Linux Audio

Check our new training course

Loading...
  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
 36static int fd_map;	/* File descriptor for file being modified. */
 37static int mmap_failed; /* Boolean flag. */
 38static void *ehdr_curr; /* current ElfXX_Ehdr *  for resource cleanup */
 39static char gpfx;	/* prefix for global symbol name (sometimes '_') */
 40static struct stat sb;	/* Remember .st_size, etc. */
 41static jmp_buf jmpenv;	/* setjmp/longjmp per-file error escape */
 42static const char *altmcount;	/* alternate mcount symbol name */
 43static int warn_on_notrace_sect; /* warn when section has mcount not being recorded */
 44
 45/* setjmp() return values */
 46enum {
 47	SJ_SETJMP = 0,  /* hardwired first return */
 48	SJ_FAIL,
 49	SJ_SUCCEED
 50};
 51
 52/* Per-file resource cleanup when multiple files. */
 53static void
 54cleanup(void)
 55{
 56	if (!mmap_failed)
 57		munmap(ehdr_curr, sb.st_size);
 58	else
 59		free(ehdr_curr);
 60	close(fd_map);
 61}
 62
 63static void __attribute__((noreturn))
 64fail_file(void)
 65{
 66	cleanup();
 67	longjmp(jmpenv, SJ_FAIL);
 68}
 69
 70static void __attribute__((noreturn))
 71succeed_file(void)
 72{
 73	cleanup();
 74	longjmp(jmpenv, SJ_SUCCEED);
 75}
 76
 77/* ulseek, uread, ...:  Check return value for errors. */
 78
 79static off_t
 80ulseek(int const fd, off_t const offset, int const whence)
 81{
 82	off_t const w = lseek(fd, offset, whence);
 83	if (w == (off_t)-1) {
 84		perror("lseek");
 85		fail_file();
 86	}
 87	return w;
 88}
 89
 90static size_t
 91uread(int const fd, void *const buf, size_t const count)
 92{
 93	size_t const n = read(fd, buf, count);
 94	if (n != count) {
 95		perror("read");
 96		fail_file();
 97	}
 98	return n;
 99}
100
101static size_t
102uwrite(int const fd, void const *const buf, size_t const count)
103{
104	size_t const n = write(fd, buf, count);
105	if (n != count) {
106		perror("write");
107		fail_file();
108	}
109	return n;
110}
111
112static void *
113umalloc(size_t size)
114{
115	void *const addr = malloc(size);
116	if (addr == 0) {
117		fprintf(stderr, "malloc failed: %zu bytes\n", size);
118		fail_file();
119	}
120	return addr;
121}
122
123static unsigned char ideal_nop5_x86_64[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
124static unsigned char ideal_nop5_x86_32[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
125static unsigned char *ideal_nop;
126
127static char rel_type_nop;
128
129static int (*make_nop)(void *map, size_t const offset);
130
131static int make_nop_x86(void *map, size_t const offset)
132{
133	uint32_t *ptr;
134	unsigned char *op;
135
136	/* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
137	ptr = map + offset;
138	if (*ptr != 0)
139		return -1;
140
141	op = map + offset - 1;
142	if (*op != 0xe8)
143		return -1;
144
145	/* convert to nop */
146	ulseek(fd_map, offset - 1, SEEK_SET);
147	uwrite(fd_map, ideal_nop, 5);
148	return 0;
149}
150
151/*
152 * Get the whole file as a programming convenience in order to avoid
153 * malloc+lseek+read+free of many pieces.  If successful, then mmap
154 * avoids copying unused pieces; else just read the whole file.
155 * Open for both read and write; new info will be appended to the file.
156 * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
157 * do not propagate to the file until an explicit overwrite at the last.
158 * This preserves most aspects of consistency (all except .st_size)
159 * for simultaneous readers of the file while we are appending to it.
160 * However, multiple writers still are bad.  We choose not to use
161 * locking because it is expensive and the use case of kernel build
162 * makes multiple writers unlikely.
163 */
164static void *mmap_file(char const *fname)
165{
166	void *addr;
167
168	fd_map = open(fname, O_RDWR);
169	if (fd_map < 0 || fstat(fd_map, &sb) < 0) {
170		perror(fname);
171		fail_file();
172	}
173	if (!S_ISREG(sb.st_mode)) {
174		fprintf(stderr, "not a regular file: %s\n", fname);
175		fail_file();
176	}
177	addr = mmap(0, sb.st_size, PROT_READ|PROT_WRITE, MAP_PRIVATE,
178		    fd_map, 0);
179	mmap_failed = 0;
180	if (addr == MAP_FAILED) {
181		mmap_failed = 1;
182		addr = umalloc(sb.st_size);
183		uread(fd_map, addr, sb.st_size);
184	}
185	return addr;
186}
187
188/* w8rev, w8nat, ...: Handle endianness. */
189
190static uint64_t w8rev(uint64_t const x)
191{
192	return   ((0xff & (x >> (0 * 8))) << (7 * 8))
193	       | ((0xff & (x >> (1 * 8))) << (6 * 8))
194	       | ((0xff & (x >> (2 * 8))) << (5 * 8))
195	       | ((0xff & (x >> (3 * 8))) << (4 * 8))
196	       | ((0xff & (x >> (4 * 8))) << (3 * 8))
197	       | ((0xff & (x >> (5 * 8))) << (2 * 8))
198	       | ((0xff & (x >> (6 * 8))) << (1 * 8))
199	       | ((0xff & (x >> (7 * 8))) << (0 * 8));
200}
201
202static uint32_t w4rev(uint32_t const x)
203{
204	return   ((0xff & (x >> (0 * 8))) << (3 * 8))
205	       | ((0xff & (x >> (1 * 8))) << (2 * 8))
206	       | ((0xff & (x >> (2 * 8))) << (1 * 8))
207	       | ((0xff & (x >> (3 * 8))) << (0 * 8));
208}
209
210static uint32_t w2rev(uint16_t const x)
211{
212	return   ((0xff & (x >> (0 * 8))) << (1 * 8))
213	       | ((0xff & (x >> (1 * 8))) << (0 * 8));
214}
215
216static uint64_t w8nat(uint64_t const x)
217{
218	return x;
219}
220
221static uint32_t w4nat(uint32_t const x)
222{
223	return x;
224}
225
226static uint32_t w2nat(uint16_t const x)
227{
228	return x;
229}
230
231static uint64_t (*w8)(uint64_t);
232static uint32_t (*w)(uint32_t);
233static uint32_t (*w2)(uint16_t);
234
235/* Names of the sections that could contain calls to mcount. */
236static int
237is_mcounted_section_name(char const *const txtname)
238{
239	return strcmp(".text",           txtname) == 0 ||
240		strcmp(".ref.text",      txtname) == 0 ||
241		strcmp(".sched.text",    txtname) == 0 ||
242		strcmp(".spinlock.text", txtname) == 0 ||
243		strcmp(".irqentry.text", txtname) == 0 ||
244		strcmp(".kprobes.text", txtname) == 0 ||
245		strcmp(".text.unlikely", txtname) == 0;
246}
247
248/* 32 bit and 64 bit are very similar */
249#include "recordmcount.h"
250#define RECORD_MCOUNT_64
251#include "recordmcount.h"
252
253/* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
254 * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
255 * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
256 * to imply the order of the members; the spec does not say so.
257 *	typedef unsigned char Elf64_Byte;
258 * fails on MIPS64 because their <elf.h> already has it!
259 */
260
261typedef uint8_t myElf64_Byte;		/* Type for a 8-bit quantity.  */
262
263union mips_r_info {
264	Elf64_Xword r_info;
265	struct {
266		Elf64_Word r_sym;		/* Symbol index.  */
267		myElf64_Byte r_ssym;		/* Special symbol.  */
268		myElf64_Byte r_type3;		/* Third relocation.  */
269		myElf64_Byte r_type2;		/* Second relocation.  */
270		myElf64_Byte r_type;		/* First relocation.  */
271	} r_mips;
272};
273
274static uint64_t MIPS64_r_sym(Elf64_Rel const *rp)
275{
276	return w(((union mips_r_info){ .r_info = rp->r_info }).r_mips.r_sym);
277}
278
279static void MIPS64_r_info(Elf64_Rel *const rp, unsigned sym, unsigned type)
280{
281	rp->r_info = ((union mips_r_info){
282		.r_mips = { .r_sym = w(sym), .r_type = type }
283	}).r_info;
284}
285
286static void
287do_file(char const *const fname)
288{
289	Elf32_Ehdr *const ehdr = mmap_file(fname);
290	unsigned int reltype = 0;
291
292	ehdr_curr = ehdr;
293	w = w4nat;
294	w2 = w2nat;
295	w8 = w8nat;
296	switch (ehdr->e_ident[EI_DATA]) {
297		static unsigned int const endian = 1;
298	default:
299		fprintf(stderr, "unrecognized ELF data encoding %d: %s\n",
300			ehdr->e_ident[EI_DATA], fname);
301		fail_file();
302		break;
303	case ELFDATA2LSB:
304		if (*(unsigned char const *)&endian != 1) {
305			/* main() is big endian, file.o is little endian. */
306			w = w4rev;
307			w2 = w2rev;
308			w8 = w8rev;
309		}
310		break;
311	case ELFDATA2MSB:
312		if (*(unsigned char const *)&endian != 0) {
313			/* main() is little endian, file.o is big endian. */
314			w = w4rev;
315			w2 = w2rev;
316			w8 = w8rev;
317		}
318		break;
319	}  /* end switch */
320	if (memcmp(ELFMAG, ehdr->e_ident, SELFMAG) != 0
321	||  w2(ehdr->e_type) != ET_REL
322	||  ehdr->e_ident[EI_VERSION] != EV_CURRENT) {
323		fprintf(stderr, "unrecognized ET_REL file %s\n", fname);
324		fail_file();
325	}
326
327	gpfx = 0;
328	switch (w2(ehdr->e_machine)) {
329	default:
330		fprintf(stderr, "unrecognized e_machine %d %s\n",
331			w2(ehdr->e_machine), fname);
332		fail_file();
333		break;
334	case EM_386:
335		reltype = R_386_32;
336		make_nop = make_nop_x86;
337		ideal_nop = ideal_nop5_x86_32;
338		mcount_adjust_32 = -1;
339		break;
340	case EM_ARM:	 reltype = R_ARM_ABS32;
341			 altmcount = "__gnu_mcount_nc";
342			 break;
343	case EM_IA_64:	 reltype = R_IA64_IMM64;   gpfx = '_'; break;
344	case EM_MIPS:	 /* reltype: e_class    */ gpfx = '_'; break;
345	case EM_PPC:	 reltype = R_PPC_ADDR32;   gpfx = '_'; break;
346	case EM_PPC64:	 reltype = R_PPC64_ADDR64; gpfx = '_'; break;
347	case EM_S390:    /* reltype: e_class    */ gpfx = '_'; break;
348	case EM_SH:	 reltype = R_SH_DIR32;                 break;
349	case EM_SPARCV9: reltype = R_SPARC_64;     gpfx = '_'; break;
350	case EM_X86_64:
351		make_nop = make_nop_x86;
352		ideal_nop = ideal_nop5_x86_64;
353		reltype = R_X86_64_64;
354		mcount_adjust_64 = -1;
355		break;
356	}  /* end switch */
357
358	switch (ehdr->e_ident[EI_CLASS]) {
359	default:
360		fprintf(stderr, "unrecognized ELF class %d %s\n",
361			ehdr->e_ident[EI_CLASS], fname);
362		fail_file();
363		break;
364	case ELFCLASS32:
365		if (w2(ehdr->e_ehsize) != sizeof(Elf32_Ehdr)
366		||  w2(ehdr->e_shentsize) != sizeof(Elf32_Shdr)) {
367			fprintf(stderr,
368				"unrecognized ET_REL file: %s\n", fname);
369			fail_file();
370		}
371		if (w2(ehdr->e_machine) == EM_S390) {
372			reltype = R_390_32;
373			mcount_adjust_32 = -4;
374		}
375		if (w2(ehdr->e_machine) == EM_MIPS) {
376			reltype = R_MIPS_32;
377			is_fake_mcount32 = MIPS32_is_fake_mcount;
378		}
379		do32(ehdr, fname, reltype);
380		break;
381	case ELFCLASS64: {
382		Elf64_Ehdr *const ghdr = (Elf64_Ehdr *)ehdr;
383		if (w2(ghdr->e_ehsize) != sizeof(Elf64_Ehdr)
384		||  w2(ghdr->e_shentsize) != sizeof(Elf64_Shdr)) {
385			fprintf(stderr,
386				"unrecognized ET_REL file: %s\n", fname);
387			fail_file();
388		}
389		if (w2(ghdr->e_machine) == EM_S390) {
390			reltype = R_390_64;
391			mcount_adjust_64 = -8;
392		}
393		if (w2(ghdr->e_machine) == EM_MIPS) {
394			reltype = R_MIPS_64;
395			Elf64_r_sym = MIPS64_r_sym;
396			Elf64_r_info = MIPS64_r_info;
397			is_fake_mcount64 = MIPS64_is_fake_mcount;
398		}
399		do64(ghdr, fname, reltype);
400		break;
401	}
402	}  /* end switch */
403
404	cleanup();
405}
406
407int
408main(int argc, char *argv[])
409{
410	const char ftrace[] = "/ftrace.o";
411	int ftrace_size = sizeof(ftrace) - 1;
412	int n_error = 0;  /* gcc-4.3.0 false positive complaint */
413	int c;
414	int i;
415
416	while ((c = getopt(argc, argv, "w")) >= 0) {
417		switch (c) {
418		case 'w':
419			warn_on_notrace_sect = 1;
420			break;
421		default:
422			fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
423			return 0;
424		}
425	}
426
427	if ((argc - optind) < 1) {
428		fprintf(stderr, "usage: recordmcount [-w] file.o...\n");
429		return 0;
430	}
431
432	/* Process each file in turn, allowing deep failure. */
433	for (i = optind; i < argc; i++) {
434		char *file = argv[i];
435		int const sjval = setjmp(jmpenv);
436		int len;
437
438		/*
439		 * The file kernel/trace/ftrace.o references the mcount
440		 * function but does not call it. Since ftrace.o should
441		 * not be traced anyway, we just skip it.
442		 */
443		len = strlen(file);
444		if (len >= ftrace_size &&
445		    strcmp(file + (len - ftrace_size), ftrace) == 0)
446			continue;
447
448		switch (sjval) {
449		default:
450			fprintf(stderr, "internal error: %s\n", file);
451			exit(1);
452			break;
453		case SJ_SETJMP:    /* normal sequence */
454			/* Avoid problems if early cleanup() */
455			fd_map = -1;
456			ehdr_curr = NULL;
457			mmap_failed = 1;
458			do_file(file);
459			break;
460		case SJ_FAIL:    /* error in do_file or below */
461			++n_error;
462			break;
463		case SJ_SUCCEED:    /* premature success */
464			/* do nothing */
465			break;
466		}  /* end switch */
467	}
468	return !!n_error;
469}
470
471