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).
7 * Restructured to fit Linux format, as well as other updates:
8 * Copyright 2010 Steven Rostedt <srostedt@redhat.com>, Red Hat Inc.
12 * Strategy: alter the .o file in-place.
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.]
24 #include <sys/types.h>
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.
43 #ifndef R_METAG_ADDR32
44 #define R_METAG_ADDR32 2
47 #define R_METAG_NONE 3
51 #define EM_AARCH64 183
52 #define R_AARCH64_NONE 0
53 #define R_AARCH64_ABS64 257
56 static int fd_map
; /* File descriptor for file being modified. */
57 static int mmap_failed
; /* Boolean flag. */
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 static void *file_map
; /* pointer of the mapped file */
64 static void *file_end
; /* pointer to the end of the mapped file */
65 static int file_updated
; /* flag to state file was changed */
66 static void *file_ptr
; /* current file pointer location */
67 static void *file_append
; /* added to the end of the file */
68 static size_t file_append_size
; /* how much is added to end of file */
70 /* setjmp() return values */
72 SJ_SETJMP
= 0, /* hardwired first return */
77 /* Per-file resource cleanup when multiple files. */
82 munmap(file_map
, sb
.st_size
);
92 static void __attribute__((noreturn
))
96 longjmp(jmpenv
, SJ_FAIL
);
99 static void __attribute__((noreturn
))
103 longjmp(jmpenv
, SJ_SUCCEED
);
106 /* ulseek, uread, ...: Check return value for errors. */
109 ulseek(int const fd
, off_t
const offset
, int const whence
)
113 file_ptr
= file_map
+ offset
;
119 file_ptr
= file_map
+ (sb
.st_size
- offset
);
122 if (file_ptr
< file_map
) {
123 fprintf(stderr
, "lseek: seek before file\n");
126 return file_ptr
- file_map
;
130 uread(int const fd
, void *const buf
, size_t const count
)
132 size_t const n
= read(fd
, buf
, count
);
141 uwrite(int const fd
, void const *const buf
, size_t const count
)
148 if (file_ptr
+ count
>= file_end
) {
149 off_t aoffset
= (file_ptr
+ count
) - file_end
;
151 if (aoffset
> file_append_size
) {
152 file_append
= realloc(file_append
, aoffset
);
153 file_append_size
= aoffset
;
159 if (file_ptr
< file_end
) {
160 cnt
= file_end
- file_ptr
;
163 idx
= aoffset
- count
;
168 memcpy(file_ptr
, buf
, cnt
);
171 memcpy(file_append
+ idx
, buf
+ cnt
, count
- cnt
);
180 void *const addr
= malloc(size
);
182 fprintf(stderr
, "malloc failed: %zu bytes\n", size
);
188 static unsigned char ideal_nop5_x86_64
[5] = { 0x0f, 0x1f, 0x44, 0x00, 0x00 };
189 static unsigned char ideal_nop5_x86_32
[5] = { 0x3e, 0x8d, 0x74, 0x26, 0x00 };
190 static unsigned char *ideal_nop
;
192 static char rel_type_nop
;
194 static int (*make_nop
)(void *map
, size_t const offset
);
196 static int make_nop_x86(void *map
, size_t const offset
)
201 /* Confirm we have 0xe8 0x0 0x0 0x0 0x0 */
206 op
= map
+ offset
- 1;
211 ulseek(fd_map
, offset
- 1, SEEK_SET
);
212 uwrite(fd_map
, ideal_nop
, 5);
216 static unsigned char ideal_nop4_arm_le
[4] = { 0x00, 0x00, 0xa0, 0xe1 }; /* mov r0, r0 */
217 static unsigned char ideal_nop4_arm_be
[4] = { 0xe1, 0xa0, 0x00, 0x00 }; /* mov r0, r0 */
218 static unsigned char *ideal_nop4_arm
;
220 static unsigned char bl_mcount_arm_le
[4] = { 0xfe, 0xff, 0xff, 0xeb }; /* bl */
221 static unsigned char bl_mcount_arm_be
[4] = { 0xeb, 0xff, 0xff, 0xfe }; /* bl */
222 static unsigned char *bl_mcount_arm
;
224 static unsigned char push_arm_le
[4] = { 0x04, 0xe0, 0x2d, 0xe5 }; /* push {lr} */
225 static unsigned char push_arm_be
[4] = { 0xe5, 0x2d, 0xe0, 0x04 }; /* push {lr} */
226 static unsigned char *push_arm
;
228 static unsigned char ideal_nop2_thumb_le
[2] = { 0x00, 0xbf }; /* nop */
229 static unsigned char ideal_nop2_thumb_be
[2] = { 0xbf, 0x00 }; /* nop */
230 static unsigned char *ideal_nop2_thumb
;
232 static unsigned char push_bl_mcount_thumb_le
[6] = { 0x00, 0xb5, 0xff, 0xf7, 0xfe, 0xff }; /* push {lr}, bl */
233 static unsigned char push_bl_mcount_thumb_be
[6] = { 0xb5, 0x00, 0xf7, 0xff, 0xff, 0xfe }; /* push {lr}, bl */
234 static unsigned char *push_bl_mcount_thumb
;
236 static int make_nop_arm(void *map
, size_t const offset
)
244 if (memcmp(ptr
, bl_mcount_arm
, 4) == 0) {
245 if (memcmp(ptr
- 4, push_arm
, 4) == 0) {
249 ideal_nop
= ideal_nop4_arm
;
251 } else if (memcmp(ptr
- 2, push_bl_mcount_thumb
, 6) == 0) {
255 ideal_nop
= ideal_nop2_thumb
;
260 ulseek(fd_map
, off
, SEEK_SET
);
263 uwrite(fd_map
, ideal_nop
, nop_size
);
269 static unsigned char ideal_nop4_arm64
[4] = {0x1f, 0x20, 0x03, 0xd5};
270 static int make_nop_arm64(void *map
, size_t const offset
)
275 /* bl <_mcount> is 0x94000000 before relocation */
276 if (*ptr
!= 0x94000000)
280 ulseek(fd_map
, offset
, SEEK_SET
);
281 uwrite(fd_map
, ideal_nop
, 4);
286 * Get the whole file as a programming convenience in order to avoid
287 * malloc+lseek+read+free of many pieces. If successful, then mmap
288 * avoids copying unused pieces; else just read the whole file.
289 * Open for both read and write; new info will be appended to the file.
290 * Use MAP_PRIVATE so that a few changes to the in-memory ElfXX_Ehdr
291 * do not propagate to the file until an explicit overwrite at the last.
292 * This preserves most aspects of consistency (all except .st_size)
293 * for simultaneous readers of the file while we are appending to it.
294 * However, multiple writers still are bad. We choose not to use
295 * locking because it is expensive and the use case of kernel build
296 * makes multiple writers unlikely.
298 static void *mmap_file(char const *fname
)
300 fd_map
= open(fname
, O_RDONLY
);
301 if (fd_map
< 0 || fstat(fd_map
, &sb
) < 0) {
305 if (!S_ISREG(sb
.st_mode
)) {
306 fprintf(stderr
, "not a regular file: %s\n", fname
);
309 file_map
= mmap(0, sb
.st_size
, PROT_READ
|PROT_WRITE
, MAP_PRIVATE
,
312 if (file_map
== MAP_FAILED
) {
314 file_map
= umalloc(sb
.st_size
);
315 uread(fd_map
, file_map
, sb
.st_size
);
319 file_end
= file_map
+ sb
.st_size
;
324 static void write_file(const char *fname
)
326 char tmp_file
[strlen(fname
) + 4];
332 sprintf(tmp_file
, "%s.rc", fname
);
335 * After reading the entire file into memory, delete it
336 * and write it back, to prevent weird side effects of modifying
337 * an object file in place.
339 fd_map
= open(tmp_file
, O_WRONLY
| O_TRUNC
| O_CREAT
, sb
.st_mode
);
344 n
= write(fd_map
, file_map
, sb
.st_size
);
345 if (n
!= sb
.st_size
) {
349 if (file_append_size
) {
350 n
= write(fd_map
, file_append
, file_append_size
);
351 if (n
!= file_append_size
) {
357 if (rename(tmp_file
, fname
) < 0) {
363 /* w8rev, w8nat, ...: Handle endianness. */
365 static uint64_t w8rev(uint64_t const x
)
367 return ((0xff & (x
>> (0 * 8))) << (7 * 8))
368 | ((0xff & (x
>> (1 * 8))) << (6 * 8))
369 | ((0xff & (x
>> (2 * 8))) << (5 * 8))
370 | ((0xff & (x
>> (3 * 8))) << (4 * 8))
371 | ((0xff & (x
>> (4 * 8))) << (3 * 8))
372 | ((0xff & (x
>> (5 * 8))) << (2 * 8))
373 | ((0xff & (x
>> (6 * 8))) << (1 * 8))
374 | ((0xff & (x
>> (7 * 8))) << (0 * 8));
377 static uint32_t w4rev(uint32_t const x
)
379 return ((0xff & (x
>> (0 * 8))) << (3 * 8))
380 | ((0xff & (x
>> (1 * 8))) << (2 * 8))
381 | ((0xff & (x
>> (2 * 8))) << (1 * 8))
382 | ((0xff & (x
>> (3 * 8))) << (0 * 8));
385 static uint32_t w2rev(uint16_t const x
)
387 return ((0xff & (x
>> (0 * 8))) << (1 * 8))
388 | ((0xff & (x
>> (1 * 8))) << (0 * 8));
391 static uint64_t w8nat(uint64_t const x
)
396 static uint32_t w4nat(uint32_t const x
)
401 static uint32_t w2nat(uint16_t const x
)
406 static uint64_t (*w8
)(uint64_t);
407 static uint32_t (*w
)(uint32_t);
408 static uint32_t (*w2
)(uint16_t);
410 /* Names of the sections that could contain calls to mcount. */
412 is_mcounted_section_name(char const *const txtname
)
414 return strcmp(".text", txtname
) == 0 ||
415 strcmp(".ref.text", txtname
) == 0 ||
416 strcmp(".sched.text", txtname
) == 0 ||
417 strcmp(".spinlock.text", txtname
) == 0 ||
418 strcmp(".irqentry.text", txtname
) == 0 ||
419 strcmp(".softirqentry.text", txtname
) == 0 ||
420 strcmp(".kprobes.text", txtname
) == 0 ||
421 strcmp(".cpuidle.text", txtname
) == 0 ||
422 strcmp(".text.unlikely", txtname
) == 0;
425 /* 32 bit and 64 bit are very similar */
426 #include "recordmcount.h"
427 #define RECORD_MCOUNT_64
428 #include "recordmcount.h"
430 /* 64-bit EM_MIPS has weird ELF64_Rela.r_info.
431 * http://techpubs.sgi.com/library/manuals/4000/007-4658-001/pdf/007-4658-001.pdf
432 * We interpret Table 29 Relocation Operation (Elf64_Rel, Elf64_Rela) [p.40]
433 * to imply the order of the members; the spec does not say so.
434 * typedef unsigned char Elf64_Byte;
435 * fails on MIPS64 because their <elf.h> already has it!
438 typedef uint8_t myElf64_Byte
; /* Type for a 8-bit quantity. */
443 Elf64_Word r_sym
; /* Symbol index. */
444 myElf64_Byte r_ssym
; /* Special symbol. */
445 myElf64_Byte r_type3
; /* Third relocation. */
446 myElf64_Byte r_type2
; /* Second relocation. */
447 myElf64_Byte r_type
; /* First relocation. */
451 static uint64_t MIPS64_r_sym(Elf64_Rel
const *rp
)
453 return w(((union mips_r_info
){ .r_info
= rp
->r_info
}).r_mips
.r_sym
);
456 static void MIPS64_r_info(Elf64_Rel
*const rp
, unsigned sym
, unsigned type
)
458 rp
->r_info
= ((union mips_r_info
){
459 .r_mips
= { .r_sym
= w(sym
), .r_type
= type
}
464 do_file(char const *const fname
)
466 Elf32_Ehdr
*const ehdr
= mmap_file(fname
);
467 unsigned int reltype
= 0;
472 switch (ehdr
->e_ident
[EI_DATA
]) {
473 static unsigned int const endian
= 1;
475 fprintf(stderr
, "unrecognized ELF data encoding %d: %s\n",
476 ehdr
->e_ident
[EI_DATA
], fname
);
480 if (*(unsigned char const *)&endian
!= 1) {
481 /* main() is big endian, file.o is little endian. */
486 ideal_nop4_arm
= ideal_nop4_arm_le
;
487 bl_mcount_arm
= bl_mcount_arm_le
;
488 push_arm
= push_arm_le
;
489 ideal_nop2_thumb
= ideal_nop2_thumb_le
;
490 push_bl_mcount_thumb
= push_bl_mcount_thumb_le
;
493 if (*(unsigned char const *)&endian
!= 0) {
494 /* main() is little endian, file.o is big endian. */
499 ideal_nop4_arm
= ideal_nop4_arm_be
;
500 bl_mcount_arm
= bl_mcount_arm_be
;
501 push_arm
= push_arm_be
;
502 ideal_nop2_thumb
= ideal_nop2_thumb_be
;
503 push_bl_mcount_thumb
= push_bl_mcount_thumb_be
;
506 if (memcmp(ELFMAG
, ehdr
->e_ident
, SELFMAG
) != 0
507 || w2(ehdr
->e_type
) != ET_REL
508 || ehdr
->e_ident
[EI_VERSION
] != EV_CURRENT
) {
509 fprintf(stderr
, "unrecognized ET_REL file %s\n", fname
);
514 switch (w2(ehdr
->e_machine
)) {
516 fprintf(stderr
, "unrecognized e_machine %d %s\n",
517 w2(ehdr
->e_machine
), fname
);
522 rel_type_nop
= R_386_NONE
;
523 make_nop
= make_nop_x86
;
524 ideal_nop
= ideal_nop5_x86_32
;
525 mcount_adjust_32
= -1;
527 case EM_ARM
: reltype
= R_ARM_ABS32
;
528 altmcount
= "__gnu_mcount_nc";
529 make_nop
= make_nop_arm
;
530 rel_type_nop
= R_ARM_NONE
;
533 reltype
= R_AARCH64_ABS64
;
534 make_nop
= make_nop_arm64
;
535 rel_type_nop
= R_AARCH64_NONE
;
536 ideal_nop
= ideal_nop4_arm64
;
539 case EM_IA_64
: reltype
= R_IA64_IMM64
; gpfx
= '_'; break;
540 case EM_METAG
: reltype
= R_METAG_ADDR32
;
541 altmcount
= "_mcount_wrapper";
542 rel_type_nop
= R_METAG_NONE
;
543 /* We happen to have the same requirement as MIPS */
544 is_fake_mcount32
= MIPS32_is_fake_mcount
;
546 case EM_MIPS
: /* reltype: e_class */ gpfx
= '_'; break;
547 case EM_PPC
: reltype
= R_PPC_ADDR32
; gpfx
= '_'; break;
548 case EM_PPC64
: reltype
= R_PPC64_ADDR64
; gpfx
= '_'; break;
549 case EM_S390
: /* reltype: e_class */ gpfx
= '_'; break;
550 case EM_SH
: reltype
= R_SH_DIR32
; break;
551 case EM_SPARCV9
: reltype
= R_SPARC_64
; gpfx
= '_'; break;
553 make_nop
= make_nop_x86
;
554 ideal_nop
= ideal_nop5_x86_64
;
555 reltype
= R_X86_64_64
;
556 rel_type_nop
= R_X86_64_NONE
;
557 mcount_adjust_64
= -1;
561 switch (ehdr
->e_ident
[EI_CLASS
]) {
563 fprintf(stderr
, "unrecognized ELF class %d %s\n",
564 ehdr
->e_ident
[EI_CLASS
], fname
);
568 if (w2(ehdr
->e_ehsize
) != sizeof(Elf32_Ehdr
)
569 || w2(ehdr
->e_shentsize
) != sizeof(Elf32_Shdr
)) {
571 "unrecognized ET_REL file: %s\n", fname
);
574 if (w2(ehdr
->e_machine
) == EM_MIPS
) {
576 is_fake_mcount32
= MIPS32_is_fake_mcount
;
578 do32(ehdr
, fname
, reltype
);
581 Elf64_Ehdr
*const ghdr
= (Elf64_Ehdr
*)ehdr
;
582 if (w2(ghdr
->e_ehsize
) != sizeof(Elf64_Ehdr
)
583 || w2(ghdr
->e_shentsize
) != sizeof(Elf64_Shdr
)) {
585 "unrecognized ET_REL file: %s\n", fname
);
588 if (w2(ghdr
->e_machine
) == EM_S390
) {
590 mcount_adjust_64
= -14;
592 if (w2(ghdr
->e_machine
) == EM_MIPS
) {
594 Elf64_r_sym
= MIPS64_r_sym
;
595 Elf64_r_info
= MIPS64_r_info
;
596 is_fake_mcount64
= MIPS64_is_fake_mcount
;
598 do64(ghdr
, fname
, reltype
);
608 main(int argc
, char *argv
[])
610 const char ftrace
[] = "/ftrace.o";
611 int ftrace_size
= sizeof(ftrace
) - 1;
612 int n_error
= 0; /* gcc-4.3.0 false positive complaint */
616 while ((c
= getopt(argc
, argv
, "w")) >= 0) {
619 warn_on_notrace_sect
= 1;
622 fprintf(stderr
, "usage: recordmcount [-w] file.o...\n");
627 if ((argc
- optind
) < 1) {
628 fprintf(stderr
, "usage: recordmcount [-w] file.o...\n");
632 /* Process each file in turn, allowing deep failure. */
633 for (i
= optind
; i
< argc
; i
++) {
634 char *file
= argv
[i
];
635 int const sjval
= setjmp(jmpenv
);
639 * The file kernel/trace/ftrace.o references the mcount
640 * function but does not call it. Since ftrace.o should
641 * not be traced anyway, we just skip it.
644 if (len
>= ftrace_size
&&
645 strcmp(file
+ (len
- ftrace_size
), ftrace
) == 0)
650 fprintf(stderr
, "internal error: %s\n", file
);
653 case SJ_SETJMP
: /* normal sequence */
654 /* Avoid problems if early cleanup() */
662 case SJ_FAIL
: /* error in do_file or below */
663 fprintf(stderr
, "%s: failed\n", file
);
666 case SJ_SUCCEED
: /* premature success */