1
linux/tools/perf/builtin-annotate.c
Arnaldo Carvalho de Melo 48fb4fdd6b perf annotate: Handle samples not at objdump output addr boundaries
Without this patch we get this for need_resched:

[root@mica ~]# perf annotate need_resched

------------------------------------------------
 Percent |      Source code & Disassembly of vmlinux
------------------------------------------------
         :
         :
         :      Disassembly of section .text:
         :
         :      ffffffff810095ed <need_resched>:
         :              return (state & TASK_INTERRUPTIBLE) || __fatal_signal_pending(p);
         :      }
         :
         :      static inline int need_resched(void)
         :      {
    0.00 :      ffffffff810095ed:       55                      push   %rbp
         :              return unlikely(test_thread_flag(TIF_NEED_RESCHED));
    0.00 :      ffffffff810095ee:       be 03 00 00 00          mov    $0x3,%esi
         :
         :      static inline struct thread_info *current_thread_info(void)
         :      {
         :              struct thread_info *ti;
         :              ti = (void *)(percpu_read_stable(kernel_stack) +
    0.00 :      ffffffff810095f3:       65 48 8b 3c 25 48 b5    mov    %gs:0xb548,%rdi
    0.00 :      ffffffff810095fa:       00 00
         :              return (state & TASK_INTERRUPTIBLE) || __fatal_signal_pending(p);
         :      }
         :
         :      static inline int need_resched(void)
         :      {
    0.00 :      ffffffff810095fc:       48 89 e5                mov    %rsp,%rbp
         :              return unlikely(test_thread_flag(TIF_NEED_RESCHED));
    0.00 :      ffffffff810095ff:       48 81 ef d8 1f 00 00    sub    $0x1fd8,%rdi
    0.00 :      ffffffff81009606:       e8 9d ff ff ff          callq  ffffffff810095a8 <test_ti_thread_flag>
         :      }
    0.00 :      ffffffff8100960b:       c9                      leaveq
    0.00 :      ffffffff8100960c:       85 c0                   test   %eax,%eax
    0.00 :      ffffffff8100960e:       0f 95 c0                setne  %al
    0.00 :      ffffffff81009611:       0f b6 c0                movzbl %al,%eax
         :      Disassembly of section .vsyscall_0:
         :      Disassembly of section .vsyscall_fn:
         :      Disassembly of section .vsyscall_1:
         :      Disassembly of section .vsyscall_2:
         :      Disassembly of section .init.text:
         :      Disassembly of section .altinstr_replacement:
         :      Disassembly of section .exit.text:
[root@mica ~]#

But from the 'perf report' result we know that there are hits
for need_resched on a 4 way machine mostly doing nothing, so
after adding code to show what is in each hist offset and
collapsing IP hits for what happens between objdump lines we
get, for the same perf.data file:

[root@mica ~]# perf annotate -v need_resched

------------------------------------------------
 Percent |      Source code & Disassembly of vmlinux
------------------------------------------------
         :
         :
         :      Disassembly of section .text:
         :
         :      ffffffff810095ed <need_resched>:
         :              return (state & TASK_INTERRUPTIBLE) || __fatal_signal_pending(p);
         :      }
         :
         :      static inline int need_resched(void)
         :      {
    0.00 :      ffffffff810095ed:       55                      push   %rbp
         :              return unlikely(test_thread_flag(TIF_NEED_RESCHED));
   52.78 :      ffffffff810095ee:       be 03 00 00 00          mov    $0x3,%esi
         :
         :      static inline struct thread_info *current_thread_info(void)
         :      {
         :              struct thread_info *ti;
         :              ti = (void *)(percpu_read_stable(kernel_stack) +
    0.00 :      ffffffff810095f3:       65 48 8b 3c 25 48 b5    mov    %gs:0xb548,%rdi
    0.00 :      ffffffff810095fa:       00 00
         :              return (state & TASK_INTERRUPTIBLE) || __fatal_signal_pending(p);
         :      }
         :
         :      static inline int need_resched(void)
         :      {
    0.00 :      ffffffff810095fc:       48 89 e5                mov    %rsp,%rbp
         :              return unlikely(test_thread_flag(TIF_NEED_RESCHED));
    9.72 :      ffffffff810095ff:       48 81 ef d8 1f 00 00    sub    $0x1fd8,%rdi
    0.00 :      ffffffff81009606:       e8 9d ff ff ff          callq  ffffffff810095a8 <test_ti_thread_flag>
         :      }
    0.00 :      ffffffff8100960b:       c9                      leaveq
    0.00 :      ffffffff8100960c:       85 c0                   test   %eax,%eax
   37.50 :      ffffffff8100960e:       0f 95 c0                setne  %al
    0.00 :      ffffffff81009611:       0f b6 c0                movzbl %al,%eax
         :      Disassembly of section .vsyscall_0:
         :      Disassembly of section .vsyscall_fn:
         :      Disassembly of section .vsyscall_1:
         :      Disassembly of section .vsyscall_2:
         :      Disassembly of section .init.text:
         :      Disassembly of section .altinstr_replacement:
         :      Disassembly of section .exit.text:
[root@mica ~]#

And now 'perf annotate -v', verbose mode, will show the hits per
precise IP, so that one can make sense of the attribution to
each objdumop line:

[root@mica ~]# perf annotate -v need_resched
Looking at the vmlinux_path (5 entries long)
Using /lib/modules/2.6.33-rc8-tip-00784-g3471df5-dirty/build/vmlinux
for symbols annotate_sym: filename=/lib/modules/2.6.33-rc8-tip-00784-g3471df5-dirty/build/vmlinux, sym=need_resched, start=0xffffffff810095ed, end=0xffffffff81009614

------------------------------------------------
 Percent |      Source code & Disassembly of vmlinux
------------------------------------------------
                ffffffff810095f1: 152
                ffffffff81009603: 28
                ffffffff8100960f: 55
                ffffffff81009610: 53
                          h->sum: 288
<SNIP same annotation>

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Miller <davem@davemloft.net>
Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <1267194194-15670-1-git-send-email-acme@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-02-26 15:42:49 +01:00

634 lines
14 KiB
C

/*
* builtin-annotate.c
*
* Builtin annotate command: Analyze the perf.data input file,
* look up and read DSOs and symbol information and display
* a histogram of results, along various sorting keys.
*/
#include "builtin.h"
#include "util/util.h"
#include "util/color.h"
#include <linux/list.h>
#include "util/cache.h"
#include <linux/rbtree.h>
#include "util/symbol.h"
#include "util/string.h"
#include "perf.h"
#include "util/debug.h"
#include "util/event.h"
#include "util/parse-options.h"
#include "util/parse-events.h"
#include "util/thread.h"
#include "util/sort.h"
#include "util/hist.h"
#include "util/session.h"
static char const *input_name = "perf.data";
static int force;
static int full_paths;
static int print_line;
struct sym_hist {
u64 sum;
u64 ip[0];
};
struct sym_ext {
struct rb_node node;
double percent;
char *path;
};
struct sym_priv {
struct sym_hist *hist;
struct sym_ext *ext;
};
static const char *sym_hist_filter;
static int sym__alloc_hist(struct symbol *self)
{
struct sym_priv *priv = symbol__priv(self);
const int size = (sizeof(*priv->hist) +
(self->end - self->start) * sizeof(u64));
priv->hist = zalloc(size);
return priv->hist == NULL ? -1 : 0;
}
/*
* collect histogram counts
*/
static int annotate__hist_hit(struct hist_entry *he, u64 ip)
{
unsigned int sym_size, offset;
struct symbol *sym = he->sym;
struct sym_priv *priv;
struct sym_hist *h;
he->count++;
if (!sym || !he->map)
return 0;
priv = symbol__priv(sym);
if (priv->hist == NULL && sym__alloc_hist(sym) < 0)
return -ENOMEM;
sym_size = sym->end - sym->start;
offset = ip - sym->start;
pr_debug3("%s: ip=%#Lx\n", __func__, he->map->unmap_ip(he->map, ip));
if (offset >= sym_size)
return 0;
h = priv->hist;
h->sum++;
h->ip[offset]++;
pr_debug3("%#Lx %s: count++ [ip: %#Lx, %#Lx] => %Ld\n", he->sym->start,
he->sym->name, ip, ip - he->sym->start, h->ip[offset]);
return 0;
}
static int perf_session__add_hist_entry(struct perf_session *self,
struct addr_location *al, u64 count)
{
bool hit;
struct hist_entry *he;
if (sym_hist_filter != NULL &&
(al->sym == NULL || strcmp(sym_hist_filter, al->sym->name) != 0)) {
/* We're only interested in a symbol named sym_hist_filter */
if (al->sym != NULL) {
rb_erase(&al->sym->rb_node,
&al->map->dso->symbols[al->map->type]);
symbol__delete(al->sym);
}
return 0;
}
he = __perf_session__add_hist_entry(self, al, NULL, count, &hit);
if (he == NULL)
return -ENOMEM;
return annotate__hist_hit(he, al->addr);
}
static int process_sample_event(event_t *event, struct perf_session *session)
{
struct addr_location al;
dump_printf("(IP, %d): %d: %#Lx\n", event->header.misc,
event->ip.pid, event->ip.ip);
if (event__preprocess_sample(event, session, &al, NULL) < 0) {
pr_warning("problem processing %d event, skipping it.\n",
event->header.type);
return -1;
}
if (!al.filtered && perf_session__add_hist_entry(session, &al, 1)) {
pr_warning("problem incrementing symbol count, "
"skipping event\n");
return -1;
}
return 0;
}
struct objdump_line {
struct list_head node;
s64 offset;
char *line;
};
static struct objdump_line *objdump_line__new(s64 offset, char *line)
{
struct objdump_line *self = malloc(sizeof(*self));
if (self != NULL) {
self->offset = offset;
self->line = line;
}
return self;
}
static void objdump_line__free(struct objdump_line *self)
{
free(self->line);
free(self);
}
static void objdump__add_line(struct list_head *head, struct objdump_line *line)
{
list_add_tail(&line->node, head);
}
static struct objdump_line *objdump__get_next_ip_line(struct list_head *head,
struct objdump_line *pos)
{
list_for_each_entry_continue(pos, head, node)
if (pos->offset >= 0)
return pos;
return NULL;
}
static int parse_line(FILE *file, struct hist_entry *he,
struct list_head *head)
{
struct symbol *sym = he->sym;
struct objdump_line *objdump_line;
char *line = NULL, *tmp, *tmp2;
size_t line_len;
s64 line_ip, offset = -1;
char *c;
if (getline(&line, &line_len, file) < 0)
return -1;
if (!line)
return -1;
c = strchr(line, '\n');
if (c)
*c = 0;
line_ip = -1;
/*
* Strip leading spaces:
*/
tmp = line;
while (*tmp) {
if (*tmp != ' ')
break;
tmp++;
}
if (*tmp) {
/*
* Parse hexa addresses followed by ':'
*/
line_ip = strtoull(tmp, &tmp2, 16);
if (*tmp2 != ':')
line_ip = -1;
}
if (line_ip != -1) {
u64 start = map__rip_2objdump(he->map, sym->start);
offset = line_ip - start;
}
objdump_line = objdump_line__new(offset, line);
if (objdump_line == NULL) {
free(line);
return -1;
}
objdump__add_line(head, objdump_line);
return 0;
}
static int objdump_line__print(struct objdump_line *self,
struct list_head *head,
struct hist_entry *he, u64 len)
{
struct symbol *sym = he->sym;
static const char *prev_line;
static const char *prev_color;
if (self->offset != -1) {
const char *path = NULL;
unsigned int hits = 0;
double percent = 0.0;
const char *color;
struct sym_priv *priv = symbol__priv(sym);
struct sym_ext *sym_ext = priv->ext;
struct sym_hist *h = priv->hist;
s64 offset = self->offset;
struct objdump_line *next = objdump__get_next_ip_line(head, self);
while (offset < (s64)len &&
(next == NULL || offset < next->offset)) {
if (sym_ext) {
if (path == NULL)
path = sym_ext[offset].path;
percent += sym_ext[offset].percent;
} else
hits += h->ip[offset];
++offset;
}
if (sym_ext == NULL && h->sum)
percent = 100.0 * hits / h->sum;
color = get_percent_color(percent);
/*
* Also color the filename and line if needed, with
* the same color than the percentage. Don't print it
* twice for close colored ip with the same filename:line
*/
if (path) {
if (!prev_line || strcmp(prev_line, path)
|| color != prev_color) {
color_fprintf(stdout, color, " %s", path);
prev_line = path;
prev_color = color;
}
}
color_fprintf(stdout, color, " %7.2f", percent);
printf(" : ");
color_fprintf(stdout, PERF_COLOR_BLUE, "%s\n", self->line);
} else {
if (!*self->line)
printf(" :\n");
else
printf(" : %s\n", self->line);
}
return 0;
}
static struct rb_root root_sym_ext;
static void insert_source_line(struct sym_ext *sym_ext)
{
struct sym_ext *iter;
struct rb_node **p = &root_sym_ext.rb_node;
struct rb_node *parent = NULL;
while (*p != NULL) {
parent = *p;
iter = rb_entry(parent, struct sym_ext, node);
if (sym_ext->percent > iter->percent)
p = &(*p)->rb_left;
else
p = &(*p)->rb_right;
}
rb_link_node(&sym_ext->node, parent, p);
rb_insert_color(&sym_ext->node, &root_sym_ext);
}
static void free_source_line(struct hist_entry *he, int len)
{
struct sym_priv *priv = symbol__priv(he->sym);
struct sym_ext *sym_ext = priv->ext;
int i;
if (!sym_ext)
return;
for (i = 0; i < len; i++)
free(sym_ext[i].path);
free(sym_ext);
priv->ext = NULL;
root_sym_ext = RB_ROOT;
}
/* Get the filename:line for the colored entries */
static void
get_source_line(struct hist_entry *he, int len, const char *filename)
{
struct symbol *sym = he->sym;
u64 start;
int i;
char cmd[PATH_MAX * 2];
struct sym_ext *sym_ext;
struct sym_priv *priv = symbol__priv(sym);
struct sym_hist *h = priv->hist;
if (!h->sum)
return;
sym_ext = priv->ext = calloc(len, sizeof(struct sym_ext));
if (!priv->ext)
return;
start = he->map->unmap_ip(he->map, sym->start);
for (i = 0; i < len; i++) {
char *path = NULL;
size_t line_len;
u64 offset;
FILE *fp;
sym_ext[i].percent = 100.0 * h->ip[i] / h->sum;
if (sym_ext[i].percent <= 0.5)
continue;
offset = start + i;
sprintf(cmd, "addr2line -e %s %016llx", filename, offset);
fp = popen(cmd, "r");
if (!fp)
continue;
if (getline(&path, &line_len, fp) < 0 || !line_len)
goto next;
sym_ext[i].path = malloc(sizeof(char) * line_len + 1);
if (!sym_ext[i].path)
goto next;
strcpy(sym_ext[i].path, path);
insert_source_line(&sym_ext[i]);
next:
pclose(fp);
}
}
static void print_summary(const char *filename)
{
struct sym_ext *sym_ext;
struct rb_node *node;
printf("\nSorted summary for file %s\n", filename);
printf("----------------------------------------------\n\n");
if (RB_EMPTY_ROOT(&root_sym_ext)) {
printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
return;
}
node = rb_first(&root_sym_ext);
while (node) {
double percent;
const char *color;
char *path;
sym_ext = rb_entry(node, struct sym_ext, node);
percent = sym_ext->percent;
color = get_percent_color(percent);
path = sym_ext->path;
color_fprintf(stdout, color, " %7.2f %s", percent, path);
node = rb_next(node);
}
}
static void hist_entry__print_hits(struct hist_entry *self)
{
struct symbol *sym = self->sym;
struct sym_priv *priv = symbol__priv(sym);
struct sym_hist *h = priv->hist;
u64 len = sym->end - sym->start, offset;
for (offset = 0; offset < len; ++offset)
if (h->ip[offset] != 0)
printf("%*Lx: %Lu\n", BITS_PER_LONG / 2,
sym->start + offset, h->ip[offset]);
printf("%*s: %Lu\n", BITS_PER_LONG / 2, "h->sum", h->sum);
}
static void annotate_sym(struct hist_entry *he)
{
struct map *map = he->map;
struct dso *dso = map->dso;
struct symbol *sym = he->sym;
const char *filename = dso->long_name, *d_filename;
u64 len;
char command[PATH_MAX*2];
FILE *file;
LIST_HEAD(head);
struct objdump_line *pos, *n;
if (!filename)
return;
pr_debug("%s: filename=%s, sym=%s, start=%#Lx, end=%#Lx\n", __func__,
filename, sym->name, map->unmap_ip(map, sym->start),
map->unmap_ip(map, sym->end));
if (full_paths)
d_filename = filename;
else
d_filename = basename(filename);
len = sym->end - sym->start;
if (print_line) {
get_source_line(he, len, filename);
print_summary(filename);
}
printf("\n\n------------------------------------------------\n");
printf(" Percent | Source code & Disassembly of %s\n", d_filename);
printf("------------------------------------------------\n");
if (verbose >= 2)
printf("annotating [%p] %30s : [%p] %30s\n",
dso, dso->long_name, sym, sym->name);
sprintf(command, "objdump --start-address=0x%016Lx --stop-address=0x%016Lx -dS %s|grep -v %s",
map__rip_2objdump(map, sym->start),
map__rip_2objdump(map, sym->end),
filename, filename);
if (verbose >= 3)
printf("doing: %s\n", command);
file = popen(command, "r");
if (!file)
return;
while (!feof(file)) {
if (parse_line(file, he, &head) < 0)
break;
}
pclose(file);
if (verbose)
hist_entry__print_hits(he);
list_for_each_entry_safe(pos, n, &head, node) {
objdump_line__print(pos, &head, he, len);
list_del(&pos->node);
objdump_line__free(pos);
}
if (print_line)
free_source_line(he, len);
}
static void perf_session__find_annotations(struct perf_session *self)
{
struct rb_node *nd;
for (nd = rb_first(&self->hists); nd; nd = rb_next(nd)) {
struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
struct sym_priv *priv;
if (he->sym == NULL)
continue;
priv = symbol__priv(he->sym);
if (priv->hist == NULL)
continue;
annotate_sym(he);
/*
* Since we have a hist_entry per IP for the same symbol, free
* he->sym->hist to signal we already processed this symbol.
*/
free(priv->hist);
priv->hist = NULL;
}
}
static struct perf_event_ops event_ops = {
.sample = process_sample_event,
.mmap = event__process_mmap,
.comm = event__process_comm,
.fork = event__process_task,
};
static int __cmd_annotate(void)
{
int ret;
struct perf_session *session;
session = perf_session__new(input_name, O_RDONLY, force);
if (session == NULL)
return -ENOMEM;
ret = perf_session__process_events(session, &event_ops);
if (ret)
goto out_delete;
if (dump_trace) {
event__print_totals();
goto out_delete;
}
if (verbose > 3)
perf_session__fprintf(session, stdout);
if (verbose > 2)
dsos__fprintf(stdout);
perf_session__collapse_resort(session);
perf_session__output_resort(session, session->event_total[0]);
perf_session__find_annotations(session);
out_delete:
perf_session__delete(session);
return ret;
}
static const char * const annotate_usage[] = {
"perf annotate [<options>] <command>",
NULL
};
static const struct option options[] = {
OPT_STRING('i', "input", &input_name, "file",
"input file name"),
OPT_STRING('s', "symbol", &sym_hist_filter, "symbol",
"symbol to annotate"),
OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
OPT_BOOLEAN('v', "verbose", &verbose,
"be more verbose (show symbol address, etc)"),
OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
"dump raw trace in ASCII"),
OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
"file", "vmlinux pathname"),
OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
"load module symbols - WARNING: use only with -k and LIVE kernel"),
OPT_BOOLEAN('l', "print-line", &print_line,
"print matching source lines (may be slow)"),
OPT_BOOLEAN('P', "full-paths", &full_paths,
"Don't shorten the displayed pathnames"),
OPT_END()
};
int cmd_annotate(int argc, const char **argv, const char *prefix __used)
{
argc = parse_options(argc, argv, options, annotate_usage, 0);
symbol_conf.priv_size = sizeof(struct sym_priv);
symbol_conf.try_vmlinux_path = true;
if (symbol__init() < 0)
return -1;
setup_sorting(annotate_usage, options);
if (argc) {
/*
* Special case: if there's an argument left then assume tha
* it's a symbol filter:
*/
if (argc > 1)
usage_with_options(annotate_usage, options);
sym_hist_filter = argv[0];
}
setup_pager();
if (field_sep && *field_sep == '.') {
pr_err("'.' is the only non valid --field-separator argument\n");
return -1;
}
return __cmd_annotate();
}