diff options
author | John Reiser <jreiser@bitwagon.com> | 2010-10-13 15:12:54 -0400 |
---|---|---|
committer | Steven Rostedt <rostedt@goodmis.org> | 2010-10-14 16:44:34 -0400 |
commit | 81d3858d3131a589cade0d8b57f95cc1fc699b89 (patch) | |
tree | a79260842b6de29a065e52f891e1527336e6e999 /scripts/recordmcount.c | |
parent | 3cba11d32bb4b24c3ba257043595772df4b9c7b5 (diff) |
ftrace: Add C version of recordmcount compile time code
Currently, the mcount callers are found with a perl script that does
an objdump on every file in the kernel. This is a C version of that
same code which should increase the performance time of compiling
the kernel with dynamic ftrace enabled.
Signed-off-by: John Reiser <jreiser@bitwagon.com>
[ Updated the code to include .text.unlikely section as well as
changing the format to follow Linux coding style. ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Diffstat (limited to 'scripts/recordmcount.c')
-rw-r--r-- | scripts/recordmcount.c | 885 |
1 files changed, 885 insertions, 0 deletions
diff --git a/scripts/recordmcount.c b/scripts/recordmcount.c new file mode 100644 index 000000000000..34f32be17090 --- /dev/null +++ b/scripts/recordmcount.c | |||
@@ -0,0 +1,885 @@ | |||
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 | /* Append the new shstrtab, Elf32_Shdr[], __mcount_loc and its relocations. */ | ||
216 | static void append32(Elf32_Ehdr *const ehdr, | ||
217 | Elf32_Shdr *const shstr, | ||
218 | uint32_t const *const mloc0, | ||
219 | uint32_t const *const mlocp, | ||
220 | Elf32_Rel const *const mrel0, | ||
221 | Elf32_Rel const *const mrelp, | ||
222 | unsigned int const rel_entsize, | ||
223 | unsigned int const symsec_sh_link) | ||
224 | { | ||
225 | /* Begin constructing output file */ | ||
226 | Elf32_Shdr mcsec; | ||
227 | char const *mc_name = (sizeof(Elf32_Rela) == rel_entsize) | ||
228 | ? ".rela__mcount_loc" | ||
229 | : ".rel__mcount_loc"; | ||
230 | unsigned const old_shnum = w2(ehdr->e_shnum); | ||
231 | uint32_t const old_shoff = w(ehdr->e_shoff); | ||
232 | uint32_t const old_shstr_sh_size = w(shstr->sh_size); | ||
233 | uint32_t const old_shstr_sh_offset = w(shstr->sh_offset); | ||
234 | uint32_t t = 1 + strlen(mc_name) + w(shstr->sh_size); | ||
235 | uint32_t new_e_shoff; | ||
236 | |||
237 | shstr->sh_size = w(t); | ||
238 | shstr->sh_offset = w(sb.st_size); | ||
239 | t += sb.st_size; | ||
240 | t += (3u & -t); /* 4-byte align */ | ||
241 | new_e_shoff = t; | ||
242 | |||
243 | /* body for new shstrtab */ | ||
244 | ulseek(fd_map, sb.st_size, SEEK_SET); | ||
245 | uwrite(fd_map, old_shstr_sh_offset + (void *)ehdr, old_shstr_sh_size); | ||
246 | uwrite(fd_map, mc_name, 1 + strlen(mc_name)); | ||
247 | |||
248 | /* old(modified) Elf32_Shdr table, 4-byte aligned */ | ||
249 | ulseek(fd_map, t, SEEK_SET); | ||
250 | t += sizeof(Elf32_Shdr) * old_shnum; | ||
251 | uwrite(fd_map, old_shoff + (void *)ehdr, | ||
252 | sizeof(Elf32_Shdr) * old_shnum); | ||
253 | |||
254 | /* new sections __mcount_loc and .rel__mcount_loc */ | ||
255 | t += 2*sizeof(mcsec); | ||
256 | mcsec.sh_name = w((sizeof(Elf32_Rela) == rel_entsize) + strlen(".rel") | ||
257 | + old_shstr_sh_size); | ||
258 | mcsec.sh_type = w(SHT_PROGBITS); | ||
259 | mcsec.sh_flags = w(SHF_ALLOC); | ||
260 | mcsec.sh_addr = 0; | ||
261 | mcsec.sh_offset = w(t); | ||
262 | mcsec.sh_size = w((void *)mlocp - (void *)mloc0); | ||
263 | mcsec.sh_link = 0; | ||
264 | mcsec.sh_info = 0; | ||
265 | mcsec.sh_addralign = w(4); | ||
266 | mcsec.sh_entsize = w(4); | ||
267 | uwrite(fd_map, &mcsec, sizeof(mcsec)); | ||
268 | |||
269 | mcsec.sh_name = w(old_shstr_sh_size); | ||
270 | mcsec.sh_type = (sizeof(Elf32_Rela) == rel_entsize) | ||
271 | ? w(SHT_RELA) | ||
272 | : w(SHT_REL); | ||
273 | mcsec.sh_flags = 0; | ||
274 | mcsec.sh_addr = 0; | ||
275 | mcsec.sh_offset = w((void *)mlocp - (void *)mloc0 + t); | ||
276 | mcsec.sh_size = w((void *)mrelp - (void *)mrel0); | ||
277 | mcsec.sh_link = w(symsec_sh_link); | ||
278 | mcsec.sh_info = w(old_shnum); | ||
279 | mcsec.sh_addralign = w(4); | ||
280 | mcsec.sh_entsize = w(rel_entsize); | ||
281 | uwrite(fd_map, &mcsec, sizeof(mcsec)); | ||
282 | |||
283 | uwrite(fd_map, mloc0, (void *)mlocp - (void *)mloc0); | ||
284 | uwrite(fd_map, mrel0, (void *)mrelp - (void *)mrel0); | ||
285 | |||
286 | ehdr->e_shoff = w(new_e_shoff); | ||
287 | ehdr->e_shnum = w2(2 + w2(ehdr->e_shnum)); /* {.rel,}__mcount_loc */ | ||
288 | ulseek(fd_map, 0, SEEK_SET); | ||
289 | uwrite(fd_map, ehdr, sizeof(*ehdr)); | ||
290 | } | ||
291 | |||
292 | /* | ||
293 | * append64 and append32 (and other analogous pairs) could be templated | ||
294 | * using C++, but the complexity is high. (For an example, look at p_elf.h | ||
295 | * in the source for UPX, http://upx.sourceforge.net) So: remember to make | ||
296 | * the corresponding change in the routine for the other size. | ||
297 | */ | ||
298 | static void append64(Elf64_Ehdr *const ehdr, | ||
299 | Elf64_Shdr *const shstr, | ||
300 | uint64_t const *const mloc0, | ||
301 | uint64_t const *const mlocp, | ||
302 | Elf64_Rel const *const mrel0, | ||
303 | Elf64_Rel const *const mrelp, | ||
304 | unsigned int const rel_entsize, | ||
305 | unsigned int const symsec_sh_link) | ||
306 | { | ||
307 | /* Begin constructing output file */ | ||
308 | Elf64_Shdr mcsec; | ||
309 | char const *mc_name = (sizeof(Elf64_Rela) == rel_entsize) | ||
310 | ? ".rela__mcount_loc" | ||
311 | : ".rel__mcount_loc"; | ||
312 | unsigned const old_shnum = w2(ehdr->e_shnum); | ||
313 | uint64_t const old_shoff = w8(ehdr->e_shoff); | ||
314 | uint64_t const old_shstr_sh_size = w8(shstr->sh_size); | ||
315 | uint64_t const old_shstr_sh_offset = w8(shstr->sh_offset); | ||
316 | uint64_t t = 1 + strlen(mc_name) + w8(shstr->sh_size); | ||
317 | uint64_t new_e_shoff; | ||
318 | |||
319 | shstr->sh_size = w8(t); | ||
320 | shstr->sh_offset = w8(sb.st_size); | ||
321 | t += sb.st_size; | ||
322 | t += (7u & -t); /* 8-byte align */ | ||
323 | new_e_shoff = t; | ||
324 | |||
325 | /* body for new shstrtab */ | ||
326 | ulseek(fd_map, sb.st_size, SEEK_SET); | ||
327 | uwrite(fd_map, old_shstr_sh_offset + (void *)ehdr, old_shstr_sh_size); | ||
328 | uwrite(fd_map, mc_name, 1 + strlen(mc_name)); | ||
329 | |||
330 | /* old(modified) Elf64_Shdr table, 8-byte aligned */ | ||
331 | ulseek(fd_map, t, SEEK_SET); | ||
332 | t += sizeof(Elf64_Shdr) * old_shnum; | ||
333 | uwrite(fd_map, old_shoff + (void *)ehdr, | ||
334 | sizeof(Elf64_Shdr) * old_shnum); | ||
335 | |||
336 | /* new sections __mcount_loc and .rel__mcount_loc */ | ||
337 | t += 2*sizeof(mcsec); | ||
338 | mcsec.sh_name = w((sizeof(Elf64_Rela) == rel_entsize) + strlen(".rel") | ||
339 | + old_shstr_sh_size); | ||
340 | mcsec.sh_type = w(SHT_PROGBITS); | ||
341 | mcsec.sh_flags = w8(SHF_ALLOC); | ||
342 | mcsec.sh_addr = 0; | ||
343 | mcsec.sh_offset = w8(t); | ||
344 | mcsec.sh_size = w8((void *)mlocp - (void *)mloc0); | ||
345 | mcsec.sh_link = 0; | ||
346 | mcsec.sh_info = 0; | ||
347 | mcsec.sh_addralign = w8(8); | ||
348 | mcsec.sh_entsize = w8(8); | ||
349 | uwrite(fd_map, &mcsec, sizeof(mcsec)); | ||
350 | |||
351 | mcsec.sh_name = w(old_shstr_sh_size); | ||
352 | mcsec.sh_type = (sizeof(Elf64_Rela) == rel_entsize) | ||
353 | ? w(SHT_RELA) | ||
354 | : w(SHT_REL); | ||
355 | mcsec.sh_flags = 0; | ||
356 | mcsec.sh_addr = 0; | ||
357 | mcsec.sh_offset = w8((void *)mlocp - (void *)mloc0 + t); | ||
358 | mcsec.sh_size = w8((void *)mrelp - (void *)mrel0); | ||
359 | mcsec.sh_link = w(symsec_sh_link); | ||
360 | mcsec.sh_info = w(old_shnum); | ||
361 | mcsec.sh_addralign = w8(8); | ||
362 | mcsec.sh_entsize = w8(rel_entsize); | ||
363 | uwrite(fd_map, &mcsec, sizeof(mcsec)); | ||
364 | |||
365 | uwrite(fd_map, mloc0, (void *)mlocp - (void *)mloc0); | ||
366 | uwrite(fd_map, mrel0, (void *)mrelp - (void *)mrel0); | ||
367 | |||
368 | ehdr->e_shoff = w8(new_e_shoff); | ||
369 | ehdr->e_shnum = w2(2 + w2(ehdr->e_shnum)); /* {.rel,}__mcount_loc */ | ||
370 | ulseek(fd_map, 0, SEEK_SET); | ||
371 | uwrite(fd_map, ehdr, sizeof(*ehdr)); | ||
372 | } | ||
373 | |||
374 | /* | ||
375 | * Look at the relocations in order to find the calls to mcount. | ||
376 | * Accumulate the section offsets that are found, and their relocation info, | ||
377 | * onto the end of the existing arrays. | ||
378 | */ | ||
379 | static uint32_t *sift32_rel_mcount(uint32_t *mlocp, | ||
380 | unsigned const offbase, | ||
381 | Elf32_Rel **const mrelpp, | ||
382 | Elf32_Shdr const *const relhdr, | ||
383 | Elf32_Ehdr const *const ehdr, | ||
384 | unsigned const recsym, | ||
385 | uint32_t const recval, | ||
386 | unsigned const reltype) | ||
387 | { | ||
388 | uint32_t *const mloc0 = mlocp; | ||
389 | Elf32_Rel *mrelp = *mrelpp; | ||
390 | Elf32_Shdr *const shdr0 = (Elf32_Shdr *)(w(ehdr->e_shoff) | ||
391 | + (void *)ehdr); | ||
392 | unsigned const symsec_sh_link = w(relhdr->sh_link); | ||
393 | Elf32_Shdr const *const symsec = &shdr0[symsec_sh_link]; | ||
394 | Elf32_Sym const *const sym0 = (Elf32_Sym const *)(w(symsec->sh_offset) | ||
395 | + (void *)ehdr); | ||
396 | |||
397 | Elf32_Shdr const *const strsec = &shdr0[w(symsec->sh_link)]; | ||
398 | char const *const str0 = (char const *)(w(strsec->sh_offset) | ||
399 | + (void *)ehdr); | ||
400 | |||
401 | Elf32_Rel const *const rel0 = (Elf32_Rel const *)(w(relhdr->sh_offset) | ||
402 | + (void *)ehdr); | ||
403 | unsigned rel_entsize = w(relhdr->sh_entsize); | ||
404 | unsigned const nrel = w(relhdr->sh_size) / rel_entsize; | ||
405 | Elf32_Rel const *relp = rel0; | ||
406 | |||
407 | unsigned mcountsym = 0; | ||
408 | unsigned t; | ||
409 | |||
410 | for (t = nrel; t; --t) { | ||
411 | if (!mcountsym) { | ||
412 | Elf32_Sym const *const symp = | ||
413 | &sym0[ELF32_R_SYM(w(relp->r_info))]; | ||
414 | |||
415 | if (0 == strcmp((('_' == gpfx) ? "_mcount" : "mcount"), | ||
416 | &str0[w(symp->st_name)])) | ||
417 | mcountsym = ELF32_R_SYM(w(relp->r_info)); | ||
418 | } | ||
419 | if (mcountsym == ELF32_R_SYM(w(relp->r_info))) { | ||
420 | uint32_t const addend = w(w(relp->r_offset) - recval); | ||
421 | mrelp->r_offset = w(offbase | ||
422 | + ((void *)mlocp - (void *)mloc0)); | ||
423 | mrelp->r_info = w(ELF32_R_INFO(recsym, reltype)); | ||
424 | if (sizeof(Elf32_Rela) == rel_entsize) { | ||
425 | ((Elf32_Rela *)mrelp)->r_addend = addend; | ||
426 | *mlocp++ = 0; | ||
427 | } else | ||
428 | *mlocp++ = addend; | ||
429 | |||
430 | mrelp = (Elf32_Rel *)(rel_entsize + (void *)mrelp); | ||
431 | } | ||
432 | relp = (Elf32_Rel const *)(rel_entsize + (void *)relp); | ||
433 | } | ||
434 | *mrelpp = mrelp; | ||
435 | return mlocp; | ||
436 | } | ||
437 | |||
438 | static uint64_t *sift64_rel_mcount(uint64_t *mlocp, | ||
439 | unsigned const offbase, | ||
440 | Elf64_Rel **const mrelpp, | ||
441 | Elf64_Shdr const *const relhdr, | ||
442 | Elf64_Ehdr const *const ehdr, | ||
443 | unsigned const recsym, | ||
444 | uint64_t const recval, | ||
445 | unsigned const reltype) | ||
446 | { | ||
447 | uint64_t *const mloc0 = mlocp; | ||
448 | Elf64_Rel *mrelp = *mrelpp; | ||
449 | Elf64_Shdr *const shdr0 = (Elf64_Shdr *)(w8(ehdr->e_shoff) | ||
450 | + (void *)ehdr); | ||
451 | unsigned const symsec_sh_link = w(relhdr->sh_link); | ||
452 | Elf64_Shdr const *const symsec = &shdr0[symsec_sh_link]; | ||
453 | Elf64_Sym const *const sym0 = (Elf64_Sym const *)(w8(symsec->sh_offset) | ||
454 | + (void *)ehdr); | ||
455 | |||
456 | Elf64_Shdr const *const strsec = &shdr0[w(symsec->sh_link)]; | ||
457 | char const *const str0 = (char const *)(w8(strsec->sh_offset) | ||
458 | + (void *)ehdr); | ||
459 | |||
460 | Elf64_Rel const *const rel0 = (Elf64_Rel const *)(w8(relhdr->sh_offset) | ||
461 | + (void *)ehdr); | ||
462 | unsigned rel_entsize = w8(relhdr->sh_entsize); | ||
463 | unsigned const nrel = w8(relhdr->sh_size) / rel_entsize; | ||
464 | Elf64_Rel const *relp = rel0; | ||
465 | |||
466 | unsigned mcountsym = 0; | ||
467 | unsigned t; | ||
468 | |||
469 | for (t = nrel; 0 != t; --t) { | ||
470 | if (!mcountsym) { | ||
471 | Elf64_Sym const *const symp = | ||
472 | &sym0[ELF64_R_SYM(w8(relp->r_info))]; | ||
473 | char const *symname = &str0[w(symp->st_name)]; | ||
474 | |||
475 | if ('.' == symname[0]) | ||
476 | ++symname; /* ppc64 hack */ | ||
477 | if (0 == strcmp((('_' == gpfx) ? "_mcount" : "mcount"), | ||
478 | symname)) | ||
479 | mcountsym = ELF64_R_SYM(w8(relp->r_info)); | ||
480 | } | ||
481 | |||
482 | if (mcountsym == ELF64_R_SYM(w8(relp->r_info))) { | ||
483 | uint64_t const addend = w8(w8(relp->r_offset) - recval); | ||
484 | |||
485 | mrelp->r_offset = w8(offbase | ||
486 | + ((void *)mlocp - (void *)mloc0)); | ||
487 | mrelp->r_info = w8(ELF64_R_INFO(recsym, reltype)); | ||
488 | if (sizeof(Elf64_Rela) == rel_entsize) { | ||
489 | ((Elf64_Rela *)mrelp)->r_addend = addend; | ||
490 | *mlocp++ = 0; | ||
491 | } else | ||
492 | *mlocp++ = addend; | ||
493 | |||
494 | mrelp = (Elf64_Rel *)(rel_entsize + (void *)mrelp); | ||
495 | } | ||
496 | relp = (Elf64_Rel const *)(rel_entsize + (void *)relp); | ||
497 | } | ||
498 | *mrelpp = mrelp; | ||
499 | |||
500 | return mlocp; | ||
501 | } | ||
502 | |||
503 | /* | ||
504 | * Find a symbol in the given section, to be used as the base for relocating | ||
505 | * the table of offsets of calls to mcount. A local or global symbol suffices, | ||
506 | * but avoid a Weak symbol because it may be overridden; the change in value | ||
507 | * would invalidate the relocations of the offsets of the calls to mcount. | ||
508 | * Often the found symbol will be the unnamed local symbol generated by | ||
509 | * GNU 'as' for the start of each section. For example: | ||
510 | * Num: Value Size Type Bind Vis Ndx Name | ||
511 | * 2: 00000000 0 SECTION LOCAL DEFAULT 1 | ||
512 | */ | ||
513 | static unsigned find32_secsym_ndx(unsigned const txtndx, | ||
514 | char const *const txtname, | ||
515 | uint32_t *const recvalp, | ||
516 | Elf32_Shdr const *const symhdr, | ||
517 | Elf32_Ehdr const *const ehdr) | ||
518 | { | ||
519 | Elf32_Sym const *const sym0 = (Elf32_Sym const *)(w(symhdr->sh_offset) | ||
520 | + (void *)ehdr); | ||
521 | unsigned const nsym = w(symhdr->sh_size) / w(symhdr->sh_entsize); | ||
522 | Elf32_Sym const *symp; | ||
523 | unsigned t; | ||
524 | |||
525 | for (symp = sym0, t = nsym; t; --t, ++symp) { | ||
526 | unsigned int const st_bind = ELF32_ST_BIND(symp->st_info); | ||
527 | |||
528 | if (txtndx == w2(symp->st_shndx) | ||
529 | /* avoid STB_WEAK */ | ||
530 | && (STB_LOCAL == st_bind || STB_GLOBAL == st_bind)) { | ||
531 | *recvalp = w(symp->st_value); | ||
532 | return symp - sym0; | ||
533 | } | ||
534 | } | ||
535 | fprintf(stderr, "Cannot find symbol for section %d: %s.\n", | ||
536 | txtndx, txtname); | ||
537 | fail_file(); | ||
538 | } | ||
539 | |||
540 | static unsigned find64_secsym_ndx(unsigned const txtndx, | ||
541 | char const *const txtname, | ||
542 | uint64_t *const recvalp, | ||
543 | Elf64_Shdr const *const symhdr, | ||
544 | Elf64_Ehdr const *const ehdr) | ||
545 | { | ||
546 | Elf64_Sym const *const sym0 = (Elf64_Sym const *)(w8(symhdr->sh_offset) | ||
547 | + (void *)ehdr); | ||
548 | unsigned const nsym = w8(symhdr->sh_size) / w8(symhdr->sh_entsize); | ||
549 | Elf64_Sym const *symp; | ||
550 | unsigned t; | ||
551 | |||
552 | for (symp = sym0, t = nsym; t; --t, ++symp) { | ||
553 | unsigned int const st_bind = ELF64_ST_BIND(symp->st_info); | ||
554 | |||
555 | if (txtndx == w2(symp->st_shndx) | ||
556 | /* avoid STB_WEAK */ | ||
557 | && (STB_LOCAL == st_bind || STB_GLOBAL == st_bind)) { | ||
558 | *recvalp = w8(symp->st_value); | ||
559 | return symp - sym0; | ||
560 | } | ||
561 | } | ||
562 | fprintf(stderr, "Cannot find symbol for section %d: %s.\n", | ||
563 | txtndx, txtname); | ||
564 | fail_file(); | ||
565 | } | ||
566 | |||
567 | /* | ||
568 | * Evade ISO C restriction: no declaration after statement in | ||
569 | * has32_rel_mcount. | ||
570 | */ | ||
571 | static char const * | ||
572 | __has32_rel_mcount(Elf32_Shdr const *const relhdr, /* is SHT_REL or SHT_RELA */ | ||
573 | Elf32_Shdr const *const shdr0, | ||
574 | char const *const shstrtab, | ||
575 | char const *const fname) | ||
576 | { | ||
577 | /* .sh_info depends on .sh_type == SHT_REL[,A] */ | ||
578 | Elf32_Shdr const *const txthdr = &shdr0[w(relhdr->sh_info)]; | ||
579 | char const *const txtname = &shstrtab[w(txthdr->sh_name)]; | ||
580 | |||
581 | if (0 == strcmp("__mcount_loc", txtname)) { | ||
582 | fprintf(stderr, "warning: __mcount_loc already exists: %s\n", | ||
583 | fname); | ||
584 | succeed_file(); | ||
585 | } | ||
586 | if (SHT_PROGBITS != w(txthdr->sh_type) || | ||
587 | !is_mcounted_section_name(txtname)) | ||
588 | return NULL; | ||
589 | return txtname; | ||
590 | } | ||
591 | |||
592 | static char const *has32_rel_mcount(Elf32_Shdr const *const relhdr, | ||
593 | Elf32_Shdr const *const shdr0, | ||
594 | char const *const shstrtab, | ||
595 | char const *const fname) | ||
596 | { | ||
597 | if (SHT_REL != w(relhdr->sh_type) && SHT_RELA != w(relhdr->sh_type)) | ||
598 | return NULL; | ||
599 | return __has32_rel_mcount(relhdr, shdr0, shstrtab, fname); | ||
600 | } | ||
601 | |||
602 | static char const *__has64_rel_mcount(Elf64_Shdr const *const relhdr, | ||
603 | Elf64_Shdr const *const shdr0, | ||
604 | char const *const shstrtab, | ||
605 | char const *const fname) | ||
606 | { | ||
607 | /* .sh_info depends on .sh_type == SHT_REL[,A] */ | ||
608 | Elf64_Shdr const *const txthdr = &shdr0[w(relhdr->sh_info)]; | ||
609 | char const *const txtname = &shstrtab[w(txthdr->sh_name)]; | ||
610 | |||
611 | if (0 == strcmp("__mcount_loc", txtname)) { | ||
612 | fprintf(stderr, "warning: __mcount_loc already exists: %s\n", | ||
613 | fname); | ||
614 | succeed_file(); | ||
615 | } | ||
616 | if (SHT_PROGBITS != w(txthdr->sh_type) || | ||
617 | !is_mcounted_section_name(txtname)) | ||
618 | return NULL; | ||
619 | return txtname; | ||
620 | } | ||
621 | |||
622 | static char const *has64_rel_mcount(Elf64_Shdr const *const relhdr, | ||
623 | Elf64_Shdr const *const shdr0, | ||
624 | char const *const shstrtab, | ||
625 | char const *const fname) | ||
626 | { | ||
627 | if (SHT_REL != w(relhdr->sh_type) && SHT_RELA != w(relhdr->sh_type)) | ||
628 | return NULL; | ||
629 | return __has64_rel_mcount(relhdr, shdr0, shstrtab, fname); | ||
630 | } | ||
631 | |||
632 | static unsigned tot32_relsize(Elf32_Shdr const *const shdr0, | ||
633 | unsigned nhdr, | ||
634 | const char *const shstrtab, | ||
635 | const char *const fname) | ||
636 | { | ||
637 | unsigned totrelsz = 0; | ||
638 | Elf32_Shdr const *shdrp = shdr0; | ||
639 | for (; 0 != nhdr; --nhdr, ++shdrp) { | ||
640 | if (has32_rel_mcount(shdrp, shdr0, shstrtab, fname)) | ||
641 | totrelsz += w(shdrp->sh_size); | ||
642 | } | ||
643 | return totrelsz; | ||
644 | } | ||
645 | |||
646 | static unsigned tot64_relsize(Elf64_Shdr const *const shdr0, | ||
647 | unsigned nhdr, | ||
648 | const char *const shstrtab, | ||
649 | const char *const fname) | ||
650 | { | ||
651 | unsigned totrelsz = 0; | ||
652 | Elf64_Shdr const *shdrp = shdr0; | ||
653 | |||
654 | for (; nhdr; --nhdr, ++shdrp) { | ||
655 | if (has64_rel_mcount(shdrp, shdr0, shstrtab, fname)) | ||
656 | totrelsz += w8(shdrp->sh_size); | ||
657 | } | ||
658 | return totrelsz; | ||
659 | } | ||
660 | |||
661 | /* Overall supervision for Elf32 ET_REL file. */ | ||
662 | static void | ||
663 | do32(Elf32_Ehdr *const ehdr, char const *const fname, unsigned const reltype) | ||
664 | { | ||
665 | Elf32_Shdr *const shdr0 = (Elf32_Shdr *)(w(ehdr->e_shoff) | ||
666 | + (void *)ehdr); | ||
667 | unsigned const nhdr = w2(ehdr->e_shnum); | ||
668 | Elf32_Shdr *const shstr = &shdr0[w2(ehdr->e_shstrndx)]; | ||
669 | char const *const shstrtab = (char const *)(w(shstr->sh_offset) | ||
670 | + (void *)ehdr); | ||
671 | |||
672 | Elf32_Shdr const *relhdr; | ||
673 | unsigned k; | ||
674 | |||
675 | /* Upper bound on space: assume all relevant relocs are for mcount. */ | ||
676 | unsigned const totrelsz = tot32_relsize(shdr0, nhdr, shstrtab, fname); | ||
677 | Elf32_Rel *const mrel0 = umalloc(totrelsz); | ||
678 | Elf32_Rel * mrelp = mrel0; | ||
679 | |||
680 | /* 2*sizeof(address) <= sizeof(Elf32_Rel) */ | ||
681 | uint32_t *const mloc0 = umalloc(totrelsz>>1); | ||
682 | uint32_t * mlocp = mloc0; | ||
683 | |||
684 | unsigned rel_entsize = 0; | ||
685 | unsigned symsec_sh_link = 0; | ||
686 | |||
687 | for (relhdr = shdr0, k = nhdr; k; --k, ++relhdr) { | ||
688 | char const *const txtname = has32_rel_mcount(relhdr, shdr0, | ||
689 | shstrtab, fname); | ||
690 | if (txtname) { | ||
691 | uint32_t recval = 0; | ||
692 | unsigned const recsym = find32_secsym_ndx( | ||
693 | w(relhdr->sh_info), txtname, &recval, | ||
694 | &shdr0[symsec_sh_link = w(relhdr->sh_link)], | ||
695 | ehdr); | ||
696 | |||
697 | rel_entsize = w(relhdr->sh_entsize); | ||
698 | mlocp = sift32_rel_mcount(mlocp, | ||
699 | (void *)mlocp - (void *)mloc0, &mrelp, | ||
700 | relhdr, ehdr, recsym, recval, reltype); | ||
701 | } | ||
702 | } | ||
703 | if (mloc0 != mlocp) { | ||
704 | append32(ehdr, shstr, mloc0, mlocp, mrel0, mrelp, | ||
705 | rel_entsize, symsec_sh_link); | ||
706 | } | ||
707 | free(mrel0); | ||
708 | free(mloc0); | ||
709 | } | ||
710 | |||
711 | static void | ||
712 | do64(Elf64_Ehdr *const ehdr, char const *const fname, unsigned const reltype) | ||
713 | { | ||
714 | Elf64_Shdr *const shdr0 = (Elf64_Shdr *)(w8(ehdr->e_shoff) | ||
715 | + (void *)ehdr); | ||
716 | unsigned const nhdr = w2(ehdr->e_shnum); | ||
717 | Elf64_Shdr *const shstr = &shdr0[w2(ehdr->e_shstrndx)]; | ||
718 | char const *const shstrtab = (char const *)(w8(shstr->sh_offset) | ||
719 | + (void *)ehdr); | ||
720 | |||
721 | Elf64_Shdr const *relhdr; | ||
722 | unsigned k; | ||
723 | |||
724 | /* Upper bound on space: assume all relevant relocs are for mcount. */ | ||
725 | unsigned const totrelsz = tot64_relsize(shdr0, nhdr, shstrtab, fname); | ||
726 | Elf64_Rel *const mrel0 = umalloc(totrelsz); | ||
727 | Elf64_Rel * mrelp = mrel0; | ||
728 | |||
729 | /* 2*sizeof(address) <= sizeof(Elf64_Rel) */ | ||
730 | uint64_t *const mloc0 = umalloc(totrelsz>>1); | ||
731 | uint64_t * mlocp = mloc0; | ||
732 | |||
733 | unsigned rel_entsize = 0; | ||
734 | unsigned symsec_sh_link = 0; | ||
735 | |||
736 | for ((relhdr = shdr0), k = nhdr; k; --k, ++relhdr) { | ||
737 | char const *const txtname = has64_rel_mcount(relhdr, shdr0, | ||
738 | shstrtab, fname); | ||
739 | if (txtname) { | ||
740 | uint64_t recval = 0; | ||
741 | unsigned const recsym = find64_secsym_ndx( | ||
742 | w(relhdr->sh_info), txtname, &recval, | ||
743 | &shdr0[symsec_sh_link = w(relhdr->sh_link)], | ||
744 | ehdr); | ||
745 | |||
746 | rel_entsize = w8(relhdr->sh_entsize); | ||
747 | mlocp = sift64_rel_mcount(mlocp, | ||
748 | (void *)mlocp - (void *)mloc0, &mrelp, | ||
749 | relhdr, ehdr, recsym, recval, reltype); | ||
750 | } | ||
751 | } | ||
752 | if (mloc0 != mlocp) { | ||
753 | append64(ehdr, shstr, mloc0, mlocp, mrel0, mrelp, | ||
754 | rel_entsize, symsec_sh_link); | ||
755 | } | ||
756 | free(mrel0); | ||
757 | free(mloc0); | ||
758 | } | ||
759 | |||
760 | static void | ||
761 | do_file(char const *const fname) | ||
762 | { | ||
763 | Elf32_Ehdr *const ehdr = mmap_file(fname); | ||
764 | unsigned int reltype = 0; | ||
765 | |||
766 | ehdr_curr = ehdr; | ||
767 | w = w4nat; | ||
768 | w2 = w2nat; | ||
769 | w8 = w8nat; | ||
770 | switch (ehdr->e_ident[EI_DATA]) { | ||
771 | static unsigned int const endian = 1; | ||
772 | default: { | ||
773 | fprintf(stderr, "unrecognized ELF data encoding %d: %s\n", | ||
774 | ehdr->e_ident[EI_DATA], fname); | ||
775 | fail_file(); | ||
776 | } break; | ||
777 | case ELFDATA2LSB: { | ||
778 | if (1 != *(unsigned char const *)&endian) { | ||
779 | /* main() is big endian, file.o is little endian. */ | ||
780 | w = w4rev; | ||
781 | w2 = w2rev; | ||
782 | w8 = w8rev; | ||
783 | } | ||
784 | } break; | ||
785 | case ELFDATA2MSB: { | ||
786 | if (0 != *(unsigned char const *)&endian) { | ||
787 | /* main() is little endian, file.o is big endian. */ | ||
788 | w = w4rev; | ||
789 | w2 = w2rev; | ||
790 | w8 = w8rev; | ||
791 | } | ||
792 | } break; | ||
793 | } /* end switch */ | ||
794 | if (0 != memcmp(ELFMAG, ehdr->e_ident, SELFMAG) | ||
795 | || ET_REL != w2(ehdr->e_type) | ||
796 | || EV_CURRENT != ehdr->e_ident[EI_VERSION]) { | ||
797 | fprintf(stderr, "unrecognized ET_REL file %s\n", fname); | ||
798 | fail_file(); | ||
799 | } | ||
800 | |||
801 | gpfx = 0; | ||
802 | switch (w2(ehdr->e_machine)) { | ||
803 | default: { | ||
804 | fprintf(stderr, "unrecognized e_machine %d %s\n", | ||
805 | w2(ehdr->e_machine), fname); | ||
806 | fail_file(); | ||
807 | } break; | ||
808 | case EM_386: reltype = R_386_32; break; | ||
809 | case EM_ARM: reltype = R_ARM_ABS32; break; | ||
810 | case EM_IA_64: reltype = R_IA64_IMM64; gpfx = '_'; break; | ||
811 | case EM_PPC: reltype = R_PPC_ADDR32; gpfx = '_'; break; | ||
812 | case EM_PPC64: reltype = R_PPC64_ADDR64; gpfx = '_'; break; | ||
813 | case EM_S390: /* reltype: e_class */ gpfx = '_'; break; | ||
814 | case EM_SH: reltype = R_SH_DIR32; break; | ||
815 | case EM_SPARCV9: reltype = R_SPARC_64; gpfx = '_'; break; | ||
816 | case EM_X86_64: reltype = R_X86_64_64; break; | ||
817 | } /* end switch */ | ||
818 | |||
819 | switch (ehdr->e_ident[EI_CLASS]) { | ||
820 | default: { | ||
821 | fprintf(stderr, "unrecognized ELF class %d %s\n", | ||
822 | ehdr->e_ident[EI_CLASS], fname); | ||
823 | fail_file(); | ||
824 | } break; | ||
825 | case ELFCLASS32: { | ||
826 | if (sizeof(Elf32_Ehdr) != w2(ehdr->e_ehsize) | ||
827 | || sizeof(Elf32_Shdr) != w2(ehdr->e_shentsize)) { | ||
828 | fprintf(stderr, | ||
829 | "unrecognized ET_REL file: %s\n", fname); | ||
830 | fail_file(); | ||
831 | } | ||
832 | if (EM_S390 == w2(ehdr->e_machine)) | ||
833 | reltype = R_390_32; | ||
834 | do32(ehdr, fname, reltype); | ||
835 | } break; | ||
836 | case ELFCLASS64: { | ||
837 | Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr; | ||
838 | if (sizeof(Elf64_Ehdr) != w2(ghdr->e_ehsize) | ||
839 | || sizeof(Elf64_Shdr) != w2(ghdr->e_shentsize)) { | ||
840 | fprintf(stderr, | ||
841 | "unrecognized ET_REL file: %s\n", fname); | ||
842 | fail_file(); | ||
843 | } | ||
844 | if (EM_S390 == w2(ghdr->e_machine)) | ||
845 | reltype = R_390_64; | ||
846 | do64(ghdr, fname, reltype); | ||
847 | } break; | ||
848 | } /* end switch */ | ||
849 | |||
850 | cleanup(); | ||
851 | } | ||
852 | |||
853 | int | ||
854 | main(int argc, char const *argv[]) | ||
855 | { | ||
856 | int n_error = 0; /* gcc-4.3.0 false positive complaint */ | ||
857 | if (argc <= 1) | ||
858 | fprintf(stderr, "usage: recordmcount file.o...\n"); | ||
859 | else /* Process each file in turn, allowing deep failure. */ | ||
860 | for (--argc, ++argv; 0 < argc; --argc, ++argv) { | ||
861 | int const sjval = setjmp(jmpenv); | ||
862 | switch (sjval) { | ||
863 | default: { | ||
864 | fprintf(stderr, "internal error: %s\n", argv[0]); | ||
865 | exit(1); | ||
866 | } break; | ||
867 | case SJ_SETJMP: { /* normal sequence */ | ||
868 | /* Avoid problems if early cleanup() */ | ||
869 | fd_map = -1; | ||
870 | ehdr_curr = NULL; | ||
871 | mmap_failed = 1; | ||
872 | do_file(argv[0]); | ||
873 | } break; | ||
874 | case SJ_FAIL: { /* error in do_file or below */ | ||
875 | ++n_error; | ||
876 | } break; | ||
877 | case SJ_SUCCEED: { /* premature success */ | ||
878 | /* do nothing */ | ||
879 | } break; | ||
880 | } /* end switch */ | ||
881 | } | ||
882 | return !!n_error; | ||
883 | } | ||
884 | |||
885 | |||