1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * recordmcount.c: construct a table of the locations of calls to 'mcount'
4 * so that ftrace can find them quickly.
5 * Copyright 2009 John F. Reiser <jreiser@BitWagon.com>. All rights reserved.
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 <stdio.h>
31 #include <stdlib.h>
32 #include <string.h>
33 #include <unistd.h>
34
35 #ifndef EM_AARCH64
36 #define EM_AARCH64 183
37 #define R_AARCH64_NONE 0
38 #define R_AARCH64_ABS64 257
39 #endif
40
41 #define R_ARM_PC24 1
42 #define R_ARM_THM_CALL 10
43 #define R_ARM_CALL 28
44
45 static int fd_map; /* File descriptor for file being modified. */
46 static int mmap_failed; /* Boolean flag. */
47 static char gpfx; /* prefix for global symbol name (sometimes '_') */
48 static struct stat sb; /* Remember .st_size, etc. */
49 static const char *altmcount; /* alternate mcount symbol name */
50 static int warn_on_notrace_sect; /* warn when section has mcount not being recorded */
51 static void *file_map; /* pointer of the mapped file */
52 static void *file_end; /* pointer to the end of the mapped file */
53 static int file_updated; /* flag to state file was changed */
54 static void *file_ptr; /* current file pointer location */
55
56 static void *file_append; /* added to the end of the file */
57 static size_t file_append_size; /* how much is added to end of file */
58
59 /* Per-file resource cleanup when multiple files. */
file_append_cleanup(void)60 static void file_append_cleanup(void)
61 {
62 free(file_append);
63 file_append = NULL;
64 file_append_size = 0;
65 file_updated = 0;
66 }
67
mmap_cleanup(void)68 static void mmap_cleanup(void)
69 {
70 if (!mmap_failed)
71 munmap(file_map, sb.st_size);
72 else
73 free(file_map);
74 file_map = NULL;
75 }
76
77 /* ulseek, uwrite, ...: Check return value for errors. */
78
ulseek(off_t const offset,int const whence)79 static off_t ulseek(off_t const offset, int const whence)
80 {
81 switch (whence) {
82 case SEEK_SET:
83 file_ptr = file_map + offset;
84 break;
85 case SEEK_CUR:
86 file_ptr += offset;
87 break;
88 case SEEK_END:
89 file_ptr = file_map + (sb.st_size - offset);
90 break;
91 }
92 if (file_ptr < file_map) {
93 fprintf(stderr, "lseek: seek before file\n");
94 return -1;
95 }
96 return file_ptr - file_map;
97 }
98
uwrite(void const * const buf,size_t const count)99 static ssize_t uwrite(void const *const buf, size_t const count)
100 {
101 size_t cnt = count;
102 off_t idx = 0;
103
104 file_updated = 1;
105
106 if (file_ptr + count >= file_end) {
107 off_t aoffset = (file_ptr + count) - file_end;
108
109 if (aoffset > file_append_size) {
110 file_append = realloc(file_append, aoffset);
111 file_append_size = aoffset;
112 }
113 if (!file_append) {
114 perror("write");
115 file_append_cleanup();
116 mmap_cleanup();
117 return -1;
118 }
119 if (file_ptr < file_end) {
120 cnt = file_end - file_ptr;
121 } else {
122 cnt = 0;
123 idx = aoffset - count;
124 }
125 }
126
127 if (cnt)
128 memcpy(file_ptr, buf, cnt);
129
130 if (cnt < count)
131 memcpy(file_append + idx, buf + cnt, count - cnt);
132
133 file_ptr += count;
134 return count;
135 }
136
umalloc(size_t size)137 static void * umalloc(size_t size)
138 {
139 void *const addr = malloc(size);
140 if (addr == 0) {
141 fprintf(stderr, "malloc failed: %zu bytes\n", size);
142 file_append_cleanup();
143 mmap_cleanup();
144 return NULL;
145 }
146 return addr;
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 */
mmap_file(char const * fname)162 static void *mmap_file(char const *fname)
163 {
164 /* Avoid problems if early cleanup() */
165 fd_map = -1;
166 mmap_failed = 1;
167 file_map = NULL;
168 file_ptr = NULL;
169 file_updated = 0;
170 sb.st_size = 0;
171
172 fd_map = open(fname, O_RDONLY);
173 if (fd_map < 0) {
174 perror(fname);
175 return NULL;
176 }
177 if (fstat(fd_map, &sb) < 0) {
178 perror(fname);
179 goto out;
180 }
181 if (!S_ISREG(sb.st_mode)) {
182 fprintf(stderr, "not a regular file: %s\n", fname);
183 goto out;
184 }
185 file_map = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
186 fd_map, 0);
187 if (file_map == MAP_FAILED) {
188 mmap_failed = 1;
189 file_map = umalloc(sb.st_size);
190 if (!file_map) {
191 perror(fname);
192 goto out;
193 }
194 if (read(fd_map, file_map, sb.st_size) != sb.st_size) {
195 perror(fname);
196 free(file_map);
197 file_map = NULL;
198 goto out;
199 }
200 } else
201 mmap_failed = 0;
202 out:
203 close(fd_map);
204 fd_map = -1;
205
206 file_end = file_map + sb.st_size;
207
208 return file_map;
209 }
210
211
212 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
213 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
214 static unsigned char *ideal_nop;
215
216 static char rel_type_nop;
217
218 static int (*make_nop)(void *map, size_t const offset);
219
make_nop_x86(void * map,size_t const offset)220 static int make_nop_x86(void *map, size_t const offset)
221 {
222 uint32_t *ptr;
223 unsigned char *op;
224
225 /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
226 ptr = map + offset;
227 if (*ptr != 0)
228 return -1;
229
230 op = map + offset - 1;
231 if (*op != 0xe8)
232 return -1;
233
234 /* convert to nop */
235 if (ulseek(offset - 1, SEEK_SET) < 0)
236 return -1;
237 if (uwrite(ideal_nop, 5) < 0)
238 return -1;
239 return 0;
240 }
241
242 static unsigned char ideal_nop4_arm_le[4] = { 0x00, 0x00, 0xa0, 0xe1 }; /* mov r0, r0 */
243 static unsigned char ideal_nop4_arm_be[4] = { 0xe1, 0xa0, 0x00, 0x00 }; /* mov r0, r0 */
244 static unsigned char *ideal_nop4_arm;
245
246 static unsigned char bl_mcount_arm_le[4] = { 0xfe, 0xff, 0xff, 0xeb }; /* bl */
247 static unsigned char bl_mcount_arm_be[4] = { 0xeb, 0xff, 0xff, 0xfe }; /* bl */
248 static unsigned char *bl_mcount_arm;
249
250 static unsigned char push_arm_le[4] = { 0x04, 0xe0, 0x2d, 0xe5 }; /* push {lr} */
251 static unsigned char push_arm_be[4] = { 0xe5, 0x2d, 0xe0, 0x04 }; /* push {lr} */
252 static unsigned char *push_arm;
253
254 static unsigned char ideal_nop2_thumb_le[2] = { 0x00, 0xbf }; /* nop */
255 static unsigned char ideal_nop2_thumb_be[2] = { 0xbf, 0x00 }; /* nop */
256 static unsigned char *ideal_nop2_thumb;
257
258 static unsigned char push_bl_mcount_thumb_le[6] = { 0x00, 0xb5, 0xff, 0xf7, 0xfe, 0xff }; /* push {lr}, bl */
259 static unsigned char push_bl_mcount_thumb_be[6] = { 0xb5, 0x00, 0xf7, 0xff, 0xff, 0xfe }; /* push {lr}, bl */
260 static unsigned char *push_bl_mcount_thumb;
261
make_nop_arm(void * map,size_t const offset)262 static int make_nop_arm(void *map, size_t const offset)
263 {
264 char *ptr;
265 int cnt = 1;
266 int nop_size;
267 size_t off = offset;
268
269 ptr = map + offset;
270 if (memcmp(ptr, bl_mcount_arm, 4) == 0) {
271 if (memcmp(ptr - 4, push_arm, 4) == 0) {
272 off -= 4;
273 cnt = 2;
274 }
275 ideal_nop = ideal_nop4_arm;
276 nop_size = 4;
277 } else if (memcmp(ptr - 2, push_bl_mcount_thumb, 6) == 0) {
278 cnt = 3;
279 nop_size = 2;
280 off -= 2;
281 ideal_nop = ideal_nop2_thumb;
282 } else
283 return -1;
284
285 /* Convert to nop */
286 if (ulseek(off, SEEK_SET) < 0)
287 return -1;
288
289 do {
290 if (uwrite(ideal_nop, nop_size) < 0)
291 return -1;
292 } while (--cnt > 0);
293
294 return 0;
295 }
296
297 static unsigned char ideal_nop4_arm64[4] = {0x1f, 0x20, 0x03, 0xd5};
make_nop_arm64(void * map,size_t const offset)298 static int make_nop_arm64(void *map, size_t const offset)
299 {
300 uint32_t *ptr;
301
302 ptr = map + offset;
303 /* bl <_mcount> is 0x94000000 before relocation */
304 if (*ptr != 0x94000000)
305 return -1;
306
307 /* Convert to nop */
308 if (ulseek(offset, SEEK_SET) < 0)
309 return -1;
310 if (uwrite(ideal_nop, 4) < 0)
311 return -1;
312 return 0;
313 }
314
write_file(const char * fname)315 static int write_file(const char *fname)
316 {
317 char tmp_file[strlen(fname) + 4];
318 size_t n;
319
320 if (!file_updated)
321 return 0;
322
323 sprintf(tmp_file, "%s.rc", fname);
324
325 /*
326 * After reading the entire file into memory, delete it
327 * and write it back, to prevent weird side effects of modifying
328 * an object file in place.
329 */
330 fd_map = open(tmp_file, O_WRONLY | O_TRUNC | O_CREAT, sb.st_mode);
331 if (fd_map < 0) {
332 perror(fname);
333 return -1;
334 }
335 n = write(fd_map, file_map, sb.st_size);
336 if (n != sb.st_size) {
337 perror("write");
338 close(fd_map);
339 return -1;
340 }
341 if (file_append_size) {
342 n = write(fd_map, file_append, file_append_size);
343 if (n != file_append_size) {
344 perror("write");
345 close(fd_map);
346 return -1;
347 }
348 }
349 close(fd_map);
350 if (rename(tmp_file, fname) < 0) {
351 perror(fname);
352 return -1;
353 }
354 return 0;
355 }
356
357 /* w8rev, w8nat, ...: Handle endianness. */
358
w8rev(uint64_t const x)359 static uint64_t w8rev(uint64_t const x)
360 {
361 return ((0xff & (x >> (0 * 8))) << (7 * 8))
362 | ((0xff & (x >> (1 * 8))) << (6 * 8))
363 | ((0xff & (x >> (2 * 8))) << (5 * 8))
364 | ((0xff & (x >> (3 * 8))) << (4 * 8))
365 | ((0xff & (x >> (4 * 8))) << (3 * 8))
366 | ((0xff & (x >> (5 * 8))) << (2 * 8))
367 | ((0xff & (x >> (6 * 8))) << (1 * 8))
368 | ((0xff & (x >> (7 * 8))) << (0 * 8));
369 }
370
w4rev(uint32_t const x)371 static uint32_t w4rev(uint32_t const x)
372 {
373 return ((0xff & (x >> (0 * 8))) << (3 * 8))
374 | ((0xff & (x >> (1 * 8))) << (2 * 8))
375 | ((0xff & (x >> (2 * 8))) << (1 * 8))
376 | ((0xff & (x >> (3 * 8))) << (0 * 8));
377 }
378
w2rev(uint16_t const x)379 static uint32_t w2rev(uint16_t const x)
380 {
381 return ((0xff & (x >> (0 * 8))) << (1 * 8))
382 | ((0xff & (x >> (1 * 8))) << (0 * 8));
383 }
384
w8nat(uint64_t const x)385 static uint64_t w8nat(uint64_t const x)
386 {
387 return x;
388 }
389
w4nat(uint32_t const x)390 static uint32_t w4nat(uint32_t const x)
391 {
392 return x;
393 }
394
w2nat(uint16_t const x)395 static uint32_t w2nat(uint16_t const x)
396 {
397 return x;
398 }
399
400 static uint64_t (*w8)(uint64_t);
401 static uint32_t (*w)(uint32_t);
402 static uint32_t (*w2)(uint16_t);
403
404 /* Names of the sections that could contain calls to mcount. */
is_mcounted_section_name(char const * const txtname)405 static int is_mcounted_section_name(char const *const txtname)
406 {
407 return strncmp(".text", txtname, 5) == 0 ||
408 strcmp(".init.text", txtname) == 0 ||
409 strcmp(".ref.text", txtname) == 0 ||
410 strcmp(".sched.text", txtname) == 0 ||
411 strcmp(".spinlock.text", txtname) == 0 ||
412 strcmp(".irqentry.text", txtname) == 0 ||
413 strcmp(".softirqentry.text", txtname) == 0 ||
414 strcmp(".kprobes.text", txtname) == 0 ||
415 strcmp(".cpuidle.text", txtname) == 0 ||
416 (strncmp(".text.", txtname, 6) == 0 &&
417 strcmp(".text..ftrace", txtname) != 0);
418 }
419
420 static char const *already_has_rel_mcount = "success"; /* our work here is done! */
421
422 /* 32 bit and 64 bit are very similar */
423 #include "recordmcount.h"
424 #define RECORD_MCOUNT_64
425 #include "recordmcount.h"
426
arm_is_fake_mcount(Elf32_Rel const * rp)427 static int arm_is_fake_mcount(Elf32_Rel const *rp)
428 {
429 switch (ELF32_R_TYPE(w(rp->r_info))) {
430 case R_ARM_THM_CALL:
431 case R_ARM_CALL:
432 case R_ARM_PC24:
433 return 0;
434 }
435
436 return 1;
437 }
438
439 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
440 * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
441 * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
442 * to imply the order of the members; the spec does not say so.
443 * typedef unsigned char Elf64_Byte;
444 * fails on MIPS64 because their <elf.h> already has it!
445 */
446
447 typedef uint8_t myElf64_Byte; /* Type for a 8-bit quantity. */
448
449 union mips_r_info {
450 Elf64_Xword r_info;
451 struct {
452 Elf64_Word r_sym; /* Symbol index. */
453 myElf64_Byte r_ssym; /* Special symbol. */
454 myElf64_Byte r_type3; /* Third relocation. */
455 myElf64_Byte r_type2; /* Second relocation. */
456 myElf64_Byte r_type; /* First relocation. */
457 } r_mips;
458 };
459
MIPS64_r_sym(Elf64_Rel const * rp)460 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
461 {
462 return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
463 }
464
MIPS64_r_info(Elf64_Rel * const rp,unsigned sym,unsigned type)465 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
466 {
467 rp->r_info = ((union mips_r_info){
468 .r_mips = { .r_sym = w(sym), .r_type = type }
469 }).r_info;
470 }
471
do_file(char const * const fname)472 static int do_file(char const *const fname)
473 {
474 unsigned int reltype = 0;
475 Elf32_Ehdr *ehdr;
476 int rc = -1;
477
478 ehdr = mmap_file(fname);
479 if (!ehdr)
480 goto out;
481
482 w = w4nat;
483 w2 = w2nat;
484 w8 = w8nat;
485 switch (ehdr->e_ident[EI_DATA]) {
486 static unsigned int const endian = 1;
487 default:
488 fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
489 ehdr->e_ident[EI_DATA], fname);
490 goto out;
491 case ELFDATA2LSB:
492 if (*(unsigned char const *)&endian != 1) {
493 /* main() is big endian, file.o is little endian. */
494 w = w4rev;
495 w2 = w2rev;
496 w8 = w8rev;
497 }
498 ideal_nop4_arm = ideal_nop4_arm_le;
499 bl_mcount_arm = bl_mcount_arm_le;
500 push_arm = push_arm_le;
501 ideal_nop2_thumb = ideal_nop2_thumb_le;
502 push_bl_mcount_thumb = push_bl_mcount_thumb_le;
503 break;
504 case ELFDATA2MSB:
505 if (*(unsigned char const *)&endian != 0) {
506 /* main() is little endian, file.o is big endian. */
507 w = w4rev;
508 w2 = w2rev;
509 w8 = w8rev;
510 }
511 ideal_nop4_arm = ideal_nop4_arm_be;
512 bl_mcount_arm = bl_mcount_arm_be;
513 push_arm = push_arm_be;
514 ideal_nop2_thumb = ideal_nop2_thumb_be;
515 push_bl_mcount_thumb = push_bl_mcount_thumb_be;
516 break;
517 } /* end switch */
518 if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0 ||
519 w2(ehdr->e_type) != ET_REL ||
520 ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
521 fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
522 goto out;
523 }
524
525 gpfx = '_';
526 switch (w2(ehdr->e_machine)) {
527 default:
528 fprintf(stderr, "unrecognized e_machine %u %s\n",
529 w2(ehdr->e_machine), fname);
530 goto out;
531 case EM_386:
532 reltype = R_386_32;
533 rel_type_nop = R_386_NONE;
534 make_nop = make_nop_x86;
535 ideal_nop = ideal_nop5_x86_32;
536 mcount_adjust_32 = -1;
537 gpfx = 0;
538 break;
539 case EM_ARM:
540 reltype = R_ARM_ABS32;
541 altmcount = "__gnu_mcount_nc";
542 make_nop = make_nop_arm;
543 rel_type_nop = R_ARM_NONE;
544 is_fake_mcount32 = arm_is_fake_mcount;
545 gpfx = 0;
546 break;
547 case EM_AARCH64:
548 reltype = R_AARCH64_ABS64;
549 make_nop = make_nop_arm64;
550 rel_type_nop = R_AARCH64_NONE;
551 ideal_nop = ideal_nop4_arm64;
552 break;
553 case EM_IA_64: reltype = R_IA64_IMM64; break;
554 case EM_MIPS: /* reltype: e_class */ break;
555 case EM_PPC: reltype = R_PPC_ADDR32; break;
556 case EM_PPC64: reltype = R_PPC64_ADDR64; break;
557 case EM_S390: /* reltype: e_class */ break;
558 case EM_SH: reltype = R_SH_DIR32; gpfx = 0; break;
559 case EM_SPARCV9: reltype = R_SPARC_64; break;
560 case EM_X86_64:
561 make_nop = make_nop_x86;
562 ideal_nop = ideal_nop5_x86_64;
563 reltype = R_X86_64_64;
564 rel_type_nop = R_X86_64_NONE;
565 mcount_adjust_64 = -1;
566 gpfx = 0;
567 break;
568 } /* end switch */
569
570 switch (ehdr->e_ident[EI_CLASS]) {
571 default:
572 fprintf(stderr, "unrecognized ELF class %d %s\n",
573 ehdr->e_ident[EI_CLASS], fname);
574 goto out;
575 case ELFCLASS32:
576 if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
577 || w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
578 fprintf(stderr,
579 "unrecognized ET_REL file: %s\n", fname);
580 goto out;
581 }
582 if (w2(ehdr->e_machine) == EM_MIPS) {
583 reltype = R_MIPS_32;
584 is_fake_mcount32 = MIPS32_is_fake_mcount;
585 }
586 if (do32(ehdr, fname, reltype) < 0)
587 goto out;
588 break;
589 case ELFCLASS64: {
590 Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
591 if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
592 || w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
593 fprintf(stderr,
594 "unrecognized ET_REL file: %s\n", fname);
595 goto out;
596 }
597 if (w2(ghdr->e_machine) == EM_S390) {
598 reltype = R_390_64;
599 mcount_adjust_64 = -14;
600 }
601 if (w2(ghdr->e_machine) == EM_MIPS) {
602 reltype = R_MIPS_64;
603 Elf64_r_sym = MIPS64_r_sym;
604 Elf64_r_info = MIPS64_r_info;
605 is_fake_mcount64 = MIPS64_is_fake_mcount;
606 }
607 if (do64(ghdr, fname, reltype) < 0)
608 goto out;
609 break;
610 }
611 } /* end switch */
612
613 rc = write_file(fname);
614 out:
615 file_append_cleanup();
616 mmap_cleanup();
617 return rc;
618 }
619
main(int argc,char * argv[])620 int main(int argc, char *argv[])
621 {
622 const char ftrace[] = "/ftrace.o";
623 int ftrace_size = sizeof(ftrace) - 1;
624 int n_error = 0; /* gcc-4.3.0 false positive complaint */
625 int c;
626 int i;
627
628 while ((c = getopt(argc, argv, "w")) >= 0) {
629 switch (c) {
630 case 'w':
631 warn_on_notrace_sect = 1;
632 break;
633 default:
634 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
635 return 0;
636 }
637 }
638
639 if ((argc - optind) < 1) {
640 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
641 return 0;
642 }
643
644 /* Process each file in turn, allowing deep failure. */
645 for (i = optind; i < argc; i++) {
646 char *file = argv[i];
647 int len;
648
649 /*
650 * The file kernel/trace/ftrace.o references the mcount
651 * function but does not call it. Since ftrace.o should
652 * not be traced anyway, we just skip it.
653 */
654 len = strlen(file);
655 if (len >= ftrace_size &&
656 strcmp(file + (len - ftrace_size), ftrace) == 0)
657 continue;
658
659 if (do_file(file)) {
660 fprintf(stderr, "%s: failed\n", file);
661 ++n_error;
662 }
663 }
664 return !!n_error;
665 }
666