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
50 #ifndef EM_AARCH64
51 #define EM_AARCH64 183
52 #define R_AARCH64_ABS64 257
53 #endif
54
55 static int fd_map; /* File descriptor for file being modified. */
56 static int mmap_failed; /* Boolean flag. */
57 static void *ehdr_curr; /* current ElfXX_Ehdr * for resource cleanup */
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
64 /* setjmp() return values */
65 enum {
66 SJ_SETJMP = 0, /* hardwired first return */
67 SJ_FAIL,
68 SJ_SUCCEED
69 };
70
71 /* Per-file resource cleanup when multiple files. */
72 static void
cleanup(void)73 cleanup(void)
74 {
75 if (!mmap_failed)
76 munmap(ehdr_curr, sb.st_size);
77 else
78 free(ehdr_curr);
79 close(fd_map);
80 }
81
82 static void __attribute__((noreturn))
fail_file(void)83 fail_file(void)
84 {
85 cleanup();
86 longjmp(jmpenv, SJ_FAIL);
87 }
88
89 static void __attribute__((noreturn))
succeed_file(void)90 succeed_file(void)
91 {
92 cleanup();
93 longjmp(jmpenv, SJ_SUCCEED);
94 }
95
96 /* ulseek, uread, ...: Check return value for errors. */
97
98 static off_t
ulseek(int const fd,off_t const offset,int const whence)99 ulseek(int const fd, off_t const offset, int const whence)
100 {
101 off_t const w = lseek(fd, offset, whence);
102 if (w == (off_t)-1) {
103 perror("lseek");
104 fail_file();
105 }
106 return w;
107 }
108
109 static size_t
uread(int const fd,void * const buf,size_t const count)110 uread(int const fd, void *const buf, size_t const count)
111 {
112 size_t const n = read(fd, buf, count);
113 if (n != count) {
114 perror("read");
115 fail_file();
116 }
117 return n;
118 }
119
120 static size_t
uwrite(int const fd,void const * const buf,size_t const count)121 uwrite(int const fd, void const *const buf, size_t const count)
122 {
123 size_t const n = write(fd, buf, count);
124 if (n != count) {
125 perror("write");
126 fail_file();
127 }
128 return n;
129 }
130
131 static void *
umalloc(size_t size)132 umalloc(size_t size)
133 {
134 void *const addr = malloc(size);
135 if (addr == 0) {
136 fprintf(stderr, "malloc failed: %zu bytes\n", size);
137 fail_file();
138 }
139 return addr;
140 }
141
142 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
143 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
144 static unsigned char *ideal_nop;
145
146 static char rel_type_nop;
147
148 static int (*make_nop)(void *map, size_t const offset);
149
make_nop_x86(void * map,size_t const offset)150 static int make_nop_x86(void *map, size_t const offset)
151 {
152 uint32_t *ptr;
153 unsigned char *op;
154
155 /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
156 ptr = map + offset;
157 if (*ptr != 0)
158 return -1;
159
160 op = map + offset - 1;
161 if (*op != 0xe8)
162 return -1;
163
164 /* convert to nop */
165 ulseek(fd_map, offset - 1, SEEK_SET);
166 uwrite(fd_map, ideal_nop, 5);
167 return 0;
168 }
169
170 /*
171 * Get the whole file as a programming convenience in order to avoid
172 * malloc+lseek+read+free of many pieces. If successful, then mmap
173 * avoids copying unused pieces; else just read the whole file.
174 * Open for both read and write; new info will be appended to the file.
175 * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
176 * do not propagate to the file until an explicit overwrite at the last.
177 * This preserves most aspects of consistency (all except .st_size)
178 * for simultaneous readers of the file while we are appending to it.
179 * However, multiple writers still are bad. We choose not to use
180 * locking because it is expensive and the use case of kernel build
181 * makes multiple writers unlikely.
182 */
mmap_file(char const * fname)183 static void *mmap_file(char const *fname)
184 {
185 void *addr;
186
187 fd_map = open(fname, O_RDWR);
188 if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
189 perror(fname);
190 fail_file();
191 }
192 if (!S_ISREG(sb.st_mode)) {
193 fprintf(stderr, "not a regular file: %s\n", fname);
194 fail_file();
195 }
196 addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
197 fd_map, 0);
198 mmap_failed = 0;
199 if (addr == MAP_FAILED) {
200 mmap_failed = 1;
201 addr = umalloc(sb.st_size);
202 uread(fd_map, addr, sb.st_size);
203 }
204 return addr;
205 }
206
207 /* w8rev, w8nat, ...: Handle endianness. */
208
w8rev(uint64_t const x)209 static uint64_t w8rev(uint64_t const x)
210 {
211 return ((0xff & (x >> (0 * 8))) << (7 * 8))
212 | ((0xff & (x >> (1 * 8))) << (6 * 8))
213 | ((0xff & (x >> (2 * 8))) << (5 * 8))
214 | ((0xff & (x >> (3 * 8))) << (4 * 8))
215 | ((0xff & (x >> (4 * 8))) << (3 * 8))
216 | ((0xff & (x >> (5 * 8))) << (2 * 8))
217 | ((0xff & (x >> (6 * 8))) << (1 * 8))
218 | ((0xff & (x >> (7 * 8))) << (0 * 8));
219 }
220
w4rev(uint32_t const x)221 static uint32_t w4rev(uint32_t const x)
222 {
223 return ((0xff & (x >> (0 * 8))) << (3 * 8))
224 | ((0xff & (x >> (1 * 8))) << (2 * 8))
225 | ((0xff & (x >> (2 * 8))) << (1 * 8))
226 | ((0xff & (x >> (3 * 8))) << (0 * 8));
227 }
228
w2rev(uint16_t const x)229 static uint32_t w2rev(uint16_t const x)
230 {
231 return ((0xff & (x >> (0 * 8))) << (1 * 8))
232 | ((0xff & (x >> (1 * 8))) << (0 * 8));
233 }
234
w8nat(uint64_t const x)235 static uint64_t w8nat(uint64_t const x)
236 {
237 return x;
238 }
239
w4nat(uint32_t const x)240 static uint32_t w4nat(uint32_t const x)
241 {
242 return x;
243 }
244
w2nat(uint16_t const x)245 static uint32_t w2nat(uint16_t const x)
246 {
247 return x;
248 }
249
250 static uint64_t (*w8)(uint64_t);
251 static uint32_t (*w)(uint32_t);
252 static uint32_t (*w2)(uint16_t);
253
254 /* Names of the sections that could contain calls to mcount. */
255 static int
is_mcounted_section_name(char const * const txtname)256 is_mcounted_section_name(char const *const txtname)
257 {
258 return strcmp(".text", txtname) == 0 ||
259 strcmp(".ref.text", txtname) == 0 ||
260 strcmp(".sched.text", txtname) == 0 ||
261 strcmp(".spinlock.text", txtname) == 0 ||
262 strcmp(".irqentry.text", txtname) == 0 ||
263 strcmp(".kprobes.text", txtname) == 0 ||
264 strcmp(".text.unlikely", txtname) == 0;
265 }
266
267 /* 32 bit and 64 bit are very similar */
268 #include "recordmcount.h"
269 #define RECORD_MCOUNT_64
270 #include "recordmcount.h"
271
272 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
273 * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
274 * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
275 * to imply the order of the members; the spec does not say so.
276 * typedef unsigned char Elf64_Byte;
277 * fails on MIPS64 because their <elf.h> already has it!
278 */
279
280 typedef uint8_t myElf64_Byte; /* Type for a 8-bit quantity. */
281
282 union mips_r_info {
283 Elf64_Xword r_info;
284 struct {
285 Elf64_Word r_sym; /* Symbol index. */
286 myElf64_Byte r_ssym; /* Special symbol. */
287 myElf64_Byte r_type3; /* Third relocation. */
288 myElf64_Byte r_type2; /* Second relocation. */
289 myElf64_Byte r_type; /* First relocation. */
290 } r_mips;
291 };
292
MIPS64_r_sym(Elf64_Rel const * rp)293 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
294 {
295 return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
296 }
297
MIPS64_r_info(Elf64_Rel * const rp,unsigned sym,unsigned type)298 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
299 {
300 rp->r_info = ((union mips_r_info){
301 .r_mips = { .r_sym = w(sym), .r_type = type }
302 }).r_info;
303 }
304
305 static void
do_file(char const * const fname)306 do_file(char const *const fname)
307 {
308 Elf32_Ehdr *const ehdr = mmap_file(fname);
309 unsigned int reltype = 0;
310
311 ehdr_curr = ehdr;
312 w = w4nat;
313 w2 = w2nat;
314 w8 = w8nat;
315 switch (ehdr->e_ident[EI_DATA]) {
316 static unsigned int const endian = 1;
317 default:
318 fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
319 ehdr->e_ident[EI_DATA], fname);
320 fail_file();
321 break;
322 case ELFDATA2LSB:
323 if (*(unsigned char const *)&endian != 1) {
324 /* main() is big endian, file.o is little endian. */
325 w = w4rev;
326 w2 = w2rev;
327 w8 = w8rev;
328 }
329 break;
330 case ELFDATA2MSB:
331 if (*(unsigned char const *)&endian != 0) {
332 /* main() is little endian, file.o is big endian. */
333 w = w4rev;
334 w2 = w2rev;
335 w8 = w8rev;
336 }
337 break;
338 } /* end switch */
339 if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
340 || w2(ehdr->e_type) != ET_REL
341 || ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
342 fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
343 fail_file();
344 }
345
346 gpfx = 0;
347 switch (w2(ehdr->e_machine)) {
348 default:
349 fprintf(stderr, "unrecognized e_machine %d %s\n",
350 w2(ehdr->e_machine), fname);
351 fail_file();
352 break;
353 case EM_386:
354 reltype = R_386_32;
355 make_nop = make_nop_x86;
356 ideal_nop = ideal_nop5_x86_32;
357 mcount_adjust_32 = -1;
358 break;
359 case EM_ARM: reltype = R_ARM_ABS32;
360 altmcount = "__gnu_mcount_nc";
361 break;
362 case EM_AARCH64:
363 reltype = R_AARCH64_ABS64; gpfx = '_'; break;
364 case EM_IA_64: reltype = R_IA64_IMM64; gpfx = '_'; break;
365 case EM_METAG: reltype = R_METAG_ADDR32;
366 altmcount = "_mcount_wrapper";
367 rel_type_nop = R_METAG_NONE;
368 /* We happen to have the same requirement as MIPS */
369 is_fake_mcount32 = MIPS32_is_fake_mcount;
370 break;
371 case EM_MIPS: /* reltype: e_class */ gpfx = '_'; break;
372 case EM_PPC: reltype = R_PPC_ADDR32; gpfx = '_'; break;
373 case EM_PPC64: reltype = R_PPC64_ADDR64; gpfx = '_'; break;
374 case EM_S390: /* reltype: e_class */ gpfx = '_'; break;
375 case EM_SH: reltype = R_SH_DIR32; break;
376 case EM_SPARCV9: reltype = R_SPARC_64; gpfx = '_'; break;
377 case EM_X86_64:
378 make_nop = make_nop_x86;
379 ideal_nop = ideal_nop5_x86_64;
380 reltype = R_X86_64_64;
381 mcount_adjust_64 = -1;
382 break;
383 } /* end switch */
384
385 switch (ehdr->e_ident[EI_CLASS]) {
386 default:
387 fprintf(stderr, "unrecognized ELF class %d %s\n",
388 ehdr->e_ident[EI_CLASS], fname);
389 fail_file();
390 break;
391 case ELFCLASS32:
392 if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
393 || w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
394 fprintf(stderr,
395 "unrecognized ET_REL file: %s\n", fname);
396 fail_file();
397 }
398 if (w2(ehdr->e_machine) == EM_MIPS) {
399 reltype = R_MIPS_32;
400 is_fake_mcount32 = MIPS32_is_fake_mcount;
401 }
402 do32(ehdr, fname, reltype);
403 break;
404 case ELFCLASS64: {
405 Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
406 if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
407 || w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
408 fprintf(stderr,
409 "unrecognized ET_REL file: %s\n", fname);
410 fail_file();
411 }
412 if (w2(ghdr->e_machine) == EM_S390) {
413 reltype = R_390_64;
414 mcount_adjust_64 = -8;
415 }
416 if (w2(ghdr->e_machine) == EM_MIPS) {
417 reltype = R_MIPS_64;
418 Elf64_r_sym = MIPS64_r_sym;
419 Elf64_r_info = MIPS64_r_info;
420 is_fake_mcount64 = MIPS64_is_fake_mcount;
421 }
422 do64(ghdr, fname, reltype);
423 break;
424 }
425 } /* end switch */
426
427 cleanup();
428 }
429
430 int
main(int argc,char * argv[])431 main(int argc, char *argv[])
432 {
433 const char ftrace[] = "/ftrace.o";
434 int ftrace_size = sizeof(ftrace) - 1;
435 int n_error = 0; /* gcc-4.3.0 false positive complaint */
436 int c;
437 int i;
438
439 while ((c = getopt(argc, argv, "w")) >= 0) {
440 switch (c) {
441 case 'w':
442 warn_on_notrace_sect = 1;
443 break;
444 default:
445 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
446 return 0;
447 }
448 }
449
450 if ((argc - optind) < 1) {
451 fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
452 return 0;
453 }
454
455 /* Process each file in turn, allowing deep failure. */
456 for (i = optind; i < argc; i++) {
457 char *file = argv[i];
458 int const sjval = setjmp(jmpenv);
459 int len;
460
461 /*
462 * The file kernel/trace/ftrace.o references the mcount
463 * function but does not call it. Since ftrace.o should
464 * not be traced anyway, we just skip it.
465 */
466 len = strlen(file);
467 if (len >= ftrace_size &&
468 strcmp(file + (len - ftrace_size), ftrace) == 0)
469 continue;
470
471 switch (sjval) {
472 default:
473 fprintf(stderr, "internal error: %s\n", file);
474 exit(1);
475 break;
476 case SJ_SETJMP: /* normal sequence */
477 /* Avoid problems if early cleanup() */
478 fd_map = -1;
479 ehdr_curr = NULL;
480 mmap_failed = 1;
481 do_file(file);
482 break;
483 case SJ_FAIL: /* error in do_file or below */
484 ++n_error;
485 break;
486 case SJ_SUCCEED: /* premature success */
487 /* do nothing */
488 break;
489 } /* end switch */
490 }
491 return !!n_error;
492 }
493