[PATCH 2/8] perf, tool: Enable grouping logic for parsed events

From: Jiri Olsa
Date: Wed Apr 04 2012 - 17:18:25 EST

The current event grouping is basic. If you specify the '--group'
option for record/stat/top command, all the specified events become
members of a single group with the first event as a group leader.

This patch adds a functionality that allows to create event groups
based on the way they are specified on the command line. The 'group='
syntax introduced in earlier patch is used. The current '--group/-g'
option behaviour remains intact.

It is possible to create a group with following:
perf record -e group=cycles,faults ls

creating single event group containing 'cycles' and 'faults' events,
and with cycles event as a leader.

All groups are created with regards to threads and cpus. Thus
recording an event group within a 2 threads on server with
4 CPUs will create 8 separate groups.

Examples (first event in brackets is group leader):

# 1 group (cpu-clock,task-clock)
perf record --group -e cpu-clock,task-clock ls
perf record -e group=cpu-clock,task-clock ls

# 2 groups (cpu-clock,task-clock) (minor-faults,major-faults)
perf record -e "group=cpu-clock,task-clock;group=minor-faults,major-faults" ls

# 1 group (cpu-clock,task-clock,minor-faults,major-faults)
perf record --group -e cpu-clock,task-clock \
-e minor-faults,major-faults ls

# 2 groups (cpu-clock,task-clock) (minor-faults,major-faults)
perf record -v -e group=cpu-clock,task-clock \
-e group=minor-faults,major-faults -e instructions ls

# 1 group (cpu-clock,task-clock,minor-faults,major-faults,instructions)
perf record --group -e cpu-clock,task-clock \
-e minor-faults,major-faults -e instructions ls

Updated automated test to check on group_leader settings.

Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
---
tools/perf/builtin-record.c | 13 +++----
tools/perf/builtin-stat.c | 13 +++----
tools/perf/builtin-test.c | 8 ++--
tools/perf/builtin-top.c | 12 ++----
tools/perf/util/evlist.c | 34 +++++++++++++------
tools/perf/util/evlist.h | 5 ++-
tools/perf/util/evsel.c | 46 ++++++++++++++++----------
tools/perf/util/evsel.h | 10 ++---
tools/perf/util/parse-events-test.c | 61 +++++++++++++++++++++++++++++++++-
tools/perf/util/parse-events.c | 3 +-
tools/perf/util/python.c | 7 +++-
11 files changed, 144 insertions(+), 68 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 87fc68d..3e0ef7d 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -177,18 +177,18 @@ static bool perf_evlist__equal(struct perf_evlist *evlist,

static void perf_record__open(struct perf_record *rec)
{
- struct perf_evsel *pos, *first;
+ struct perf_evsel *pos;
struct perf_evlist *evlist = rec->evlist;
struct perf_session *session = rec->session;
struct perf_record_opts *opts = &rec->opts;

- first = list_entry(evlist->entries.next, struct perf_evsel, node);
-
perf_evlist__config_attrs(evlist, opts);

+ if (opts->group)
+ perf_evlist__group(evlist);
+
list_for_each_entry(pos, &evlist->entries, node) {
struct perf_event_attr *attr = &pos->attr;
- struct xyarray *group_fd = NULL;
/*
* Check if parse_single_tracepoint_event has already asked for
* PERF_SAMPLE_TIME.
@@ -203,16 +203,13 @@ static void perf_record__open(struct perf_record *rec)
*/
bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;

- if (opts->group && pos != first)
- group_fd = first->fd;
fallback_missing_features:
if (opts->exclude_guest_missing)
attr->exclude_guest = attr->exclude_host = 0;
retry_sample_id:
attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1;
try_again:
- if (perf_evsel__open(pos, evlist->cpus, evlist->threads,
- opts->group, group_fd) < 0) {
+ if (perf_evsel__open(pos, evlist->cpus, evlist->threads) < 0) {
int err = errno;

if (err == EPERM || err == EACCES) {
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index c941bb6..9dffc8f 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -282,10 +282,6 @@ static int create_perf_stat_counter(struct perf_evsel *evsel,
struct perf_evsel *first)
{
struct perf_event_attr *attr = &evsel->attr;
- struct xyarray *group_fd = NULL;
-
- if (group && evsel != first)
- group_fd = first->fd;

if (scale)
attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
@@ -294,15 +290,13 @@ static int create_perf_stat_counter(struct perf_evsel *evsel,
attr->inherit = !no_inherit;

if (system_wide)
- return perf_evsel__open_per_cpu(evsel, evsel_list->cpus,
- group, group_fd);
+ return perf_evsel__open_per_cpu(evsel, evsel_list->cpus);
if (!target_pid && !target_tid && (!group || evsel == first)) {
attr->disabled = 1;
attr->enable_on_exec = 1;
}

- return perf_evsel__open_per_thread(evsel, evsel_list->threads,
- group, group_fd);
+ return perf_evsel__open_per_thread(evsel, evsel_list->threads);
}

/*
@@ -459,6 +453,9 @@ static int run_perf_stat(int argc __used, const char **argv)
close(child_ready_pipe[0]);
}

+ if (group)
+ perf_evlist__group(evsel_list);
+
first = list_entry(evsel_list->entries.next, struct perf_evsel, node);

list_for_each_entry(counter, &evsel_list->entries, node) {
diff --git a/tools/perf/builtin-test.c b/tools/perf/builtin-test.c
index 7c61eef..c19d2d7 100644
--- a/tools/perf/builtin-test.c
+++ b/tools/perf/builtin-test.c
@@ -294,7 +294,7 @@ static int test__open_syscall_event(void)
goto out_thread_map_delete;
}

- if (perf_evsel__open_per_thread(evsel, threads, false, NULL) < 0) {
+ if (perf_evsel__open_per_thread(evsel, threads) < 0) {
pr_debug("failed to open counter: %s, "
"tweak /proc/sys/kernel/perf_event_paranoid?\n",
strerror(errno));
@@ -369,7 +369,7 @@ static int test__open_syscall_event_on_all_cpus(void)
goto out_thread_map_delete;
}

- if (perf_evsel__open(evsel, cpus, threads, false, NULL) < 0) {
+ if (perf_evsel__open(evsel, cpus, threads) < 0) {
pr_debug("failed to open counter: %s, "
"tweak /proc/sys/kernel/perf_event_paranoid?\n",
strerror(errno));
@@ -534,7 +534,7 @@ static int test__basic_mmap(void)

perf_evlist__add(evlist, evsels[i]);

- if (perf_evsel__open(evsels[i], cpus, threads, false, NULL) < 0) {
+ if (perf_evsel__open(evsels[i], cpus, threads) < 0) {
pr_debug("failed to open counter: %s, "
"tweak /proc/sys/kernel/perf_event_paranoid?\n",
strerror(errno));
@@ -736,7 +736,7 @@ static int test__PERF_RECORD(void)
* Call sys_perf_event_open on all the fds on all the evsels,
* grouping them if asked to.
*/
- err = perf_evlist__open(evlist, opts.group);
+ err = perf_evlist__open(evlist);
if (err < 0) {
pr_debug("perf_evlist__open: %s\n", strerror(errno));
goto out_delete_evlist;
diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c
index e3c63ae..38944a2 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -841,17 +841,14 @@ static void perf_top__mmap_read(struct perf_top *top)

static void perf_top__start_counters(struct perf_top *top)
{
- struct perf_evsel *counter, *first;
+ struct perf_evsel *counter;
struct perf_evlist *evlist = top->evlist;

- first = list_entry(evlist->entries.next, struct perf_evsel, node);
+ if (top->group)
+ perf_evlist__group(evlist);

list_for_each_entry(counter, &evlist->entries, node) {
struct perf_event_attr *attr = &counter->attr;
- struct xyarray *group_fd = NULL;
-
- if (top->group && counter != first)
- group_fd = first->fd;

attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;

@@ -879,8 +876,7 @@ retry_sample_id:
attr->sample_id_all = top->sample_id_all_missing ? 0 : 1;
try_again:
if (perf_evsel__open(counter, top->evlist->cpus,
- top->evlist->threads, top->group,
- group_fd) < 0) {
+ top->evlist->threads) < 0) {
int err = errno;

if (err == EPERM || err == EACCES) {
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 1986d80..486f939 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -107,6 +107,26 @@ void perf_evlist__splice_list_tail(struct perf_evlist *evlist,
evlist->nr_entries += nr_entries;
}

+void perf_evlist__group_list(struct list_head *list)
+{
+ struct perf_evsel *evsel, *first;
+
+ first = list_entry(list->next, struct perf_evsel, node);
+
+ list_for_each_entry(evsel, list, node)
+ if (evsel != first)
+ evsel->leader = first;
+}
+
+int perf_evlist__group(struct perf_evlist *evlist)
+{
+ if (!evlist->nr_entries)
+ return -EINVAL;
+
+ perf_evlist__group_list(&evlist->entries);
+ return 0;
+}
+
int perf_evlist__add_default(struct perf_evlist *evlist)
{
struct perf_event_attr attr = {
@@ -740,21 +760,13 @@ void perf_evlist__set_selected(struct perf_evlist *evlist,
evlist->selected = evsel;
}

-int perf_evlist__open(struct perf_evlist *evlist, bool group)
+int perf_evlist__open(struct perf_evlist *evlist)
{
- struct perf_evsel *evsel, *first;
+ struct perf_evsel *evsel;
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);
+ err = perf_evsel__open(evsel, evlist->cpus, evlist->threads);
if (err < 0)
goto out_err;
}
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 21f1c9e..863789a 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -78,7 +78,7 @@ struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id);

union perf_event *perf_evlist__mmap_read(struct perf_evlist *self, int idx);

-int perf_evlist__open(struct perf_evlist *evlist, bool group);
+int perf_evlist__open(struct perf_evlist *evlist);

void perf_evlist__config_attrs(struct perf_evlist *evlist,
struct perf_record_opts *opts);
@@ -122,4 +122,7 @@ void perf_evlist__splice_list_tail(struct perf_evlist *evlist,
struct list_head *list,
int nr_entries);

+void perf_evlist__group_list(struct list_head *list);
+int perf_evlist__group(struct perf_evlist *evlist);
+
#endif /* __PERF_EVLIST_H */
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 8c13dbc..445ba60 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -16,7 +16,6 @@
#include "thread_map.h"

#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
-#define GROUP_FD(group_fd, cpu) (*(int *)xyarray__entry(group_fd, cpu, 0))

int __perf_evsel__sample_size(u64 sample_type)
{
@@ -293,9 +292,24 @@ int __perf_evsel__read(struct perf_evsel *evsel,
return 0;
}

+static int get_group_fd(struct perf_evsel *evsel, int cpu, int thread)
+{
+ struct perf_evsel *leader = evsel->leader;
+
+ if (!leader)
+ return -1;
+
+ /*
+ * Leader must be already processed/open,
+ * if not it's a bug.
+ */
+ assert(leader->fd);
+
+ return FD(leader, cpu, thread);
+}
+
static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
- struct thread_map *threads, bool group,
- struct xyarray *group_fds)
+ struct thread_map *threads)
{
int cpu, thread;
unsigned long flags = 0;
@@ -311,13 +325,15 @@ static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
}

for (cpu = 0; cpu < cpus->nr; cpu++) {
- int group_fd = group_fds ? GROUP_FD(group_fds, cpu) : -1;

for (thread = 0; thread < threads->nr; thread++) {
+ int group_fd;

if (!evsel->cgrp)
pid = threads->map[thread];

+ group_fd = get_group_fd(evsel, cpu, thread);
+
FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
pid,
cpus->map[cpu],
@@ -327,8 +343,9 @@ static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
goto out_close;
}

- if (group && group_fd == -1)
- group_fd = FD(evsel, cpu, thread);
+ pr_debug("event cpu %d, thread %d, fd %d, group %d\n",
+ cpu, pid, FD(evsel, cpu, thread),
+ group_fd);
}
}

@@ -372,8 +389,7 @@ static struct {
};

int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
- struct thread_map *threads, bool group,
- struct xyarray *group_fd)
+ struct thread_map *threads)
{
if (cpus == NULL) {
/* Work around old compiler warnings about strict aliasing */
@@ -383,23 +399,19 @@ int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
if (threads == NULL)
threads = &empty_thread_map.map;

- return __perf_evsel__open(evsel, cpus, threads, group, group_fd);
+ return __perf_evsel__open(evsel, cpus, threads);
}

int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
- struct cpu_map *cpus, bool group,
- struct xyarray *group_fd)
+ struct cpu_map *cpus)
{
- return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group,
- group_fd);
+ return __perf_evsel__open(evsel, cpus, &empty_thread_map.map);
}

int perf_evsel__open_per_thread(struct perf_evsel *evsel,
- struct thread_map *threads, bool group,
- struct xyarray *group_fd)
+ struct thread_map *threads)
{
- return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group,
- group_fd);
+ return __perf_evsel__open(evsel, &empty_cpu_map.map, threads);
}

static int perf_event__parse_id_sample(const union perf_event *event, u64 type,
diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h
index 3d6b3e4..e749d5c 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -66,6 +66,7 @@ struct perf_evsel {
void *data;
} handler;
bool supported;
+ struct perf_evsel *leader;
};

struct cpu_map;
@@ -91,14 +92,11 @@ void perf_evsel__free_id(struct perf_evsel *evsel);
void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads);

int perf_evsel__open_per_cpu(struct perf_evsel *evsel,
- struct cpu_map *cpus, bool group,
- struct xyarray *group_fds);
+ struct cpu_map *cpus);
int perf_evsel__open_per_thread(struct perf_evsel *evsel,
- struct thread_map *threads, bool group,
- struct xyarray *group_fds);
+ struct thread_map *threads);
int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
- struct thread_map *threads, bool group,
- struct xyarray *group_fds);
+ struct thread_map *threads);
void perf_evsel__close(struct perf_evsel *evsel, int ncpus, int nthreads);

#define perf_evsel__match(evsel, t, c) \
diff --git a/tools/perf/util/parse-events-test.c b/tools/perf/util/parse-events-test.c
index 73ff0af..baa4caf 100644
--- a/tools/perf/util/parse-events-test.c
+++ b/tools/perf/util/parse-events-test.c
@@ -388,8 +388,7 @@ static int test__checkevent_list(struct perf_evlist *evlist)

static int test__checkevent_pmu_name(struct perf_evlist *evlist)
{
- struct perf_evsel *evsel = list_entry(evlist->entries.next,
- struct perf_evsel, node);
+ struct perf_evsel *evsel;

/* cpu/config=1,name=krava1/u */
evsel = list_entry(evlist->entries.next, struct perf_evsel, node);
@@ -431,6 +430,60 @@ static int test__checkevent_pmu_branch_type(struct perf_evlist *evlist)
return 0;
}

+static int test__checkevent_group(struct perf_evlist *evlist)
+{
+ struct perf_evsel *evsel, *leader;
+
+ TEST_ASSERT_VAL("wrong number of entries", 6 == evlist->nr_entries);
+
+ /* cycles */
+ evsel = leader = list_entry(evlist->entries.next, struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong config",
+ PERF_COUNT_HW_CPU_CYCLES == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ /* faults */
+ evsel = list_entry(evsel->node.next, struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_SOFTWARE == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong config",
+ PERF_COUNT_SW_PAGE_FAULTS == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ /* r1 */
+ evsel = list_entry(evsel->node.next, struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong config", 1 == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ /* mem:0 */
+ evsel = list_entry(evsel->node.next, struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_BREAKPOINT == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong bp_type", (HW_BREAKPOINT_R | HW_BREAKPOINT_W) ==
+ evsel->attr.bp_type);
+ TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_4 ==
+ evsel->attr.bp_len);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ /* cpu/config=2,period=1000/u */
+ evsel = leader = list_entry(evsel->node.next, struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_RAW == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong config", 2 == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong sample_period",
+ 1000 == evsel->attr.sample_period);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ /* instructions:h" */
+ evsel = list_entry(evsel->node.next, struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_HARDWARE == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong config",
+ PERF_COUNT_HW_INSTRUCTIONS == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ return 0;
+}
+
static struct test__event_st {
const char *name;
__u32 type;
@@ -544,6 +597,10 @@ static struct test__event_st {
.name = "cpu/config=1,branch_type=hv|any_ret|1,config2=2/u",
.check = test__checkevent_pmu_branch_type,
},
+ [27] = {
+ .name = "group=cycles,faults;r1,mem:0:u;group=cpu/config=2,period=1000/u,instructions:h",
+ .check = test__checkevent_group,
+ },
};

#define TEST__EVENTS_CNT (sizeof(test__events) / sizeof(struct test__event_st))
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 53c7505..5110916 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -791,8 +791,9 @@ int parse_events_add_pmu(struct list_head **list, int *idx,
return add_event(list, idx, &attr, name);
}

-int parse_events__group(struct list_head *list __used)
+int parse_events__group(struct list_head *list)
{
+ perf_evlist__group_list(list);
return 0;
}

diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c
index e03b58a..419c29e 100644
--- a/tools/perf/util/python.c
+++ b/tools/perf/util/python.c
@@ -627,7 +627,7 @@ static PyObject *pyrf_evsel__open(struct pyrf_evsel *pevsel,
* This will group just the fds for this single evsel, to group
* multiple events, use evlist.open().
*/
- if (perf_evsel__open(evsel, cpus, threads, group, NULL) < 0) {
+ if (perf_evsel__open(evsel, cpus, threads) < 0) {
PyErr_SetFromErrno(PyExc_OSError);
return NULL;
}
@@ -828,7 +828,10 @@ static PyObject *pyrf_evlist__open(struct pyrf_evlist *pevlist,
if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|OOii", kwlist, &group))
return NULL;

- if (perf_evlist__open(evlist, group) < 0) {
+ if (group)
+ perf_evlist__group(evlist);
+
+ if (perf_evlist__open(evlist) < 0) {
PyErr_SetFromErrno(PyExc_OSError);
return NULL;
}
--
1.7.7.6

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/