2011-01-30 19:46:46 +07:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
|
|
|
|
*
|
|
|
|
* Parts came from builtin-{top,stat,record}.c, see those files for further
|
|
|
|
* copyright notes.
|
|
|
|
*
|
|
|
|
* Released under the GPL v2. (and only v2, not any later version)
|
|
|
|
*/
|
2011-11-05 17:41:51 +07:00
|
|
|
#include "util.h"
|
|
|
|
#include "debugfs.h"
|
2011-01-12 07:30:02 +07:00
|
|
|
#include <poll.h>
|
2011-01-30 19:46:46 +07:00
|
|
|
#include "cpumap.h"
|
|
|
|
#include "thread_map.h"
|
2011-01-12 05:56:53 +07:00
|
|
|
#include "evlist.h"
|
|
|
|
#include "evsel.h"
|
2011-11-09 17:47:15 +07:00
|
|
|
#include <unistd.h>
|
2011-01-12 05:56:53 +07:00
|
|
|
|
2011-11-04 18:10:59 +07:00
|
|
|
#include "parse-events.h"
|
|
|
|
|
2011-01-30 19:46:46 +07:00
|
|
|
#include <sys/mman.h>
|
|
|
|
|
2011-01-13 07:39:13 +07:00
|
|
|
#include <linux/bitops.h>
|
|
|
|
#include <linux/hash.h>
|
|
|
|
|
2011-01-30 19:46:46 +07:00
|
|
|
#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
|
2011-03-10 21:15:54 +07:00
|
|
|
#define SID(e, x, y) xyarray__entry(e->sample_id, x, y)
|
2011-01-30 19:46:46 +07:00
|
|
|
|
2011-01-30 20:59:43 +07:00
|
|
|
void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus,
|
|
|
|
struct thread_map *threads)
|
2011-01-19 06:41:45 +07:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i)
|
|
|
|
INIT_HLIST_HEAD(&evlist->heads[i]);
|
|
|
|
INIT_LIST_HEAD(&evlist->entries);
|
2011-01-30 20:59:43 +07:00
|
|
|
perf_evlist__set_maps(evlist, cpus, threads);
|
2011-11-09 17:47:15 +07:00
|
|
|
evlist->workload.pid = -1;
|
2011-01-19 06:41:45 +07:00
|
|
|
}
|
|
|
|
|
2011-01-30 20:59:43 +07:00
|
|
|
struct perf_evlist *perf_evlist__new(struct cpu_map *cpus,
|
|
|
|
struct thread_map *threads)
|
2011-01-12 05:56:53 +07:00
|
|
|
{
|
|
|
|
struct perf_evlist *evlist = zalloc(sizeof(*evlist));
|
|
|
|
|
2011-01-19 06:41:45 +07:00
|
|
|
if (evlist != NULL)
|
2011-01-30 20:59:43 +07:00
|
|
|
perf_evlist__init(evlist, cpus, threads);
|
2011-01-12 05:56:53 +07:00
|
|
|
|
|
|
|
return evlist;
|
|
|
|
}
|
|
|
|
|
2011-11-08 23:41:57 +07:00
|
|
|
void perf_evlist__config_attrs(struct perf_evlist *evlist,
|
|
|
|
struct perf_record_opts *opts)
|
|
|
|
{
|
|
|
|
struct perf_evsel *evsel;
|
|
|
|
|
|
|
|
if (evlist->cpus->map[0] < 0)
|
|
|
|
opts->no_inherit = true;
|
|
|
|
|
|
|
|
list_for_each_entry(evsel, &evlist->entries, node) {
|
|
|
|
perf_evsel__config(evsel, opts);
|
|
|
|
|
|
|
|
if (evlist->nr_entries > 1)
|
|
|
|
evsel->attr.sample_type |= PERF_SAMPLE_ID;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-01-12 05:56:53 +07:00
|
|
|
static void perf_evlist__purge(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
struct perf_evsel *pos, *n;
|
|
|
|
|
|
|
|
list_for_each_entry_safe(pos, n, &evlist->entries, node) {
|
|
|
|
list_del_init(&pos->node);
|
|
|
|
perf_evsel__delete(pos);
|
|
|
|
}
|
|
|
|
|
|
|
|
evlist->nr_entries = 0;
|
|
|
|
}
|
|
|
|
|
2011-01-19 06:41:45 +07:00
|
|
|
void perf_evlist__exit(struct perf_evlist *evlist)
|
2011-01-12 05:56:53 +07:00
|
|
|
{
|
2011-01-13 07:39:13 +07:00
|
|
|
free(evlist->mmap);
|
2011-01-12 07:30:02 +07:00
|
|
|
free(evlist->pollfd);
|
2011-01-19 06:41:45 +07:00
|
|
|
evlist->mmap = NULL;
|
|
|
|
evlist->pollfd = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
void perf_evlist__delete(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
perf_evlist__purge(evlist);
|
|
|
|
perf_evlist__exit(evlist);
|
2011-01-12 05:56:53 +07:00
|
|
|
free(evlist);
|
|
|
|
}
|
|
|
|
|
|
|
|
void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry)
|
|
|
|
{
|
|
|
|
list_add_tail(&entry->node, &evlist->entries);
|
|
|
|
++evlist->nr_entries;
|
|
|
|
}
|
|
|
|
|
2011-11-04 18:10:59 +07:00
|
|
|
static void perf_evlist__splice_list_tail(struct perf_evlist *evlist,
|
|
|
|
struct list_head *list,
|
|
|
|
int nr_entries)
|
|
|
|
{
|
|
|
|
list_splice_tail(list, &evlist->entries);
|
|
|
|
evlist->nr_entries += nr_entries;
|
|
|
|
}
|
|
|
|
|
2011-01-12 05:56:53 +07:00
|
|
|
int perf_evlist__add_default(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
struct perf_event_attr attr = {
|
|
|
|
.type = PERF_TYPE_HARDWARE,
|
|
|
|
.config = PERF_COUNT_HW_CPU_CYCLES,
|
|
|
|
};
|
|
|
|
struct perf_evsel *evsel = perf_evsel__new(&attr, 0);
|
|
|
|
|
|
|
|
if (evsel == NULL)
|
2011-06-07 23:19:36 +07:00
|
|
|
goto error;
|
|
|
|
|
|
|
|
/* use strdup() because free(evsel) assumes name is allocated */
|
|
|
|
evsel->name = strdup("cycles");
|
|
|
|
if (!evsel->name)
|
|
|
|
goto error_free;
|
2011-01-12 05:56:53 +07:00
|
|
|
|
|
|
|
perf_evlist__add(evlist, evsel);
|
|
|
|
return 0;
|
2011-06-07 23:19:36 +07:00
|
|
|
error_free:
|
|
|
|
perf_evsel__delete(evsel);
|
|
|
|
error:
|
|
|
|
return -ENOMEM;
|
2011-01-12 05:56:53 +07:00
|
|
|
}
|
2011-01-12 07:30:02 +07:00
|
|
|
|
2011-11-04 18:10:59 +07:00
|
|
|
int perf_evlist__add_attrs(struct perf_evlist *evlist,
|
|
|
|
struct perf_event_attr *attrs, size_t nr_attrs)
|
|
|
|
{
|
|
|
|
struct perf_evsel *evsel, *n;
|
|
|
|
LIST_HEAD(head);
|
|
|
|
size_t i;
|
|
|
|
|
|
|
|
for (i = 0; i < nr_attrs; i++) {
|
|
|
|
evsel = perf_evsel__new(attrs + i, evlist->nr_entries + i);
|
|
|
|
if (evsel == NULL)
|
|
|
|
goto out_delete_partial_list;
|
|
|
|
list_add_tail(&evsel->node, &head);
|
|
|
|
}
|
|
|
|
|
|
|
|
perf_evlist__splice_list_tail(evlist, &head, nr_attrs);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
out_delete_partial_list:
|
|
|
|
list_for_each_entry_safe(evsel, n, &head, node)
|
|
|
|
perf_evsel__delete(evsel);
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2011-11-05 17:41:51 +07:00
|
|
|
static int trace_event__id(const char *evname)
|
|
|
|
{
|
|
|
|
char *filename, *colon;
|
|
|
|
int err = -1, fd;
|
|
|
|
|
|
|
|
if (asprintf(&filename, "%s/%s/id", tracing_events_path, evname) < 0)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
colon = strrchr(filename, ':');
|
|
|
|
if (colon != NULL)
|
|
|
|
*colon = '/';
|
|
|
|
|
|
|
|
fd = open(filename, O_RDONLY);
|
|
|
|
if (fd >= 0) {
|
|
|
|
char id[16];
|
|
|
|
if (read(fd, id, sizeof(id)) > 0)
|
|
|
|
err = atoi(id);
|
|
|
|
close(fd);
|
|
|
|
}
|
|
|
|
|
|
|
|
free(filename);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
int perf_evlist__add_tracepoints(struct perf_evlist *evlist,
|
|
|
|
const char *tracepoints[],
|
|
|
|
size_t nr_tracepoints)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
size_t i;
|
|
|
|
struct perf_event_attr *attrs = zalloc(nr_tracepoints * sizeof(*attrs));
|
|
|
|
|
|
|
|
if (attrs == NULL)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
for (i = 0; i < nr_tracepoints; i++) {
|
|
|
|
err = trace_event__id(tracepoints[i]);
|
|
|
|
|
|
|
|
if (err < 0)
|
|
|
|
goto out_free_attrs;
|
|
|
|
|
|
|
|
attrs[i].type = PERF_TYPE_TRACEPOINT;
|
|
|
|
attrs[i].config = err;
|
|
|
|
attrs[i].sample_type = (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME |
|
|
|
|
PERF_SAMPLE_CPU);
|
|
|
|
attrs[i].sample_period = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = perf_evlist__add_attrs(evlist, attrs, nr_tracepoints);
|
|
|
|
out_free_attrs:
|
|
|
|
free(attrs);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
perf tools: Save some loops using perf_evlist__id2evsel
Since we already ask for PERF_SAMPLE_ID and use it to quickly find the
associated evsel, add handler func + data to struct perf_evsel to avoid
using chains of if(strcmp(event_name)) and also to avoid all the linear
list searches via trace_event_find.
To demonstrate the technique convert 'perf sched' to it:
# perf sched record sleep 5m
And then:
Performance counter stats for '/tmp/oldperf sched lat':
646.929438 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,901 page-faults # 0.032 M/sec
1,290,144,450 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,606,158,439 instructions # 1.24 insns per cycle
339,088,395 branches # 524.151 M/sec
4,550,735 branch-misses # 1.34% of all branches
0.647524759 seconds time elapsed
Versus:
Performance counter stats for 'perf sched lat':
473.564691 task-clock # 0.999 CPUs utilized
9 context-switches # 0.000 M/sec
0 CPU-migrations # 0.000 M/sec
20,903 page-faults # 0.044 M/sec
944,367,984 cycles # 1.994 GHz
<not supported> stalled-cycles-frontend
<not supported> stalled-cycles-backend
1,442,385,571 instructions # 1.53 insns per cycle
308,383,106 branches # 651.195 M/sec
4,481,784 branch-misses # 1.45% of all branches
0.474215751 seconds time elapsed
[root@emilia ~]#
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-1kbzpl74lwi6lavpqke2u2p3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-11-29 02:57:40 +07:00
|
|
|
static struct perf_evsel *
|
|
|
|
perf_evlist__find_tracepoint_by_id(struct perf_evlist *evlist, int id)
|
|
|
|
{
|
|
|
|
struct perf_evsel *evsel;
|
|
|
|
|
|
|
|
list_for_each_entry(evsel, &evlist->entries, node) {
|
|
|
|
if (evsel->attr.type == PERF_TYPE_TRACEPOINT &&
|
|
|
|
(int)evsel->attr.config == id)
|
|
|
|
return evsel;
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
int perf_evlist__set_tracepoints_handlers(struct perf_evlist *evlist,
|
|
|
|
const struct perf_evsel_str_handler *assocs,
|
|
|
|
size_t nr_assocs)
|
|
|
|
{
|
|
|
|
struct perf_evsel *evsel;
|
|
|
|
int err;
|
|
|
|
size_t i;
|
|
|
|
|
|
|
|
for (i = 0; i < nr_assocs; i++) {
|
|
|
|
err = trace_event__id(assocs[i].name);
|
|
|
|
if (err < 0)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
evsel = perf_evlist__find_tracepoint_by_id(evlist, err);
|
|
|
|
if (evsel == NULL)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
err = -EEXIST;
|
|
|
|
if (evsel->handler.func != NULL)
|
|
|
|
goto out;
|
|
|
|
evsel->handler.func = assocs[i].handler;
|
|
|
|
}
|
|
|
|
|
|
|
|
err = 0;
|
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2011-07-25 21:06:19 +07:00
|
|
|
void perf_evlist__disable(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
int cpu, thread;
|
|
|
|
struct perf_evsel *pos;
|
|
|
|
|
|
|
|
for (cpu = 0; cpu < evlist->cpus->nr; cpu++) {
|
|
|
|
list_for_each_entry(pos, &evlist->entries, node) {
|
|
|
|
for (thread = 0; thread < evlist->threads->nr; thread++)
|
|
|
|
ioctl(FD(pos, cpu, thread), PERF_EVENT_IOC_DISABLE);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-08-25 23:17:55 +07:00
|
|
|
void perf_evlist__enable(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
int cpu, thread;
|
|
|
|
struct perf_evsel *pos;
|
|
|
|
|
|
|
|
for (cpu = 0; cpu < evlist->cpus->nr; cpu++) {
|
|
|
|
list_for_each_entry(pos, &evlist->entries, node) {
|
|
|
|
for (thread = 0; thread < evlist->threads->nr; thread++)
|
|
|
|
ioctl(FD(pos, cpu, thread), PERF_EVENT_IOC_ENABLE);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-01-30 20:59:43 +07:00
|
|
|
int perf_evlist__alloc_pollfd(struct perf_evlist *evlist)
|
2011-01-12 07:30:02 +07:00
|
|
|
{
|
2011-01-30 20:59:43 +07:00
|
|
|
int nfds = evlist->cpus->nr * evlist->threads->nr * evlist->nr_entries;
|
2011-01-12 07:30:02 +07:00
|
|
|
evlist->pollfd = malloc(sizeof(struct pollfd) * nfds);
|
|
|
|
return evlist->pollfd != NULL ? 0 : -ENOMEM;
|
|
|
|
}
|
2011-01-13 02:03:24 +07:00
|
|
|
|
|
|
|
void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd)
|
|
|
|
{
|
|
|
|
fcntl(fd, F_SETFL, O_NONBLOCK);
|
|
|
|
evlist->pollfd[evlist->nr_fds].fd = fd;
|
|
|
|
evlist->pollfd[evlist->nr_fds].events = POLLIN;
|
|
|
|
evlist->nr_fds++;
|
|
|
|
}
|
2011-01-13 07:39:13 +07:00
|
|
|
|
2011-03-10 21:15:54 +07:00
|
|
|
static void perf_evlist__id_hash(struct perf_evlist *evlist,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
int cpu, int thread, u64 id)
|
2011-03-05 08:29:39 +07:00
|
|
|
{
|
|
|
|
int hash;
|
|
|
|
struct perf_sample_id *sid = SID(evsel, cpu, thread);
|
|
|
|
|
|
|
|
sid->id = id;
|
|
|
|
sid->evsel = evsel;
|
|
|
|
hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS);
|
|
|
|
hlist_add_head(&sid->node, &evlist->heads[hash]);
|
|
|
|
}
|
|
|
|
|
2011-03-10 21:15:54 +07:00
|
|
|
void perf_evlist__id_add(struct perf_evlist *evlist, struct perf_evsel *evsel,
|
|
|
|
int cpu, int thread, u64 id)
|
|
|
|
{
|
|
|
|
perf_evlist__id_hash(evlist, evsel, cpu, thread, id);
|
|
|
|
evsel->id[evsel->ids++] = id;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int perf_evlist__id_add_fd(struct perf_evlist *evlist,
|
|
|
|
struct perf_evsel *evsel,
|
|
|
|
int cpu, int thread, int fd)
|
2011-01-30 19:46:46 +07:00
|
|
|
{
|
|
|
|
u64 read_data[4] = { 0, };
|
2011-03-05 08:29:39 +07:00
|
|
|
int id_idx = 1; /* The first entry is the counter value */
|
2011-01-30 19:46:46 +07:00
|
|
|
|
|
|
|
if (!(evsel->attr.read_format & PERF_FORMAT_ID) ||
|
|
|
|
read(fd, &read_data, sizeof(read_data)) == -1)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
|
|
|
|
++id_idx;
|
|
|
|
if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
|
|
|
|
++id_idx;
|
|
|
|
|
2011-03-10 21:15:54 +07:00
|
|
|
perf_evlist__id_add(evlist, evsel, cpu, thread, read_data[id_idx]);
|
2011-01-30 19:46:46 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-01-13 07:39:13 +07:00
|
|
|
struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id)
|
|
|
|
{
|
|
|
|
struct hlist_head *head;
|
|
|
|
struct hlist_node *pos;
|
|
|
|
struct perf_sample_id *sid;
|
|
|
|
int hash;
|
|
|
|
|
|
|
|
if (evlist->nr_entries == 1)
|
|
|
|
return list_entry(evlist->entries.next, struct perf_evsel, node);
|
|
|
|
|
|
|
|
hash = hash_64(id, PERF_EVLIST__HLIST_BITS);
|
|
|
|
head = &evlist->heads[hash];
|
|
|
|
|
|
|
|
hlist_for_each_entry(sid, pos, head, node)
|
|
|
|
if (sid->id == id)
|
|
|
|
return sid->evsel;
|
|
|
|
return NULL;
|
|
|
|
}
|
2011-01-15 19:40:59 +07:00
|
|
|
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx)
|
2011-01-15 19:40:59 +07:00
|
|
|
{
|
|
|
|
/* XXX Move this to perf.c, making it generally available */
|
|
|
|
unsigned int page_size = sysconf(_SC_PAGE_SIZE);
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
struct perf_mmap *md = &evlist->mmap[idx];
|
2011-01-15 19:40:59 +07:00
|
|
|
unsigned int head = perf_mmap__read_head(md);
|
|
|
|
unsigned int old = md->prev;
|
|
|
|
unsigned char *data = md->base + page_size;
|
2011-01-29 23:01:45 +07:00
|
|
|
union perf_event *event = NULL;
|
2011-01-15 19:40:59 +07:00
|
|
|
|
2011-01-29 18:08:13 +07:00
|
|
|
if (evlist->overwrite) {
|
2011-01-15 19:40:59 +07:00
|
|
|
/*
|
2011-01-29 18:08:13 +07:00
|
|
|
* If we're further behind than half the buffer, there's a chance
|
|
|
|
* the writer will bite our tail and mess up the samples under us.
|
|
|
|
*
|
|
|
|
* If we somehow ended up ahead of the head, we got messed up.
|
|
|
|
*
|
|
|
|
* In either case, truncate and restart at head.
|
2011-01-15 19:40:59 +07:00
|
|
|
*/
|
2011-01-29 18:08:13 +07:00
|
|
|
int diff = head - old;
|
|
|
|
if (diff > md->mask / 2 || diff < 0) {
|
|
|
|
fprintf(stderr, "WARNING: failed to keep up with mmap data.\n");
|
|
|
|
|
|
|
|
/*
|
|
|
|
* head points to a known good entry, start there.
|
|
|
|
*/
|
|
|
|
old = head;
|
|
|
|
}
|
2011-01-15 19:40:59 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
if (old != head) {
|
|
|
|
size_t size;
|
|
|
|
|
2011-01-29 23:01:45 +07:00
|
|
|
event = (union perf_event *)&data[old & md->mask];
|
2011-01-15 19:40:59 +07:00
|
|
|
size = event->header.size;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Event straddles the mmap boundary -- header should always
|
|
|
|
* be inside due to u64 alignment of output.
|
|
|
|
*/
|
|
|
|
if ((old & md->mask) + size != ((old + size) & md->mask)) {
|
|
|
|
unsigned int offset = old;
|
|
|
|
unsigned int len = min(sizeof(*event), size), cpy;
|
|
|
|
void *dst = &evlist->event_copy;
|
|
|
|
|
|
|
|
do {
|
|
|
|
cpy = min(md->mask + 1 - (offset & md->mask), len);
|
|
|
|
memcpy(dst, &data[offset & md->mask], cpy);
|
|
|
|
offset += cpy;
|
|
|
|
dst += cpy;
|
|
|
|
len -= cpy;
|
|
|
|
} while (len);
|
|
|
|
|
|
|
|
event = &evlist->event_copy;
|
|
|
|
}
|
|
|
|
|
|
|
|
old += size;
|
|
|
|
}
|
|
|
|
|
|
|
|
md->prev = old;
|
2011-01-29 18:08:13 +07:00
|
|
|
|
|
|
|
if (!evlist->overwrite)
|
|
|
|
perf_mmap__write_tail(md, old);
|
|
|
|
|
2011-01-15 19:40:59 +07:00
|
|
|
return event;
|
|
|
|
}
|
2011-01-30 19:46:46 +07:00
|
|
|
|
2011-01-30 20:59:43 +07:00
|
|
|
void perf_evlist__munmap(struct perf_evlist *evlist)
|
2011-01-30 19:46:46 +07:00
|
|
|
{
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
int i;
|
2011-01-30 19:46:46 +07:00
|
|
|
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
for (i = 0; i < evlist->nr_mmaps; i++) {
|
|
|
|
if (evlist->mmap[i].base != NULL) {
|
|
|
|
munmap(evlist->mmap[i].base, evlist->mmap_len);
|
|
|
|
evlist->mmap[i].base = NULL;
|
2011-01-30 19:46:46 +07:00
|
|
|
}
|
|
|
|
}
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
|
|
|
|
free(evlist->mmap);
|
|
|
|
evlist->mmap = NULL;
|
2011-01-30 19:46:46 +07:00
|
|
|
}
|
|
|
|
|
2011-01-30 20:59:43 +07:00
|
|
|
int perf_evlist__alloc_mmap(struct perf_evlist *evlist)
|
2011-01-30 19:46:46 +07:00
|
|
|
{
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
evlist->nr_mmaps = evlist->cpus->nr;
|
|
|
|
if (evlist->cpus->map[0] == -1)
|
|
|
|
evlist->nr_mmaps = evlist->threads->nr;
|
|
|
|
evlist->mmap = zalloc(evlist->nr_mmaps * sizeof(struct perf_mmap));
|
2011-01-30 19:46:46 +07:00
|
|
|
return evlist->mmap != NULL ? 0 : -ENOMEM;
|
|
|
|
}
|
|
|
|
|
2011-06-02 20:39:43 +07:00
|
|
|
static int __perf_evlist__mmap(struct perf_evlist *evlist,
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
int idx, int prot, int mask, int fd)
|
2011-01-30 19:46:46 +07:00
|
|
|
{
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
evlist->mmap[idx].prev = 0;
|
|
|
|
evlist->mmap[idx].mask = mask;
|
|
|
|
evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot,
|
2011-01-30 19:46:46 +07:00
|
|
|
MAP_SHARED, fd, 0);
|
2011-06-02 20:39:43 +07:00
|
|
|
if (evlist->mmap[idx].base == MAP_FAILED)
|
2011-01-30 19:46:46 +07:00
|
|
|
return -1;
|
|
|
|
|
|
|
|
perf_evlist__add_pollfd(evlist, fd);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mask)
|
|
|
|
{
|
|
|
|
struct perf_evsel *evsel;
|
|
|
|
int cpu, thread;
|
|
|
|
|
|
|
|
for (cpu = 0; cpu < evlist->cpus->nr; cpu++) {
|
|
|
|
int output = -1;
|
|
|
|
|
|
|
|
for (thread = 0; thread < evlist->threads->nr; thread++) {
|
|
|
|
list_for_each_entry(evsel, &evlist->entries, node) {
|
|
|
|
int fd = FD(evsel, cpu, thread);
|
|
|
|
|
|
|
|
if (output == -1) {
|
|
|
|
output = fd;
|
2011-06-02 20:39:43 +07:00
|
|
|
if (__perf_evlist__mmap(evlist, cpu,
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
prot, mask, output) < 0)
|
|
|
|
goto out_unmap;
|
|
|
|
} else {
|
|
|
|
if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0)
|
|
|
|
goto out_unmap;
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
|
|
|
|
perf_evlist__id_add_fd(evlist, evsel, cpu, thread, fd) < 0)
|
|
|
|
goto out_unmap;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
out_unmap:
|
|
|
|
for (cpu = 0; cpu < evlist->cpus->nr; cpu++) {
|
|
|
|
if (evlist->mmap[cpu].base != NULL) {
|
|
|
|
munmap(evlist->mmap[cpu].base, evlist->mmap_len);
|
|
|
|
evlist->mmap[cpu].base = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, int mask)
|
|
|
|
{
|
|
|
|
struct perf_evsel *evsel;
|
|
|
|
int thread;
|
|
|
|
|
|
|
|
for (thread = 0; thread < evlist->threads->nr; thread++) {
|
|
|
|
int output = -1;
|
|
|
|
|
|
|
|
list_for_each_entry(evsel, &evlist->entries, node) {
|
|
|
|
int fd = FD(evsel, 0, thread);
|
|
|
|
|
|
|
|
if (output == -1) {
|
|
|
|
output = fd;
|
2011-06-02 20:39:43 +07:00
|
|
|
if (__perf_evlist__mmap(evlist, thread,
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
prot, mask, output) < 0)
|
|
|
|
goto out_unmap;
|
|
|
|
} else {
|
|
|
|
if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0)
|
|
|
|
goto out_unmap;
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
|
|
|
|
perf_evlist__id_add_fd(evlist, evsel, 0, thread, fd) < 0)
|
|
|
|
goto out_unmap;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
out_unmap:
|
|
|
|
for (thread = 0; thread < evlist->threads->nr; thread++) {
|
|
|
|
if (evlist->mmap[thread].base != NULL) {
|
|
|
|
munmap(evlist->mmap[thread].base, evlist->mmap_len);
|
|
|
|
evlist->mmap[thread].base = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2011-01-30 19:46:46 +07:00
|
|
|
/** perf_evlist__mmap - Create per cpu maps to receive events
|
|
|
|
*
|
|
|
|
* @evlist - list of events
|
|
|
|
* @pages - map length in pages
|
|
|
|
* @overwrite - overwrite older events?
|
|
|
|
*
|
|
|
|
* If overwrite is false the user needs to signal event consuption using:
|
|
|
|
*
|
|
|
|
* struct perf_mmap *m = &evlist->mmap[cpu];
|
|
|
|
* unsigned int head = perf_mmap__read_head(m);
|
|
|
|
*
|
|
|
|
* perf_mmap__write_tail(m, head)
|
2011-01-30 20:59:43 +07:00
|
|
|
*
|
|
|
|
* Using perf_evlist__read_on_cpu does this automatically.
|
2011-01-30 19:46:46 +07:00
|
|
|
*/
|
2011-11-09 18:10:47 +07:00
|
|
|
int perf_evlist__mmap(struct perf_evlist *evlist, unsigned int pages,
|
|
|
|
bool overwrite)
|
2011-01-30 19:46:46 +07:00
|
|
|
{
|
|
|
|
unsigned int page_size = sysconf(_SC_PAGE_SIZE);
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
struct perf_evsel *evsel;
|
2011-01-30 20:59:43 +07:00
|
|
|
const struct cpu_map *cpus = evlist->cpus;
|
|
|
|
const struct thread_map *threads = evlist->threads;
|
2011-11-09 18:10:47 +07:00
|
|
|
int prot = PROT_READ | (overwrite ? 0 : PROT_WRITE), mask;
|
|
|
|
|
|
|
|
/* 512 kiB: default amount of unprivileged mlocked memory */
|
|
|
|
if (pages == UINT_MAX)
|
|
|
|
pages = (512 * 1024) / page_size;
|
|
|
|
|
|
|
|
mask = pages * page_size - 1;
|
2011-01-30 19:46:46 +07:00
|
|
|
|
2011-01-30 20:59:43 +07:00
|
|
|
if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0)
|
2011-01-30 19:46:46 +07:00
|
|
|
return -ENOMEM;
|
|
|
|
|
2011-01-30 20:59:43 +07:00
|
|
|
if (evlist->pollfd == NULL && perf_evlist__alloc_pollfd(evlist) < 0)
|
2011-01-30 19:46:46 +07:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
evlist->overwrite = overwrite;
|
|
|
|
evlist->mmap_len = (pages + 1) * page_size;
|
|
|
|
|
|
|
|
list_for_each_entry(evsel, &evlist->entries, node) {
|
|
|
|
if ((evsel->attr.read_format & PERF_FORMAT_ID) &&
|
2011-03-10 21:15:54 +07:00
|
|
|
evsel->sample_id == NULL &&
|
2011-01-30 19:46:46 +07:00
|
|
|
perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0)
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
if (evlist->cpus->map[0] == -1)
|
|
|
|
return perf_evlist__mmap_per_thread(evlist, prot, mask);
|
2011-01-30 19:46:46 +07:00
|
|
|
|
perf evlist: Fix per thread mmap setup
The PERF_EVENT_IOC_SET_OUTPUT ioctl was returning -EINVAL when using
--pid when monitoring multithreaded apps, as we can only share a ring
buffer for events on the same thread if not doing per cpu.
Fix it by using per thread ring buffers.
Tested with:
[root@felicio ~]# tuna -t 26131 -CP | nl
1 thread ctxt_switches
2 pid SCHED_ rtpri affinity voluntary nonvoluntary cmd
3 26131 OTHER 0 0,1 10814276 2397830 chromium-browse
4 642 OTHER 0 0,1 14688 0 chromium-browse
5 26148 OTHER 0 0,1 713602 115479 chromium-browse
6 26149 OTHER 0 0,1 801958 2262 chromium-browse
7 26150 OTHER 0 0,1 1271128 248 chromium-browse
8 26151 OTHER 0 0,1 3 0 chromium-browse
9 27049 OTHER 0 0,1 36796 9 chromium-browse
10 618 OTHER 0 0,1 14711 0 chromium-browse
11 661 OTHER 0 0,1 14593 0 chromium-browse
12 29048 OTHER 0 0,1 28125 0 chromium-browse
13 26143 OTHER 0 0,1 2202789 781 chromium-browse
[root@felicio ~]#
So 11 threads under pid 26131, then:
[root@felicio ~]# perf record -F 50000 --pid 26131
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fa4a2538000-7fa4a25b9000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fa4a25b9000-7fa4a263a000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
3 7fa4a263a000-7fa4a26bb000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
4 7fa4a26bb000-7fa4a273c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
5 7fa4a273c000-7fa4a27bd000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
6 7fa4a27bd000-7fa4a283e000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
7 7fa4a283e000-7fa4a28bf000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
8 7fa4a28bf000-7fa4a2940000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
9 7fa4a2940000-7fa4a29c1000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
10 7fa4a29c1000-7fa4a2a42000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
11 7fa4a2a42000-7fa4a2ac3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
11 mmaps, one per thread since we didn't specify any CPU list, so we need one
mmap per thread and:
[root@felicio ~]# perf record -F 50000 --pid 26131
^M
^C[ perf record: Woken up 79 times to write data ]
[ perf record: Captured and wrote 20.614 MB perf.data (~900639 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 371310 26131
2 96516 26148
3 95694 26149
4 95203 26150
5 7291 26143
6 87 27049
7 76 661
8 60 29048
9 47 618
10 43 642
[root@felicio ~]#
Ok, one of the threads, 26151 was quiescent, so no samples there, but all the
others are there.
Then, if I specify one CPU:
[root@felicio ~]# perf record -F 50000 --pid 26131 --cpu 1
^C[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.680 MB perf.data (~29730 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 8444 26131
2 2584 26149
3 2518 26148
4 2324 26150
5 123 26143
6 9 661
7 9 29048
[root@felicio ~]#
This machine has two cores, so fewer threads appeared on the radar, and:
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f484b922000-7f484b9a3000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Just one mmap, as now we can use just one per-cpu buffer instead of the
per-thread needed in the previous case.
For global profiling:
[root@felicio ~]# perf record -F 50000 -a
^C[ perf record: Woken up 26 times to write data ]
[ perf record: Captured and wrote 7.128 MB perf.data (~311412 samples) ]
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7fb49b435000-7fb49b4b6000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
2 7fb49b4b6000-7fb49b537000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
It uses per-cpu buffers.
For just one thread:
[root@felicio ~]# perf record -F 50000 --tid 26148
^C[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.330 MB perf.data (~14426 samples) ]
[root@felicio ~]# perf report -D | grep PERF_RECORD_SAMPLE | cut -d/ -f2 | cut -d: -f1 | sort -n | uniq -c | sort -nr | nl
1 9969 26148
[root@felicio ~]#
[root@felicio ~]# grep perf_event /proc/`pidof perf`/maps | nl
1 7f286a51b000-7f286a59c000 rwxs 00000000 00:09 4064 anon_inode:[perf_event]
[root@felicio ~]#
Tested-by: David Ahern <dsahern@gmail.com>
Tested-by: Lin Ming <ming.m.lin@intel.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Link: http://lkml.kernel.org/r/20110426204401.GB1746@ghostprotocols.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2011-05-15 19:39:00 +07:00
|
|
|
return perf_evlist__mmap_per_cpu(evlist, prot, mask);
|
2011-01-30 19:46:46 +07:00
|
|
|
}
|
2011-01-30 20:59:43 +07:00
|
|
|
|
|
|
|
int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid,
|
|
|
|
pid_t target_tid, const char *cpu_list)
|
|
|
|
{
|
|
|
|
evlist->threads = thread_map__new(target_pid, target_tid);
|
|
|
|
|
|
|
|
if (evlist->threads == NULL)
|
|
|
|
return -1;
|
|
|
|
|
2011-04-26 02:25:20 +07:00
|
|
|
if (cpu_list == NULL && target_tid != -1)
|
2011-01-30 20:59:43 +07:00
|
|
|
evlist->cpus = cpu_map__dummy_new();
|
|
|
|
else
|
|
|
|
evlist->cpus = cpu_map__new(cpu_list);
|
|
|
|
|
|
|
|
if (evlist->cpus == NULL)
|
|
|
|
goto out_delete_threads;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
out_delete_threads:
|
|
|
|
thread_map__delete(evlist->threads);
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
void perf_evlist__delete_maps(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
cpu_map__delete(evlist->cpus);
|
|
|
|
thread_map__delete(evlist->threads);
|
|
|
|
evlist->cpus = NULL;
|
|
|
|
evlist->threads = NULL;
|
|
|
|
}
|
2011-02-26 10:51:54 +07:00
|
|
|
|
|
|
|
int perf_evlist__set_filters(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
const struct thread_map *threads = evlist->threads;
|
|
|
|
const struct cpu_map *cpus = evlist->cpus;
|
|
|
|
struct perf_evsel *evsel;
|
|
|
|
char *filter;
|
|
|
|
int thread;
|
|
|
|
int cpu;
|
|
|
|
int err;
|
|
|
|
int fd;
|
|
|
|
|
|
|
|
list_for_each_entry(evsel, &evlist->entries, node) {
|
|
|
|
filter = evsel->filter;
|
|
|
|
if (!filter)
|
|
|
|
continue;
|
|
|
|
for (cpu = 0; cpu < cpus->nr; cpu++) {
|
|
|
|
for (thread = 0; thread < threads->nr; thread++) {
|
|
|
|
fd = FD(evsel, cpu, thread);
|
|
|
|
err = ioctl(fd, PERF_EVENT_IOC_SET_FILTER, filter);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2011-05-21 22:49:00 +07:00
|
|
|
|
2011-06-02 21:04:54 +07:00
|
|
|
bool perf_evlist__valid_sample_type(const struct perf_evlist *evlist)
|
2011-05-21 22:49:00 +07:00
|
|
|
{
|
2011-06-02 21:04:54 +07:00
|
|
|
struct perf_evsel *pos, *first;
|
|
|
|
|
|
|
|
pos = first = list_entry(evlist->entries.next, struct perf_evsel, node);
|
|
|
|
|
|
|
|
list_for_each_entry_continue(pos, &evlist->entries, node) {
|
|
|
|
if (first->attr.sample_type != pos->attr.sample_type)
|
|
|
|
return false;
|
2011-05-21 22:49:00 +07:00
|
|
|
}
|
|
|
|
|
2011-06-02 21:04:54 +07:00
|
|
|
return true;
|
2011-05-21 22:49:00 +07:00
|
|
|
}
|
|
|
|
|
2011-06-02 21:04:54 +07:00
|
|
|
u64 perf_evlist__sample_type(const struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
struct perf_evsel *first;
|
|
|
|
|
|
|
|
first = list_entry(evlist->entries.next, struct perf_evsel, node);
|
|
|
|
return first->attr.sample_type;
|
|
|
|
}
|
|
|
|
|
2011-11-12 07:28:50 +07:00
|
|
|
u16 perf_evlist__id_hdr_size(const struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
struct perf_evsel *first;
|
|
|
|
struct perf_sample *data;
|
|
|
|
u64 sample_type;
|
|
|
|
u16 size = 0;
|
|
|
|
|
|
|
|
first = list_entry(evlist->entries.next, struct perf_evsel, node);
|
|
|
|
|
|
|
|
if (!first->attr.sample_id_all)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
sample_type = first->attr.sample_type;
|
|
|
|
|
|
|
|
if (sample_type & PERF_SAMPLE_TID)
|
|
|
|
size += sizeof(data->tid) * 2;
|
|
|
|
|
|
|
|
if (sample_type & PERF_SAMPLE_TIME)
|
|
|
|
size += sizeof(data->time);
|
|
|
|
|
|
|
|
if (sample_type & PERF_SAMPLE_ID)
|
|
|
|
size += sizeof(data->id);
|
|
|
|
|
|
|
|
if (sample_type & PERF_SAMPLE_STREAM_ID)
|
|
|
|
size += sizeof(data->stream_id);
|
|
|
|
|
|
|
|
if (sample_type & PERF_SAMPLE_CPU)
|
|
|
|
size += sizeof(data->cpu) * 2;
|
|
|
|
out:
|
|
|
|
return size;
|
|
|
|
}
|
|
|
|
|
2011-06-02 21:04:54 +07:00
|
|
|
bool perf_evlist__valid_sample_id_all(const struct perf_evlist *evlist)
|
2011-05-21 22:49:00 +07:00
|
|
|
{
|
2011-06-02 21:04:54 +07:00
|
|
|
struct perf_evsel *pos, *first;
|
|
|
|
|
|
|
|
pos = first = list_entry(evlist->entries.next, struct perf_evsel, node);
|
|
|
|
|
|
|
|
list_for_each_entry_continue(pos, &evlist->entries, node) {
|
|
|
|
if (first->attr.sample_id_all != pos->attr.sample_id_all)
|
|
|
|
return false;
|
2011-05-21 22:49:00 +07:00
|
|
|
}
|
|
|
|
|
2011-06-02 21:04:54 +07:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool perf_evlist__sample_id_all(const struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
struct perf_evsel *first;
|
|
|
|
|
|
|
|
first = list_entry(evlist->entries.next, struct perf_evsel, node);
|
|
|
|
return first->attr.sample_id_all;
|
2011-05-21 22:49:00 +07:00
|
|
|
}
|
2011-10-06 05:11:32 +07:00
|
|
|
|
|
|
|
void perf_evlist__set_selected(struct perf_evlist *evlist,
|
|
|
|
struct perf_evsel *evsel)
|
|
|
|
{
|
|
|
|
evlist->selected = evsel;
|
|
|
|
}
|
2011-10-25 19:42:19 +07:00
|
|
|
|
|
|
|
int perf_evlist__open(struct perf_evlist *evlist, bool group)
|
|
|
|
{
|
|
|
|
struct perf_evsel *evsel, *first;
|
|
|
|
int err, ncpus, nthreads;
|
|
|
|
|
|
|
|
first = list_entry(evlist->entries.next, struct perf_evsel, node);
|
|
|
|
|
|
|
|
list_for_each_entry(evsel, &evlist->entries, node) {
|
|
|
|
struct xyarray *group_fd = NULL;
|
|
|
|
|
|
|
|
if (group && evsel != first)
|
|
|
|
group_fd = first->fd;
|
|
|
|
|
|
|
|
err = perf_evsel__open(evsel, evlist->cpus, evlist->threads,
|
|
|
|
group, group_fd);
|
|
|
|
if (err < 0)
|
|
|
|
goto out_err;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
out_err:
|
|
|
|
ncpus = evlist->cpus ? evlist->cpus->nr : 1;
|
|
|
|
nthreads = evlist->threads ? evlist->threads->nr : 1;
|
|
|
|
|
|
|
|
list_for_each_entry_reverse(evsel, &evlist->entries, node)
|
|
|
|
perf_evsel__close(evsel, ncpus, nthreads);
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
2011-11-09 17:47:15 +07:00
|
|
|
|
|
|
|
int perf_evlist__prepare_workload(struct perf_evlist *evlist,
|
|
|
|
struct perf_record_opts *opts,
|
|
|
|
const char *argv[])
|
|
|
|
{
|
|
|
|
int child_ready_pipe[2], go_pipe[2];
|
|
|
|
char bf;
|
|
|
|
|
|
|
|
if (pipe(child_ready_pipe) < 0) {
|
|
|
|
perror("failed to create 'ready' pipe");
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (pipe(go_pipe) < 0) {
|
|
|
|
perror("failed to create 'go' pipe");
|
|
|
|
goto out_close_ready_pipe;
|
|
|
|
}
|
|
|
|
|
|
|
|
evlist->workload.pid = fork();
|
|
|
|
if (evlist->workload.pid < 0) {
|
|
|
|
perror("failed to fork");
|
|
|
|
goto out_close_pipes;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!evlist->workload.pid) {
|
|
|
|
if (opts->pipe_output)
|
|
|
|
dup2(2, 1);
|
|
|
|
|
|
|
|
close(child_ready_pipe[0]);
|
|
|
|
close(go_pipe[1]);
|
|
|
|
fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Do a dummy execvp to get the PLT entry resolved,
|
|
|
|
* so we avoid the resolver overhead on the real
|
|
|
|
* execvp call.
|
|
|
|
*/
|
|
|
|
execvp("", (char **)argv);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Tell the parent we're ready to go
|
|
|
|
*/
|
|
|
|
close(child_ready_pipe[1]);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Wait until the parent tells us to go.
|
|
|
|
*/
|
|
|
|
if (read(go_pipe[0], &bf, 1) == -1)
|
|
|
|
perror("unable to read pipe");
|
|
|
|
|
|
|
|
execvp(argv[0], (char **)argv);
|
|
|
|
|
|
|
|
perror(argv[0]);
|
|
|
|
kill(getppid(), SIGUSR1);
|
|
|
|
exit(-1);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!opts->system_wide && opts->target_tid == -1 && opts->target_pid == -1)
|
|
|
|
evlist->threads->map[0] = evlist->workload.pid;
|
|
|
|
|
|
|
|
close(child_ready_pipe[1]);
|
|
|
|
close(go_pipe[0]);
|
|
|
|
/*
|
|
|
|
* wait for child to settle
|
|
|
|
*/
|
|
|
|
if (read(child_ready_pipe[0], &bf, 1) == -1) {
|
|
|
|
perror("unable to read pipe");
|
|
|
|
goto out_close_pipes;
|
|
|
|
}
|
|
|
|
|
|
|
|
evlist->workload.cork_fd = go_pipe[1];
|
|
|
|
close(child_ready_pipe[0]);
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
out_close_pipes:
|
|
|
|
close(go_pipe[0]);
|
|
|
|
close(go_pipe[1]);
|
|
|
|
out_close_ready_pipe:
|
|
|
|
close(child_ready_pipe[0]);
|
|
|
|
close(child_ready_pipe[1]);
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int perf_evlist__start_workload(struct perf_evlist *evlist)
|
|
|
|
{
|
|
|
|
if (evlist->workload.cork_fd > 0) {
|
|
|
|
/*
|
|
|
|
* Remove the cork, let it rip!
|
|
|
|
*/
|
|
|
|
return close(evlist->workload.cork_fd);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|