strcmp(".sched.text", txtname) == 0 ||
364 strcmp(".spinlock.text", txtname) == 0 ||
365 strcmp(".irqentry.text", txtname) == 0 ||
366 strcmp(".softirqentry.text", txtname) == 0 ||
367 strcmp(".kprobes.text", txtname) == 0 ||
368 strcmp(".text.unlikely", txtname) == 0;
371 /* 32 bit and 64 bit are very similar */
372 #include "recordmcount.h"
373 #define RECORD_MCOUNT_64
374 #include "recordmcount.h"
376 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
377 * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
378 * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
379 * to imply the order of the members; the spec does not say so.
380 * typedef unsigned char Elf64_Byte;
381 * fails on MIPS64 because their <elf.h> already has it!
384 typedef uint8_t myElf64_Byte; /* Type for a 8-bit quantity. */
386 union mips_r_info {
387 Elf64_Xword r_info;
388 struct {
389 Elf64_Word r_sym; /* Symbol index. */
390 myElf64_Byte r_ssym; /* Special symbol. */
391 myElf64_Byte r_type3; /* Third relocation. */
392 myElf64_Byte r_type2; /* Second relocation. */
393 myElf64_Byte r_type; /* First relocation. */
394 } r_mips;
397 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
399 return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
402 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
404 rp->r_info = ((union mips_r_info){
405 .r_mips = { .r_sym = w(sym), .r_type = type }
406 }).r_info;
409 static void
410 do_file(char const *const fname)
412 Elf32_Ehdr *const ehdr = mmap_file(fname);
413 unsigned int reltype = 0;
415 w = w4nat;
416 w2 = w2nat;
417 w8 = w8nat;
418 switch (ehdr->e_ident[EI_DATA]) {
419 static unsigned int const endian = 1;
420 default:
421 fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
422 ehdr->e_ident[EI_DATA], fname);
423 fail_file();
424 break;
425 case ELFDATA2LSB:
426 if (*(unsigned char const *)&endian != 1) {
427 /* main() is big endian, file.o is little endian. */
428 w = w4rev;
429 w2 = w2rev;
430 w8 = w8rev;
432 break;
433 case ELFDATA2MSB:
434 if (*(unsigned char const *)&endian != 0) {
435 /* main() is little endian, file.o is big endian. */
436 w = w4rev;
437 w2 = w2rev;
438 w8 = w8rev;
440 break;
441 } /* end switch */
442 if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
443 || w2(ehdr->e_type) != ET_REL
444 || ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
445 fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
446 fail_file();
449 gpfx = 0;
450 switch (w2(ehdr->e_machine)) {
451 default:
452 fprintf(stderr, "unrecognized e_machine %d %s\n",
453 w2(ehdr->e_machine), fname);
454 fail_file();
455 break;
456 case EM_386:
457 reltype = R_386_32;
458 rel_type_nop = R_386_NONE;
459 make_nop = make_nop_x86;
460 ideal_nop = ideal_nop5_x86_32;
461 mcount_adjust_32 = -1;
462 break;
463 case EM_ARM: reltype = R_ARM_ABS32;
464 altmcount = "__gnu_mcount_nc";
465 break;
466 case EM_AARCH64:
467 reltype = R_AARCH64_ABS64;
468 make_nop = make_nop_arm64;
469 rel_type_nop = R_AARCH64_NONE;
470 ideal_nop = ideal_nop4_arm64;
471 gpfx = '_';
472 break;
473 case EM_IA_64: reltype = R_IA64_IMM64; gpfx = '_'; break;
474 case EM_METAG: reltype = R_METAG_ADDR32;
475 altmcount = "_mcount_wrapper";
476 rel_type_nop = R_METAG_NONE;
477 /* We happen to have the same requirement as MIPS */
478 is_fake_mcount32 = MIPS32_is_fake_mcount;
479 break;
480 case EM_MIPS: /* reltype: e_class */ gpfx = '_'; break;
481 case EM_PPC: reltype = R_PPC_ADDR32; gpfx = '_'; break;
482 case EM_PPC64: reltype = R_PPC64_ADDR64; gpfx = '_'; break;
483 case EM_S390: /* reltype: e_class */ gpfx = '_'; break;
484 case EM_SH: reltype = R_SH_DIR32; break;
485 case EM_SPARCV9: reltype = R_SPARC_64; gpfx = '_'; break;
486 case EM_X86_64:
487 make_nop = make_nop_x86;
488 ideal_nop = ideal_nop5_x86_64;
489 reltype = R_X86_64_64;
490 rel_type_nop = R_X86_64_NONE;
491 mcount_adjust_64 = -1;
492 break;
493 } /* end switch */
495 switch (ehdr->e_ident[EI_CLASS]) {
496 default:
497 fprintf(stderr, "unrecognized ELF class %d %s\n",
498 ehdr->e_ident[EI_CLASS], fname);
499 fail_file();
500 break;
501 case ELFCLASS32:
502 if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
503 || w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
504 fprintf(stderr,
505 "unrecognized ET_REL file: %s\n", fname);
506 fail_file();
508 if (w2(ehdr->e_machine) == EM_MIPS) {
509 reltype = R_MIPS_32;
510 is_fake_mcount32 = MIPS32_is_fake_mcount;
512 do32(ehdr, fname, reltype);
513 break;
514 case ELFCLASS64: {
515 Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
516 if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
517 || w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
518 fprintf(stderr,
519 "unrecognized ET_REL file: %s\n", fname);
520 fail_file();
522 if (w2(ghdr->e_machine) == EM_S390) {
523 reltype = R_390_64;
524 mcount_adjust_64 = -14;
526 if (w2(ghdr->e_machine) == EM_MIPS) {
527 reltype = R_MIPS_64;
528 Elf64_r_sym = MIPS64_r_sym;
529 Elf64_r_info = MIPS64_r_info;
530 is_fake_mcount64 = MIPS64_is_fake_mcount;
532 do64(ghdr, fname, reltype);
533 break;
535 } /* end switch */
537 write_file(fname);
538 cleanup();
542 main(int argc, char *argv[])
544 const char ftrace[] = "/ftrace.o";
545 int ftrace_size = sizeof(ftrace) - 1;
546 int n_error = 0; /* gcc-4.3.0 false positive complaint */
547 int c;
548 int i;
550 while ((c = getopt(argc, argv, "w")) >= 0) {
551 switch (c) {
552 case 'w':
553 warn_on_notrace_sect = 1;
554 break;
555 default:
556 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
557 return 0;
561 if ((argc - optind) < 1) {
562 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
563 return 0;
566 /* Process each file in turn, allowing deep failure. */
567 for (i = optind; i < argc; i++) {
568 char *file = argv[i];
569 int const sjval = setjmp(jmpenv);
570 int len;
573 * The file kernel/trace/ftrace.o references the mcount
574 * function but does not call it. Since ftrace.o should
575 * not be traced anyway, we just skip it.
577 len = strlen(file);
578 if (len >= ftrace_size &&
579 strcmp(file + (len - ftrace_size), ftrace) == 0)
580 continue;
582 switch (sjval) {
583 default:
584 fprintf(stderr, "internal error: %s\n", file);
585 exit(1);
586 break;
587 case SJ_SETJMP: /* normal sequence */
588 /* Avoid problems if early cleanup() */
589 fd_map = -1;
590 mmap_failed = 1;
591 file_map = NULL;
592 file_ptr = NULL;
593 file_updated = 0;
594 do_file(file);
595 break;
596 case SJ_FAIL: /* error in do_file or below */
597 fprintf(stderr, "%s: failed\n", file);
598 ++n_error;
599 break;
600 case SJ_SUCCEED: /* premature success */
601 /* do nothing */
602 break;
603 } /* end switch */
605 return !!n_error;