1 /* Read and display shared object profiling data.
2 Copyright (C) 1997-2008, 2009, 2010, 2011 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
38 #include <sys/gmon_out.h>
40 #include <sys/param.h>
43 /* Get libc version number. */
44 #include "../version.h"
46 #define PACKAGE _libc_intl_domainname
50 #if BYTE_ORDER == BIG_ENDIAN
51 # define byteorder ELFDATA2MSB
52 # define byteorder_name "big-endian"
53 #elif BYTE_ORDER == LITTLE_ENDIAN
54 # define byteorder ELFDATA2LSB
55 # define byteorder_name "little-endian"
57 # error "Unknown BYTE_ORDER " BYTE_ORDER
58 # define byteorder ELFDATANONE
62 # define PATH_MAX 1024
66 extern int __profile_frequency (void);
68 /* Name and version of program. */
69 static void print_version (FILE *stream
, struct argp_state
*state
);
70 void (*argp_program_version_hook
) (FILE *, struct argp_state
*) = print_version
;
74 /* Definitions of arguments for argp functions. */
75 static const struct argp_option options
[] =
77 { NULL
, 0, NULL
, 0, N_("Output selection:") },
78 { "call-pairs", 'c', NULL
, 0,
79 N_("print list of count paths and their number of use") },
80 { "flat-profile", 'p', NULL
, 0,
81 N_("generate flat profile with counts and ticks") },
82 { "graph", 'q', NULL
, 0, N_("generate call graph") },
84 { "test", OPT_TEST
, NULL
, OPTION_HIDDEN
, NULL
},
85 { NULL
, 0, NULL
, 0, NULL
}
88 /* Short description of program. */
89 static const char doc
[] = N_("Read and display shared object profiling data.");
90 //For bug reporting instructions, please see:\n
91 //<http://www.gnu.org/software/libc/bugs.html>.\n");
93 /* Strings for arguments in help texts. */
94 static const char args_doc
[] = N_("SHOBJ [PROFDATA]");
96 /* Prototype for option handler. */
97 static error_t
parse_opt (int key
, char *arg
, struct argp_state
*state
);
99 /* Function to print some extra text in the help message. */
100 static char *more_help (int key
, const char *text
, void *input
);
102 /* Data structure to communicate with argp functions. */
103 static struct argp argp
=
105 options
, parse_opt
, args_doc
, doc
, NULL
, more_help
109 /* Operation modes. */
114 CALL_GRAPH_MODE
= 1 << 1,
117 DEFAULT_MODE
= FLAT_MODE
| CALL_GRAPH_MODE
120 /* Nozero for testing. */
123 /* Strcuture describing calls. */
124 struct here_fromstruct
126 struct here_cg_arc_record
volatile *here
;
130 /* We define a special type to address the elements of the arc table.
131 This is basically the `gmon_cg_arc_record' format but it includes
132 the room for the tag and it uses real types. */
133 struct here_cg_arc_record
138 } __attribute__ ((packed
));
147 struct arc_list
*next
;
150 static struct obstack ob_list
;
164 struct arc_list
*froms
;
165 struct arc_list
*tos
;
171 const char *name
; /* User-provided name. */
173 struct link_map
*map
;
174 const char *dynstrtab
; /* Dynamic string table of shared object. */
175 const char *soname
; /* Soname of shared object. */
179 unsigned long int kcountsize
;
180 size_t expected_size
; /* Expected size of profiling file. */
184 unsigned int hashfraction
;
188 size_t symbol_mapsize
;
189 const ElfW(Sym
) *symtab
;
193 struct obstack ob_str
;
194 struct obstack ob_sym
;
198 struct real_gmon_hist_hdr
215 struct real_gmon_hist_hdr
*hist_hdr
;
217 uint32_t narcs
; /* Number of arcs in toset. */
218 struct here_cg_arc_record
*data
;
220 struct here_fromstruct
*froms
;
223 /* Search tree for symbols. */
224 static void *symroot
;
225 static struct known_symbol
**sortsym
;
226 static size_t symidx
;
227 static uintmax_t total_ticks
;
229 /* Prototypes for local functions. */
230 static struct shobj
*load_shobj (const char *name
);
231 static void unload_shobj (struct shobj
*shobj
);
232 static struct profdata
*load_profdata (const char *name
, struct shobj
*shobj
);
233 static void unload_profdata (struct profdata
*profdata
);
234 static void count_total_ticks (struct shobj
*shobj
, struct profdata
*profdata
);
235 static void count_calls (struct shobj
*shobj
, struct profdata
*profdata
);
236 static void read_symbols (struct shobj
*shobj
);
237 static void add_arcs (struct profdata
*profdata
);
238 static void generate_flat_profile (struct profdata
*profdata
);
239 static void generate_call_graph (struct profdata
*profdata
);
240 static void generate_call_pair_list (struct profdata
*profdata
);
244 main (int argc
, char *argv
[])
247 const char *profdata
;
248 struct shobj
*shobj_handle
;
249 struct profdata
*profdata_handle
;
252 setlocale (LC_ALL
, "");
254 /* Initialize the message catalog. */
255 textdomain (_libc_intl_domainname
);
257 /* Parse and process arguments. */
258 argp_parse (&argp
, argc
, argv
, 0, &remaining
, NULL
);
260 if (argc
- remaining
== 0 || argc
- remaining
> 2)
262 /* We need exactly two non-option parameter. */
263 argp_help (&argp
, stdout
, ARGP_HELP_SEE
| ARGP_HELP_EXIT_ERR
,
264 program_invocation_short_name
);
268 /* Get parameters. */
269 shobj
= argv
[remaining
];
270 if (argc
- remaining
== 2)
271 profdata
= argv
[remaining
+ 1];
273 /* No filename for the profiling data given. We will determine it
274 from the soname of the shobj, later. */
277 /* First see whether we can load the shared object. */
278 shobj_handle
= load_shobj (shobj
);
279 if (shobj_handle
== NULL
)
282 /* We can now determine the filename for the profiling data, if
284 if (profdata
== NULL
)
290 soname
= shobj_handle
->soname
?: basename (shobj
);
291 soname_len
= strlen (soname
);
292 newp
= (char *) alloca (soname_len
+ sizeof ".profile");
293 stpcpy (mempcpy (newp
, soname
, soname_len
), ".profile");
297 /* Now see whether the profiling data file matches the given object. */
298 profdata_handle
= load_profdata (profdata
, shobj_handle
);
299 if (profdata_handle
== NULL
)
301 unload_shobj (shobj_handle
);
306 read_symbols (shobj_handle
);
308 /* Count the ticks. */
309 count_total_ticks (shobj_handle
, profdata_handle
);
311 /* Count the calls. */
312 count_calls (shobj_handle
, profdata_handle
);
314 /* Add the arc information. */
315 add_arcs (profdata_handle
);
317 /* If no mode is specified fall back to the default mode. */
322 if (mode
& FLAT_MODE
)
323 generate_flat_profile (profdata_handle
);
325 if (mode
& CALL_GRAPH_MODE
)
326 generate_call_graph (profdata_handle
);
328 if (mode
& CALL_PAIRS
)
329 generate_call_pair_list (profdata_handle
);
331 /* Free the resources. */
332 unload_shobj (shobj_handle
);
333 unload_profdata (profdata_handle
);
339 /* Handle program arguments. */
341 parse_opt (int key
, char *arg
, struct argp_state
*state
)
352 mode
|= CALL_GRAPH_MODE
;
358 return ARGP_ERR_UNKNOWN
;
365 more_help (int key
, const char *text
, void *input
)
369 case ARGP_KEY_HELP_EXTRA
:
370 /* We print some extra information. */
371 return strdup (gettext ("\
372 For bug reporting instructions, please see:\n\
373 <http://www.gnu.org/software/libc/bugs.html>.\n"));
377 return (char *) text
;
381 /* Print the version information. */
383 print_version (FILE *stream
, struct argp_state
*state
)
385 fprintf (stream
, "sprof (GNU %s) %s\n", PACKAGE
, VERSION
);
386 fprintf (stream
, gettext ("\
387 Copyright (C) %s Free Software Foundation, Inc.\n\
388 This is free software; see the source for copying conditions. There is NO\n\
389 warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n\
392 fprintf (stream
, gettext ("Written by %s.\n"), "Ulrich Drepper");
396 /* Note that we must not use `dlopen' etc. The shobj object must not
397 be loaded for use. */
398 static struct shobj
*
399 load_shobj (const char *name
)
401 struct link_map
*map
= NULL
;
402 struct shobj
*result
;
403 ElfW(Addr
) mapstart
= ~((ElfW(Addr
)) 0);
404 ElfW(Addr
) mapend
= 0;
405 const ElfW(Phdr
) *ph
;
410 size_t pagesize
= getpagesize ();
412 /* Since we use dlopen() we must be prepared to work around the sometimes
413 strange lookup rules for the shared objects. If we have a file foo.so
414 in the current directory and the user specfies foo.so on the command
415 line (without specifying a directory) we should load the file in the
416 current directory even if a normal dlopen() call would read the other
417 file. We do this by adding a directory portion to the name. */
418 if (strchr (name
, '/') == NULL
)
420 char *load_name
= (char *) alloca (strlen (name
) + 3);
421 stpcpy (stpcpy (load_name
, "./"), name
);
423 map
= (struct link_map
*) dlopen (load_name
, RTLD_LAZY
| __RTLD_SPROF
);
427 map
= (struct link_map
*) dlopen (name
, RTLD_LAZY
| __RTLD_SPROF
);
430 error (0, errno
, _("failed to load shared object `%s'"), name
);
435 /* Prepare the result. */
436 result
= (struct shobj
*) calloc (1, sizeof (struct shobj
));
439 error (0, errno
, _("cannot create internal descriptors"));
446 /* Compute the size of the sections which contain program code.
447 This must match the code in dl-profile.c (_dl_start_profile). */
448 for (ph
= map
->l_phdr
; ph
< &map
->l_phdr
[map
->l_phnum
]; ++ph
)
449 if (ph
->p_type
== PT_LOAD
&& (ph
->p_flags
& PF_X
))
451 ElfW(Addr
) start
= (ph
->p_vaddr
& ~(pagesize
- 1));
452 ElfW(Addr
) end
= ((ph
->p_vaddr
+ ph
->p_memsz
+ pagesize
- 1)
455 if (start
< mapstart
)
461 result
->lowpc
= ROUNDDOWN ((uintptr_t) (mapstart
+ map
->l_addr
),
462 HISTFRACTION
* sizeof (HISTCOUNTER
));
463 result
->highpc
= ROUNDUP ((uintptr_t) (mapend
+ map
->l_addr
),
464 HISTFRACTION
* sizeof (HISTCOUNTER
));
466 printf ("load addr: %0#*" PRIxPTR
"\n"
467 "lower bound PC: %0#*" PRIxPTR
"\n"
468 "upper bound PC: %0#*" PRIxPTR
"\n",
469 __ELF_NATIVE_CLASS
== 32 ? 10 : 18, map
->l_addr
,
470 __ELF_NATIVE_CLASS
== 32 ? 10 : 18, result
->lowpc
,
471 __ELF_NATIVE_CLASS
== 32 ? 10 : 18, result
->highpc
);
473 textsize
= result
->highpc
- result
->lowpc
;
474 result
->kcountsize
= textsize
/ HISTFRACTION
;
475 result
->hashfraction
= HASHFRACTION
;
477 printf ("hashfraction = %d\ndivider = %Zu\n",
478 result
->hashfraction
,
479 result
->hashfraction
* sizeof (struct here_fromstruct
));
480 result
->tossize
= textsize
/ HASHFRACTION
;
481 result
->fromlimit
= textsize
* ARCDENSITY
/ 100;
482 if (result
->fromlimit
< MINARCS
)
483 result
->fromlimit
= MINARCS
;
484 if (result
->fromlimit
> MAXARCS
)
485 result
->fromlimit
= MAXARCS
;
486 result
->fromssize
= result
->fromlimit
* sizeof (struct here_fromstruct
);
488 result
->expected_size
= (sizeof (struct gmon_hdr
)
489 + 4 + sizeof (struct gmon_hist_hdr
)
493 * sizeof (struct here_cg_arc_record
)));
496 printf ("expected size: %Zd\n", result
->expected_size
);
498 #define SCALE_1_TO_1 0x10000L
500 if (result
->kcountsize
< result
->highpc
- result
->lowpc
)
502 size_t range
= result
->highpc
- result
->lowpc
;
503 size_t quot
= range
/ result
->kcountsize
;
505 if (quot
>= SCALE_1_TO_1
)
507 else if (quot
>= SCALE_1_TO_1
/ 256)
508 result
->s_scale
= SCALE_1_TO_1
/ quot
;
509 else if (range
> ULONG_MAX
/ 256)
510 result
->s_scale
= ((SCALE_1_TO_1
* 256)
511 / (range
/ (result
->kcountsize
/ 256)));
513 result
->s_scale
= ((SCALE_1_TO_1
* 256)
514 / ((range
* 256) / result
->kcountsize
));
517 result
->s_scale
= SCALE_1_TO_1
;
520 printf ("s_scale: %d\n", result
->s_scale
);
522 /* Determine the dynamic string table. */
523 if (map
->l_info
[DT_STRTAB
] == NULL
)
524 result
->dynstrtab
= NULL
;
526 result
->dynstrtab
= (const char *) D_PTR (map
, l_info
[DT_STRTAB
]);
528 printf ("string table: %p\n", result
->dynstrtab
);
530 /* Determine the soname. */
531 if (map
->l_info
[DT_SONAME
] == NULL
)
532 result
->soname
= NULL
;
534 result
->soname
= result
->dynstrtab
+ map
->l_info
[DT_SONAME
]->d_un
.d_val
;
535 if (do_test
&& result
->soname
!= NULL
)
536 printf ("soname: %s\n", result
->soname
);
538 /* Now we have to load the symbol table.
540 First load the section header table. */
541 ehdr
= (ElfW(Ehdr
) *) map
->l_map_start
;
543 /* Make sure we are on the right party. */
544 if (ehdr
->e_shentsize
!= sizeof (ElfW(Shdr
)))
547 /* And we need the shared object file descriptor again. */
548 fd
= open (map
->l_name
, O_RDONLY
);
550 /* Dooh, this really shouldn't happen. We know the file is available. */
551 error (EXIT_FAILURE
, errno
, _("Reopening shared object `%s' failed"),
554 /* Map the section header. */
555 size_t size
= ehdr
->e_shnum
* sizeof (ElfW(Shdr
));
556 shdr
= (ElfW(Shdr
) *) alloca (size
);
557 if (pread (fd
, shdr
, size
, ehdr
->e_shoff
) != size
)
558 error (EXIT_FAILURE
, errno
, _("reading of section headers failed"));
560 /* Get the section header string table. */
561 char *shstrtab
= (char *) alloca (shdr
[ehdr
->e_shstrndx
].sh_size
);
562 if (pread (fd
, shstrtab
, shdr
[ehdr
->e_shstrndx
].sh_size
,
563 shdr
[ehdr
->e_shstrndx
].sh_offset
)
564 != shdr
[ehdr
->e_shstrndx
].sh_size
)
565 error (EXIT_FAILURE
, errno
,
566 _("reading of section header string table failed"));
568 /* Search for the ".symtab" section. */
569 ElfW(Shdr
) *symtab_entry
= NULL
;
570 ElfW(Shdr
) *debuglink_entry
= NULL
;
571 for (int idx
= 0; idx
< ehdr
->e_shnum
; ++idx
)
572 if (shdr
[idx
].sh_type
== SHT_SYMTAB
573 && strcmp (shstrtab
+ shdr
[idx
].sh_name
, ".symtab") == 0)
575 symtab_entry
= &shdr
[idx
];
578 else if (shdr
[idx
].sh_type
== SHT_PROGBITS
579 && strcmp (shstrtab
+ shdr
[idx
].sh_name
, ".gnu_debuglink") == 0)
580 debuglink_entry
= &shdr
[idx
];
582 /* Get the file name of the debuginfo file if necessary. */
584 if (symtab_entry
== NULL
&& debuglink_entry
!= NULL
)
586 size_t size
= debuglink_entry
->sh_size
;
587 char *debuginfo_fname
= (char *) alloca (size
+ 1);
588 debuginfo_fname
[size
] = '\0';
589 if (pread (fd
, debuginfo_fname
, size
, debuglink_entry
->sh_offset
)
592 fprintf (stderr
, _("*** Cannot read debuginfo file name: %m\n"));
596 static const char procpath
[] = "/proc/self/fd/%d";
597 char origprocname
[sizeof (procpath
) + sizeof (int) * 3];
598 snprintf (origprocname
, sizeof (origprocname
), procpath
, fd
);
599 char *origlink
= (char *) alloca (PATH_MAX
+ 1);
600 origlink
[PATH_MAX
] = '\0';
601 if (readlink (origprocname
, origlink
, PATH_MAX
) == -1)
604 /* Try to find the actual file. There are three places:
605 1. the same directory the DSO is in
606 2. in a subdir named .debug of the directory the DSO is in
607 3. in /usr/lib/debug/PATH-OF-DSO
609 char *realname
= canonicalize_file_name (origlink
);
611 if (realname
== NULL
|| (cp
= strrchr (realname
, '/')) == NULL
)
612 error (EXIT_FAILURE
, errno
, _("cannot determine file name"));
614 /* Leave the last slash in place. */
617 /* First add the debuginfo file name only. */
618 static const char usrlibdebug
[]= "/usr/lib/debug/";
619 char *workbuf
= (char *) alloca (sizeof (usrlibdebug
)
621 + strlen (debuginfo_fname
));
622 strcpy (stpcpy (workbuf
, realname
), debuginfo_fname
);
624 int fd2
= open (workbuf
, O_RDONLY
);
627 strcpy (stpcpy (stpcpy (workbuf
, realname
), ".debug/"),
629 fd2
= open (workbuf
, O_RDONLY
);
632 strcpy (stpcpy (stpcpy (workbuf
, usrlibdebug
), realname
),
634 fd2
= open (workbuf
, O_RDONLY
);
642 /* Read the ELF header. */
643 if (pread (fd2
, &ehdr2
, sizeof (ehdr2
), 0) != sizeof (ehdr2
))
644 error (EXIT_FAILURE
, errno
,
645 _("reading of ELF header failed"));
647 /* Map the section header. */
648 size_t size
= ehdr2
.e_shnum
* sizeof (ElfW(Shdr
));
649 ElfW(Shdr
) *shdr2
= (ElfW(Shdr
) *) alloca (size
);
650 if (pread (fd2
, shdr2
, size
, ehdr2
.e_shoff
) != size
)
651 error (EXIT_FAILURE
, errno
,
652 _("reading of section headers failed"));
654 /* Get the section header string table. */
655 shstrtab
= (char *) alloca (shdr2
[ehdr2
.e_shstrndx
].sh_size
);
656 if (pread (fd2
, shstrtab
, shdr2
[ehdr2
.e_shstrndx
].sh_size
,
657 shdr2
[ehdr2
.e_shstrndx
].sh_offset
)
658 != shdr2
[ehdr2
.e_shstrndx
].sh_size
)
659 error (EXIT_FAILURE
, errno
,
660 _("reading of section header string table failed"));
662 /* Search for the ".symtab" section. */
663 for (int idx
= 0; idx
< ehdr2
.e_shnum
; ++idx
)
664 if (shdr2
[idx
].sh_type
== SHT_SYMTAB
665 && strcmp (shstrtab
+ shdr2
[idx
].sh_name
, ".symtab") == 0)
667 symtab_entry
= &shdr2
[idx
];
679 if (symtab_entry
== NULL
)
681 fprintf (stderr
, _("\
682 *** The file `%s' is stripped: no detailed analysis possible\n"),
684 result
->symtab
= NULL
;
685 result
->strtab
= NULL
;
689 ElfW(Off
) min_offset
, max_offset
;
690 ElfW(Shdr
) *strtab_entry
;
692 strtab_entry
= &shdr
[symtab_entry
->sh_link
];
694 /* Find the minimum and maximum offsets that include both the symbol
695 table and the string table. */
696 if (symtab_entry
->sh_offset
< strtab_entry
->sh_offset
)
698 min_offset
= symtab_entry
->sh_offset
& ~(pagesize
- 1);
699 max_offset
= strtab_entry
->sh_offset
+ strtab_entry
->sh_size
;
703 min_offset
= strtab_entry
->sh_offset
& ~(pagesize
- 1);
704 max_offset
= symtab_entry
->sh_offset
+ symtab_entry
->sh_size
;
707 result
->symbol_map
= mmap (NULL
, max_offset
- min_offset
,
708 PROT_READ
, MAP_SHARED
|MAP_FILE
, symfd
,
710 if (result
->symbol_map
== MAP_FAILED
)
711 error (EXIT_FAILURE
, errno
, _("failed to load symbol data"));
714 = (const ElfW(Sym
) *) ((const char *) result
->symbol_map
715 + (symtab_entry
->sh_offset
- min_offset
));
716 result
->symtab_size
= symtab_entry
->sh_size
;
717 result
->strtab
= ((const char *) result
->symbol_map
718 + (strtab_entry
->sh_offset
- min_offset
));
719 result
->symbol_mapsize
= max_offset
- min_offset
;
722 /* Free the descriptor for the shared object. */
732 unload_shobj (struct shobj
*shobj
)
734 munmap (shobj
->symbol_map
, shobj
->symbol_mapsize
);
735 dlclose (shobj
->map
);
739 static struct profdata
*
740 load_profdata (const char *name
, struct shobj
*shobj
)
742 struct profdata
*result
;
748 struct here_cg_arc_record
*data
;
749 struct here_fromstruct
*froms
;
754 fd
= open (name
, O_RDONLY
);
759 if (errno
!= ENOENT
|| strchr (name
, '/') != NULL
)
760 /* The file exists but we are not allowed to read it or the
761 file does not exist and the name includes a path
765 /* A file with the given name does not exist in the current
766 directory, try it in the default location where the profiling
767 files are created. */
768 ext_name
= (char *) alloca (strlen (name
) + sizeof "/var/tmp/");
769 stpcpy (stpcpy (ext_name
, "/var/tmp/"), name
);
772 fd
= open (ext_name
, O_RDONLY
);
775 /* Even this file does not exist. */
776 error (0, errno
, _("cannot load profiling data"));
781 /* We have found the file, now make sure it is the right one for the
783 if (fstat (fd
, &st
) < 0)
785 error (0, errno
, _("while stat'ing profiling data file"));
790 if ((size_t) st
.st_size
!= shobj
->expected_size
)
793 _("profiling data file `%s' does not match shared object `%s'"),
799 /* The data file is most probably the right one for our shared
800 object. Map it now. */
801 addr
= mmap (NULL
, st
.st_size
, PROT_READ
, MAP_SHARED
|MAP_FILE
, fd
, 0);
802 if (addr
== MAP_FAILED
)
804 error (0, errno
, _("failed to mmap the profiling data file"));
809 /* We don't need the file desriptor anymore. */
812 error (0, errno
, _("error while closing the profiling data file"));
813 munmap (addr
, st
.st_size
);
817 /* Prepare the result. */
818 result
= (struct profdata
*) calloc (1, sizeof (struct profdata
));
821 error (0, errno
, _("cannot create internal descriptor"));
822 munmap (addr
, st
.st_size
);
826 /* Store the address and size so that we can later free the resources. */
828 result
->size
= st
.st_size
;
830 /* Pointer to data after the header. */
831 result
->hist
= (char *) ((struct gmon_hdr
*) addr
+ 1);
832 result
->hist_hdr
= (struct real_gmon_hist_hdr
*) ((char *) result
->hist
833 + sizeof (uint32_t));
834 result
->kcount
= (uint16_t *) ((char *) result
->hist
+ sizeof (uint32_t)
835 + sizeof (struct real_gmon_hist_hdr
));
837 /* Compute pointer to array of the arc information. */
838 narcsp
= (uint32_t *) ((char *) result
->kcount
+ shobj
->kcountsize
839 + sizeof (uint32_t));
840 result
->narcs
= *narcsp
;
841 result
->data
= (struct here_cg_arc_record
*) ((char *) narcsp
842 + sizeof (uint32_t));
844 /* Create the gmon_hdr we expect or write. */
851 if (sizeof (gmon_hdr
) != sizeof (struct gmon_hdr
)
852 || (offsetof (struct real_gmon_hdr
, cookie
)
853 != offsetof (struct gmon_hdr
, cookie
))
854 || (offsetof (struct real_gmon_hdr
, version
)
855 != offsetof (struct gmon_hdr
, version
)))
858 memcpy (&gmon_hdr
.cookie
[0], GMON_MAGIC
, sizeof (gmon_hdr
.cookie
));
859 gmon_hdr
.version
= GMON_SHOBJ_VERSION
;
860 memset (gmon_hdr
.spare
, '\0', sizeof (gmon_hdr
.spare
));
862 /* Create the hist_hdr we expect or write. */
863 struct real_gmon_hist_hdr hist_hdr
;
864 if (sizeof (hist_hdr
) != sizeof (struct gmon_hist_hdr
)
865 || (offsetof (struct real_gmon_hist_hdr
, low_pc
)
866 != offsetof (struct gmon_hist_hdr
, low_pc
))
867 || (offsetof (struct real_gmon_hist_hdr
, high_pc
)
868 != offsetof (struct gmon_hist_hdr
, high_pc
))
869 || (offsetof (struct real_gmon_hist_hdr
, hist_size
)
870 != offsetof (struct gmon_hist_hdr
, hist_size
))
871 || (offsetof (struct real_gmon_hist_hdr
, prof_rate
)
872 != offsetof (struct gmon_hist_hdr
, prof_rate
))
873 || (offsetof (struct real_gmon_hist_hdr
, dimen
)
874 != offsetof (struct gmon_hist_hdr
, dimen
))
875 || (offsetof (struct real_gmon_hist_hdr
, dimen_abbrev
)
876 != offsetof (struct gmon_hist_hdr
, dimen_abbrev
)))
879 hist_hdr
.low_pc
= (char *) shobj
->lowpc
- shobj
->map
->l_addr
;
880 hist_hdr
.high_pc
= (char *) shobj
->highpc
- shobj
->map
->l_addr
;
882 printf ("low_pc = %p\nhigh_pc = %p\n", hist_hdr
.low_pc
, hist_hdr
.high_pc
);
883 hist_hdr
.hist_size
= shobj
->kcountsize
/ sizeof (HISTCOUNTER
);
884 hist_hdr
.prof_rate
= __profile_frequency ();
885 strncpy (hist_hdr
.dimen
, "seconds", sizeof (hist_hdr
.dimen
));
886 hist_hdr
.dimen_abbrev
= 's';
888 /* Test whether the header of the profiling data is ok. */
889 if (memcmp (addr
, &gmon_hdr
, sizeof (struct gmon_hdr
)) != 0
890 || *(uint32_t *) result
->hist
!= GMON_TAG_TIME_HIST
891 || memcmp (result
->hist_hdr
, &hist_hdr
,
892 sizeof (struct gmon_hist_hdr
)) != 0
893 || narcsp
[-1] != GMON_TAG_CG_ARC
)
895 error (0, 0, _("`%s' is no correct profile data file for `%s'"),
899 if (memcmp (addr
, &gmon_hdr
, sizeof (struct gmon_hdr
)) != 0)
900 puts ("gmon_hdr differs");
901 if (*(uint32_t *) result
->hist
!= GMON_TAG_TIME_HIST
)
902 puts ("result->hist differs");
903 if (memcmp (result
->hist_hdr
, &hist_hdr
,
904 sizeof (struct gmon_hist_hdr
)) != 0)
905 puts ("hist_hdr differs");
906 if (narcsp
[-1] != GMON_TAG_CG_ARC
)
907 puts ("narcsp[-1] differs");
910 munmap (addr
, st
.st_size
);
914 /* We are pretty sure now that this is a correct input file. Set up
915 the remaining information in the result structure and return. */
916 result
->tos
= (uint16_t *) calloc (shobj
->tossize
+ shobj
->fromssize
, 1);
917 if (result
->tos
== NULL
)
919 error (0, errno
, _("cannot create internal descriptor"));
920 munmap (addr
, st
.st_size
);
925 result
->froms
= (struct here_fromstruct
*) ((char *) result
->tos
929 /* Now we have to process all the arc count entries. */
930 fromlimit
= shobj
->fromlimit
;
932 froms
= result
->froms
;
934 for (idx
= 0; idx
< MIN (*narcsp
, fromlimit
); ++idx
)
938 to_index
= (data
[idx
].self_pc
/ (shobj
->hashfraction
* sizeof (*tos
)));
939 newfromidx
= fromidx
++;
940 froms
[newfromidx
].here
= &data
[idx
];
941 froms
[newfromidx
].link
= tos
[to_index
];
942 tos
[to_index
] = newfromidx
;
950 unload_profdata (struct profdata
*profdata
)
952 free (profdata
->tos
);
953 munmap (profdata
->addr
, profdata
->size
);
959 count_total_ticks (struct shobj
*shobj
, struct profdata
*profdata
)
961 volatile uint16_t *kcount
= profdata
->kcount
;
962 size_t maxkidx
= shobj
->kcountsize
;
963 size_t factor
= 2 * (65536 / shobj
->s_scale
);
967 while (sidx
< symidx
)
969 uintptr_t start
= sortsym
[sidx
]->addr
;
970 uintptr_t end
= start
+ sortsym
[sidx
]->size
;
972 while (kidx
< maxkidx
&& factor
* kidx
< start
)
977 while (kidx
< maxkidx
&& factor
* kidx
< end
)
978 sortsym
[sidx
]->ticks
+= kcount
[kidx
++];
982 total_ticks
+= sortsym
[sidx
++]->ticks
;
988 find_symbol (uintptr_t addr
)
992 while (sidx
< symidx
)
994 uintptr_t start
= sortsym
[sidx
]->addr
;
995 uintptr_t end
= start
+ sortsym
[sidx
]->size
;
997 if (addr
>= start
&& addr
< end
)
1006 return (size_t) -1l;
1011 count_calls (struct shobj
*shobj
, struct profdata
*profdata
)
1013 struct here_cg_arc_record
*data
= profdata
->data
;
1014 uint32_t narcs
= profdata
->narcs
;
1017 for (cnt
= 0; cnt
< narcs
; ++cnt
)
1019 uintptr_t here
= data
[cnt
].self_pc
;
1022 /* Find the symbol for this address. */
1023 symbol_idx
= find_symbol (here
);
1024 if (symbol_idx
!= (size_t) -1l)
1025 sortsym
[symbol_idx
]->calls
+= data
[cnt
].count
;
1031 symorder (const void *o1
, const void *o2
)
1033 const struct known_symbol
*p1
= (const struct known_symbol
*) o1
;
1034 const struct known_symbol
*p2
= (const struct known_symbol
*) o2
;
1036 return p1
->addr
- p2
->addr
;
1041 printsym (const void *node
, VISIT value
, int level
)
1043 if (value
== leaf
|| value
== postorder
)
1044 sortsym
[symidx
++] = *(struct known_symbol
**) node
;
1049 read_symbols (struct shobj
*shobj
)
1053 /* Initialize the obstacks. */
1054 #define obstack_chunk_alloc malloc
1055 #define obstack_chunk_free free
1056 obstack_init (&shobj
->ob_str
);
1057 obstack_init (&shobj
->ob_sym
);
1058 obstack_init (&ob_list
);
1060 /* Process the symbols. */
1061 if (shobj
->symtab
!= NULL
)
1063 const ElfW(Sym
) *sym
= shobj
->symtab
;
1064 const ElfW(Sym
) *sym_end
1065 = (const ElfW(Sym
) *) ((const char *) sym
+ shobj
->symtab_size
);
1066 for (; sym
< sym_end
; sym
++)
1067 if ((ELFW(ST_TYPE
) (sym
->st_info
) == STT_FUNC
1068 || ELFW(ST_TYPE
) (sym
->st_info
) == STT_NOTYPE
)
1069 && sym
->st_size
!= 0)
1071 struct known_symbol
**existp
;
1072 struct known_symbol
*newsym
1073 = (struct known_symbol
*) obstack_alloc (&shobj
->ob_sym
,
1076 error (EXIT_FAILURE
, errno
, _("cannot allocate symbol data"));
1078 newsym
->name
= &shobj
->strtab
[sym
->st_name
];
1079 newsym
->addr
= sym
->st_value
;
1080 newsym
->size
= sym
->st_size
;
1081 newsym
->weak
= ELFW(ST_BIND
) (sym
->st_info
) == STB_WEAK
;
1082 newsym
->hidden
= (ELFW(ST_VISIBILITY
) (sym
->st_other
)
1087 existp
= tfind (newsym
, &symroot
, symorder
);
1091 tsearch (newsym
, &symroot
, symorder
);
1096 /* The function is already defined. See whether we have
1097 a better name here. */
1098 if (((*existp
)->hidden
&& !newsym
->hidden
)
1099 || ((*existp
)->name
[0] == '_' && newsym
->name
[0] != '_')
1100 || ((*existp
)->name
[0] != '_' && newsym
->name
[0] != '_'
1101 && ((*existp
)->weak
&& !newsym
->weak
)))
1104 /* We don't need the allocated memory. */
1105 obstack_free (&shobj
->ob_sym
, newsym
);
1111 /* Blarg, the binary is stripped. We have to rely on the
1112 information contained in the dynamic section of the object. */
1113 const ElfW(Sym
) *symtab
= (ElfW(Sym
) *) D_PTR (shobj
->map
,
1115 const char *strtab
= (const char *) D_PTR (shobj
->map
,
1118 /* We assume that the string table follows the symbol table,
1119 because there is no way in ELF to know the size of the
1120 dynamic symbol table without looking at the section headers. */
1121 while ((void *) symtab
< (void *) strtab
)
1123 if ((ELFW(ST_TYPE
)(symtab
->st_info
) == STT_FUNC
1124 || ELFW(ST_TYPE
)(symtab
->st_info
) == STT_NOTYPE
)
1125 && symtab
->st_size
!= 0)
1127 struct known_symbol
*newsym
;
1128 struct known_symbol
**existp
;
1131 (struct known_symbol
*) obstack_alloc (&shobj
->ob_sym
,
1134 error (EXIT_FAILURE
, errno
, _("cannot allocate symbol data"));
1136 newsym
->name
= &strtab
[symtab
->st_name
];
1137 newsym
->addr
= symtab
->st_value
;
1138 newsym
->size
= symtab
->st_size
;
1139 newsym
->weak
= ELFW(ST_BIND
) (symtab
->st_info
) == STB_WEAK
;
1140 newsym
->hidden
= (ELFW(ST_VISIBILITY
) (symtab
->st_other
)
1143 newsym
->froms
= NULL
;
1146 existp
= tfind (newsym
, &symroot
, symorder
);
1150 tsearch (newsym
, &symroot
, symorder
);
1155 /* The function is already defined. See whether we have
1156 a better name here. */
1157 if (((*existp
)->hidden
&& !newsym
->hidden
)
1158 || ((*existp
)->name
[0] == '_' && newsym
->name
[0] != '_')
1159 || ((*existp
)->name
[0] != '_' && newsym
->name
[0] != '_'
1160 && ((*existp
)->weak
&& !newsym
->weak
)))
1163 /* We don't need the allocated memory. */
1164 obstack_free (&shobj
->ob_sym
, newsym
);
1172 sortsym
= malloc (n
* sizeof (struct known_symbol
*));
1173 if (sortsym
== NULL
)
1176 twalk (symroot
, printsym
);
1181 add_arcs (struct profdata
*profdata
)
1183 uint32_t narcs
= profdata
->narcs
;
1184 struct here_cg_arc_record
*data
= profdata
->data
;
1187 for (cnt
= 0; cnt
< narcs
; ++cnt
)
1189 /* First add the incoming arc. */
1190 size_t sym_idx
= find_symbol (data
[cnt
].self_pc
);
1192 if (sym_idx
!= (size_t) -1l)
1194 struct known_symbol
*sym
= sortsym
[sym_idx
];
1195 struct arc_list
*runp
= sym
->froms
;
1198 && ((data
[cnt
].from_pc
== 0 && runp
->idx
!= (size_t) -1l)
1199 || (data
[cnt
].from_pc
!= 0
1200 && (runp
->idx
== (size_t) -1l
1201 || data
[cnt
].from_pc
< sortsym
[runp
->idx
]->addr
1202 || (data
[cnt
].from_pc
1203 >= (sortsym
[runp
->idx
]->addr
1204 + sortsym
[runp
->idx
]->size
))))))
1209 /* We need a new entry. */
1210 struct arc_list
*newp
= (struct arc_list
*)
1211 obstack_alloc (&ob_list
, sizeof (struct arc_list
));
1213 if (data
[cnt
].from_pc
== 0)
1214 newp
->idx
= (size_t) -1l;
1216 newp
->idx
= find_symbol (data
[cnt
].from_pc
);
1217 newp
->count
= data
[cnt
].count
;
1218 newp
->next
= sym
->froms
;
1222 /* Increment the counter for the found entry. */
1223 runp
->count
+= data
[cnt
].count
;
1226 /* Now add it to the appropriate outgoing list. */
1227 sym_idx
= find_symbol (data
[cnt
].from_pc
);
1228 if (sym_idx
!= (size_t) -1l)
1230 struct known_symbol
*sym
= sortsym
[sym_idx
];
1231 struct arc_list
*runp
= sym
->tos
;
1234 && (runp
->idx
== (size_t) -1l
1235 || data
[cnt
].self_pc
< sortsym
[runp
->idx
]->addr
1236 || data
[cnt
].self_pc
>= (sortsym
[runp
->idx
]->addr
1237 + sortsym
[runp
->idx
]->size
)))
1242 /* We need a new entry. */
1243 struct arc_list
*newp
= (struct arc_list
*)
1244 obstack_alloc (&ob_list
, sizeof (struct arc_list
));
1246 newp
->idx
= find_symbol (data
[cnt
].self_pc
);
1247 newp
->count
= data
[cnt
].count
;
1248 newp
->next
= sym
->tos
;
1252 /* Increment the counter for the found entry. */
1253 runp
->count
+= data
[cnt
].count
;
1260 countorder (const void *p1
, const void *p2
)
1262 struct known_symbol
*s1
= (struct known_symbol
*) p1
;
1263 struct known_symbol
*s2
= (struct known_symbol
*) p2
;
1265 if (s1
->ticks
!= s2
->ticks
)
1266 return (int) (s2
->ticks
- s1
->ticks
);
1268 if (s1
->calls
!= s2
->calls
)
1269 return (int) (s2
->calls
- s1
->calls
);
1271 return strcmp (s1
->name
, s2
->name
);
1275 static double tick_unit
;
1276 static uintmax_t cumu_ticks
;
1279 printflat (const void *node
, VISIT value
, int level
)
1281 if (value
== leaf
|| value
== postorder
)
1283 struct known_symbol
*s
= *(struct known_symbol
**) node
;
1285 cumu_ticks
+= s
->ticks
;
1287 printf ("%6.2f%10.2f%9.2f%9" PRIdMAX
"%9.2f %s\n",
1288 total_ticks
? (100.0 * s
->ticks
) / total_ticks
: 0.0,
1289 tick_unit
* cumu_ticks
,
1290 tick_unit
* s
->ticks
,
1292 s
->calls
? (s
->ticks
* 1000000) * tick_unit
/ s
->calls
: 0,
1293 /* FIXME: don't know about called functions. */
1307 generate_flat_profile (struct profdata
*profdata
)
1312 tick_unit
= 1.0 / profdata
->hist_hdr
->prof_rate
;
1314 printf ("Flat profile:\n\n"
1315 "Each sample counts as %g %s.\n",
1316 tick_unit
, profdata
->hist_hdr
->dimen
);
1317 fputs (" % cumulative self self total\n"
1318 " time seconds seconds calls us/call us/call name\n",
1321 for (n
= 0; n
< symidx
; ++n
)
1322 if (sortsym
[n
]->calls
!= 0 || sortsym
[n
]->ticks
!= 0)
1323 tsearch (sortsym
[n
], &data
, countorder
);
1325 twalk (data
, printflat
);
1327 tdestroy (data
, freenoop
);
1332 generate_call_graph (struct profdata
*profdata
)
1336 puts ("\nindex % time self children called name\n");
1338 for (cnt
= 0; cnt
< symidx
; ++cnt
)
1339 if (sortsym
[cnt
]->froms
!= NULL
|| sortsym
[cnt
]->tos
!= NULL
)
1341 struct arc_list
*runp
;
1344 /* First print the from-information. */
1345 runp
= sortsym
[cnt
]->froms
;
1346 while (runp
!= NULL
)
1348 printf (" %8.2f%8.2f%9" PRIdMAX
"/%-9" PRIdMAX
" %s",
1349 (runp
->idx
!= (size_t) -1l
1350 ? sortsym
[runp
->idx
]->ticks
* tick_unit
: 0.0),
1351 0.0, /* FIXME: what's time for the children, recursive */
1352 runp
->count
, sortsym
[cnt
]->calls
,
1353 (runp
->idx
!= (size_t) -1l ?
1354 sortsym
[runp
->idx
]->name
: "<UNKNOWN>"));
1356 if (runp
->idx
!= (size_t) -1l)
1357 printf (" [%Zd]", runp
->idx
);
1358 putchar_unlocked ('\n');
1363 /* Info abount the function itself. */
1364 n
= printf ("[%Zu]", cnt
);
1365 printf ("%*s%5.1f%8.2f%8.2f%9" PRIdMAX
" %s [%Zd]\n",
1367 total_ticks
? (100.0 * sortsym
[cnt
]->ticks
) / total_ticks
: 0,
1368 sortsym
[cnt
]->ticks
* tick_unit
,
1369 0.0, /* FIXME: what's time for the children, recursive */
1370 sortsym
[cnt
]->calls
,
1371 sortsym
[cnt
]->name
, cnt
);
1373 /* Info about the functions this function calls. */
1374 runp
= sortsym
[cnt
]->tos
;
1375 while (runp
!= NULL
)
1377 printf (" %8.2f%8.2f%9" PRIdMAX
"/",
1378 (runp
->idx
!= (size_t) -1l
1379 ? sortsym
[runp
->idx
]->ticks
* tick_unit
: 0.0),
1380 0.0, /* FIXME: what's time for the children, recursive */
1383 if (runp
->idx
!= (size_t) -1l)
1384 printf ("%-9" PRIdMAX
" %s [%Zd]\n",
1385 sortsym
[runp
->idx
]->calls
,
1386 sortsym
[runp
->idx
]->name
,
1389 fputs ("??? <UNKNOWN>\n\n", stdout
);
1394 fputs ("-----------------------------------------------\n", stdout
);
1400 generate_call_pair_list (struct profdata
*profdata
)
1404 for (cnt
= 0; cnt
< symidx
; ++cnt
)
1405 if (sortsym
[cnt
]->froms
!= NULL
|| sortsym
[cnt
]->tos
!= NULL
)
1407 struct arc_list
*runp
;
1409 /* First print the incoming arcs. */
1410 runp
= sortsym
[cnt
]->froms
;
1411 while (runp
!= NULL
)
1413 if (runp
->idx
== (size_t) -1l)
1415 <UNKNOWN> %-34s %9" PRIdMAX
"\n",
1416 sortsym
[cnt
]->name
, runp
->count
);
1420 /* Next the outgoing arcs. */
1421 runp
= sortsym
[cnt
]->tos
;
1422 while (runp
!= NULL
)
1424 printf ("%-34s %-34s %9" PRIdMAX
"\n",
1426 (runp
->idx
!= (size_t) -1l
1427 ? sortsym
[runp
->idx
]->name
: "<UNKNOWN>"),