xref: /linux/scripts/recordmcount.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 #ifndef EM_METAG
37 /* Remove this when these make it to the standard system elf.h. */
38 #define EM_METAG      174
39 #define R_METAG_ADDR32                   2
40 #define R_METAG_NONE                     3
41 #endif
42 
43 #ifndef EM_AARCH64
44 #define EM_AARCH64	183
45 #define R_AARCH64_ABS64	257
46 #endif
47 
48 static int fd_map;	/* File descriptor for file being modified. */
49 static int mmap_failed; /* Boolean flag. */
50 static void *ehdr_curr; /* current ElfXX_Ehdr *  for resource cleanup */
51 static char gpfx;	/* prefix for global symbol name (sometimes '_') */
52 static struct stat sb;	/* Remember .st_size, etc. */
53 static jmp_buf jmpenv;	/* setjmp/longjmp per-file error escape */
54 static const char *altmcount;	/* alternate mcount symbol name */
55 static int warn_on_notrace_sect; /* warn when section has mcount not being recorded */
56 
57 /* setjmp() return values */
58 enum {
59 	SJ_SETJMP = 0,  /* hardwired first return */
60 	SJ_FAIL,
61 	SJ_SUCCEED
62 };
63 
64 /* Per-file resource cleanup when multiple files. */
65 static void
66 cleanup(void)
67 {
68 	if (!mmap_failed)
69 		munmap(ehdr_curr, sb.st_size);
70 	else
71 		free(ehdr_curr);
72 	close(fd_map);
73 }
74 
75 static void __attribute__((noreturn))
76 fail_file(void)
77 {
78 	cleanup();
79 	longjmp(jmpenv, SJ_FAIL);
80 }
81 
82 static void __attribute__((noreturn))
83 succeed_file(void)
84 {
85 	cleanup();
86 	longjmp(jmpenv, SJ_SUCCEED);
87 }
88 
89 /* ulseek, uread, ...:  Check return value for errors. */
90 
91 static off_t
92 ulseek(int const fd, off_t const offset, int const whence)
93 {
94 	off_t const w = lseek(fd, offset, whence);
95 	if (w == (off_t)-1) {
96 		perror("lseek");
97 		fail_file();
98 	}
99 	return w;
100 }
101 
102 static size_t
103 uread(int const fd, void *const buf, size_t const count)
104 {
105 	size_t const n = read(fd, buf, count);
106 	if (n != count) {
107 		perror("read");
108 		fail_file();
109 	}
110 	return n;
111 }
112 
113 static size_t
114 uwrite(int const fd, void const *const buf, size_t const count)
115 {
116 	size_t const n = write(fd, buf, count);
117 	if (n != count) {
118 		perror("write");
119 		fail_file();
120 	}
121 	return n;
122 }
123 
124 static void *
125 umalloc(size_t size)
126 {
127 	void *const addr = malloc(size);
128 	if (addr == 0) {
129 		fprintf(stderr, "malloc failed: %zu bytes\n", size);
130 		fail_file();
131 	}
132 	return addr;
133 }
134 
135 static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
136 static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
137 static unsigned char *ideal_nop;
138 
139 static char rel_type_nop;
140 
141 static int (*make_nop)(void *map, size_t const offset);
142 
143 static int make_nop_x86(void *map, size_t const offset)
144 {
145 	uint32_t *ptr;
146 	unsigned char *op;
147 
148 	/* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
149 	ptr = map + offset;
150 	if (*ptr != 0)
151 		return -1;
152 
153 	op = map + offset - 1;
154 	if (*op != 0xe8)
155 		return -1;
156 
157 	/* convert to nop */
158 	ulseek(fd_map, offset - 1, SEEK_SET);
159 	uwrite(fd_map, ideal_nop, 5);
160 	return 0;
161 }
162 
163 /*
164  * Get the whole file as a programming convenience in order to avoid
165  * malloc+lseek+read+free of many pieces.  If successful, then mmap
166  * avoids copying unused pieces; else just read the whole file.
167  * Open for both read and write; new info will be appended to the file.
168  * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
169  * do not propagate to the file until an explicit overwrite at the last.
170  * This preserves most aspects of consistency (all except .st_size)
171  * for simultaneous readers of the file while we are appending to it.
172  * However, multiple writers still are bad.  We choose not to use
173  * locking because it is expensive and the use case of kernel build
174  * makes multiple writers unlikely.
175  */
176 static void *mmap_file(char const *fname)
177 {
178 	void *addr;
179 
180 	fd_map = open(fname, O_RDWR);
181 	if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
182 		perror(fname);
183 		fail_file();
184 	}
185 	if (!S_ISREG(sb.st_mode)) {
186 		fprintf(stderr, "not a regular file: %s\n", fname);
187 		fail_file();
188 	}
189 	addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
190 		    fd_map, 0);
191 	mmap_failed = 0;
192 	if (addr == MAP_FAILED) {
193 		mmap_failed = 1;
194 		addr = umalloc(sb.st_size);
195 		uread(fd_map, addr, sb.st_size);
196 	}
197 	return addr;
198 }
199 
200 /* w8rev, w8nat, ...: Handle endianness. */
201 
202 static uint64_t w8rev(uint64_t const x)
203 {
204 	return   ((0xff & (x >> (0 * 8))) << (7 * 8))
205 	       | ((0xff & (x >> (1 * 8))) << (6 * 8))
206 	       | ((0xff & (x >> (2 * 8))) << (5 * 8))
207 	       | ((0xff & (x >> (3 * 8))) << (4 * 8))
208 	       | ((0xff & (x >> (4 * 8))) << (3 * 8))
209 	       | ((0xff & (x >> (5 * 8))) << (2 * 8))
210 	       | ((0xff & (x >> (6 * 8))) << (1 * 8))
211 	       | ((0xff & (x >> (7 * 8))) << (0 * 8));
212 }
213 
214 static uint32_t w4rev(uint32_t const x)
215 {
216 	return   ((0xff & (x >> (0 * 8))) << (3 * 8))
217 	       | ((0xff & (x >> (1 * 8))) << (2 * 8))
218 	       | ((0xff & (x >> (2 * 8))) << (1 * 8))
219 	       | ((0xff & (x >> (3 * 8))) << (0 * 8));
220 }
221 
222 static uint32_t w2rev(uint16_t const x)
223 {
224 	return   ((0xff & (x >> (0 * 8))) << (1 * 8))
225 	       | ((0xff & (x >> (1 * 8))) << (0 * 8));
226 }
227 
228 static uint64_t w8nat(uint64_t const x)
229 {
230 	return x;
231 }
232 
233 static uint32_t w4nat(uint32_t const x)
234 {
235 	return x;
236 }
237 
238 static uint32_t w2nat(uint16_t const x)
239 {
240 	return x;
241 }
242 
243 static uint64_t (*w8)(uint64_t);
244 static uint32_t (*w)(uint32_t);
245 static uint32_t (*w2)(uint16_t);
246 
247 /* Names of the sections that could contain calls to mcount. */
248 static int
249 is_mcounted_section_name(char const *const txtname)
250 {
251 	return strcmp(".text",           txtname) == 0 ||
252 		strcmp(".ref.text",      txtname) == 0 ||
253 		strcmp(".sched.text",    txtname) == 0 ||
254 		strcmp(".spinlock.text", txtname) == 0 ||
255 		strcmp(".irqentry.text", txtname) == 0 ||
256 		strcmp(".kprobes.text", txtname) == 0 ||
257 		strcmp(".text.unlikely", txtname) == 0;
258 }
259 
260 /* 32 bit and 64 bit are very similar */
261 #include "recordmcount.h"
262 #define RECORD_MCOUNT_64
263 #include "recordmcount.h"
264 
265 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
266  * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
267  * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
268  * to imply the order of the members; the spec does not say so.
269  *	typedef unsigned char Elf64_Byte;
270  * fails on MIPS64 because their <elf.h> already has it!
271  */
272 
273 typedef uint8_t myElf64_Byte;		/* Type for a 8-bit quantity.  */
274 
275 union mips_r_info {
276 	Elf64_Xword r_info;
277 	struct {
278 		Elf64_Word r_sym;		/* Symbol index.  */
279 		myElf64_Byte r_ssym;		/* Special symbol.  */
280 		myElf64_Byte r_type3;		/* Third relocation.  */
281 		myElf64_Byte r_type2;		/* Second relocation.  */
282 		myElf64_Byte r_type;		/* First relocation.  */
283 	} r_mips;
284 };
285 
286 static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
287 {
288 	return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
289 }
290 
291 static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
292 {
293 	rp->r_info = ((union mips_r_info){
294 		.r_mips = { .r_sym = w(sym), .r_type = type }
295 	}).r_info;
296 }
297 
298 static void
299 do_file(char const *const fname)
300 {
301 	Elf32_Ehdr *const ehdr = mmap_file(fname);
302 	unsigned int reltype = 0;
303 
304 	ehdr_curr = ehdr;
305 	w = w4nat;
306 	w2 = w2nat;
307 	w8 = w8nat;
308 	switch (ehdr->e_ident[EI_DATA]) {
309 		static unsigned int const endian = 1;
310 	default:
311 		fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
312 			ehdr->e_ident[EI_DATA], fname);
313 		fail_file();
314 		break;
315 	case ELFDATA2LSB:
316 		if (*(unsigned char const *)&endian != 1) {
317 			/* main() is big endian, file.o is little endian. */
318 			w = w4rev;
319 			w2 = w2rev;
320 			w8 = w8rev;
321 		}
322 		break;
323 	case ELFDATA2MSB:
324 		if (*(unsigned char const *)&endian != 0) {
325 			/* main() is little endian, file.o is big endian. */
326 			w = w4rev;
327 			w2 = w2rev;
328 			w8 = w8rev;
329 		}
330 		break;
331 	}  /* end switch */
332 	if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
333 	||  w2(ehdr->e_type) != ET_REL
334 	||  ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
335 		fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
336 		fail_file();
337 	}
338 
339 	gpfx = 0;
340 	switch (w2(ehdr->e_machine)) {
341 	default:
342 		fprintf(stderr, "unrecognized e_machine %d %s\n",
343 			w2(ehdr->e_machine), fname);
344 		fail_file();
345 		break;
346 	case EM_386:
347 		reltype = R_386_32;
348 		make_nop = make_nop_x86;
349 		ideal_nop = ideal_nop5_x86_32;
350 		mcount_adjust_32 = -1;
351 		break;
352 	case EM_ARM:	 reltype = R_ARM_ABS32;
353 			 altmcount = "__gnu_mcount_nc";
354 			 break;
355 	case EM_AARCH64:
356 			 reltype = R_AARCH64_ABS64; gpfx = '_'; break;
357 	case EM_IA_64:	 reltype = R_IA64_IMM64;   gpfx = '_'; break;
358 	case EM_METAG:	 reltype = R_METAG_ADDR32;
359 			 altmcount = "_mcount_wrapper";
360 			 rel_type_nop = R_METAG_NONE;
361 			 /* We happen to have the same requirement as MIPS */
362 			 is_fake_mcount32 = MIPS32_is_fake_mcount;
363 			 break;
364 	case EM_MIPS:	 /* reltype: e_class    */ gpfx = '_'; break;
365 	case EM_PPC:	 reltype = R_PPC_ADDR32;   gpfx = '_'; break;
366 	case EM_PPC64:	 reltype = R_PPC64_ADDR64; gpfx = '_'; break;
367 	case EM_S390:    /* reltype: e_class    */ gpfx = '_'; break;
368 	case EM_SH:	 reltype = R_SH_DIR32;                 break;
369 	case EM_SPARCV9: reltype = R_SPARC_64;     gpfx = '_'; break;
370 	case EM_X86_64:
371 		make_nop = make_nop_x86;
372 		ideal_nop = ideal_nop5_x86_64;
373 		reltype = R_X86_64_64;
374 		mcount_adjust_64 = -1;
375 		break;
376 	}  /* end switch */
377 
378 	switch (ehdr->e_ident[EI_CLASS]) {
379 	default:
380 		fprintf(stderr, "unrecognized ELF class %d %s\n",
381 			ehdr->e_ident[EI_CLASS], fname);
382 		fail_file();
383 		break;
384 	case ELFCLASS32:
385 		if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
386 		||  w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
387 			fprintf(stderr,
388 				"unrecognized ET_REL file: %s\n", fname);
389 			fail_file();
390 		}
391 		if (w2(ehdr->e_machine) == EM_MIPS) {
392 			reltype = R_MIPS_32;
393 			is_fake_mcount32 = MIPS32_is_fake_mcount;
394 		}
395 		do32(ehdr, fname, reltype);
396 		break;
397 	case ELFCLASS64: {
398 		Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
399 		if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
400 		||  w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
401 			fprintf(stderr,
402 				"unrecognized ET_REL file: %s\n", fname);
403 			fail_file();
404 		}
405 		if (w2(ghdr->e_machine) == EM_S390) {
406 			reltype = R_390_64;
407 			mcount_adjust_64 = -14;
408 		}
409 		if (w2(ghdr->e_machine) == EM_MIPS) {
410 			reltype = R_MIPS_64;
411 			Elf64_r_sym = MIPS64_r_sym;
412 			Elf64_r_info = MIPS64_r_info;
413 			is_fake_mcount64 = MIPS64_is_fake_mcount;
414 		}
415 		do64(ghdr, fname, reltype);
416 		break;
417 	}
418 	}  /* end switch */
419 
420 	cleanup();
421 }
422 
423 int
424 main(int argc, char *argv[])
425 {
426 	const char ftrace[] = "/ftrace.o";
427 	int ftrace_size = sizeof(ftrace) - 1;
428 	int n_error = 0;  /* gcc-4.3.0 false positive complaint */
429 	int c;
430 	int i;
431 
432 	while ((c = getopt(argc, argv, "w")) >= 0) {
433 		switch (c) {
434 		case 'w':
435 			warn_on_notrace_sect = 1;
436 			break;
437 		default:
438 			fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
439 			return 0;
440 		}
441 	}
442 
443 	if ((argc - optind) < 1) {
444 		fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
445 		return 0;
446 	}
447 
448 	/* Process each file in turn, allowing deep failure. */
449 	for (i = optind; i < argc; i++) {
450 		char *file = argv[i];
451 		int const sjval = setjmp(jmpenv);
452 		int len;
453 
454 		/*
455 		 * The file kernel/trace/ftrace.o references the mcount
456 		 * function but does not call it. Since ftrace.o should
457 		 * not be traced anyway, we just skip it.
458 		 */
459 		len = strlen(file);
460 		if (len >= ftrace_size &&
461 		    strcmp(file + (len - ftrace_size), ftrace) == 0)
462 			continue;
463 
464 		switch (sjval) {
465 		default:
466 			fprintf(stderr, "internal error: %s\n", file);
467 			exit(1);
468 			break;
469 		case SJ_SETJMP:    /* normal sequence */
470 			/* Avoid problems if early cleanup() */
471 			fd_map = -1;
472 			ehdr_curr = NULL;
473 			mmap_failed = 1;
474 			do_file(file);
475 			break;
476 		case SJ_FAIL:    /* error in do_file or below */
477 			++n_error;
478 			break;
479 		case SJ_SUCCEED:    /* premature success */
480 			/* do nothing */
481 			break;
482 		}  /* end switch */
483 	}
484 	return !!n_error;
485 }
486