hi,
adding support for creating event groups based on the way they
are specified on the command line.
This patchset adds the '{}' style grammar to express event group,
allowing so far only the 'event modifier' as group modifier.
The sample/leader modifier changes are ready, but I'm getting
some weird numbers in tests, so I'll send it later. Meanwhile,
it'd be nice to have at least this patchset in.. ;)
Attached patches:
1/3 perf, tool: Add support to parse event group syntax
2/3 perf, tool: Enable grouping logic for parsed events
3/3 perf, test: Add automated tests for event group parsing
wbr,
jirka
---
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 | 20 ++---
tools/perf/util/evlist.h | 3 +-
tools/perf/util/evsel.c | 51 ++++++++----
tools/perf/util/evsel.h | 11 +--
tools/perf/util/parse-events-test.c | 152 +++++++++++++++++++++++++++++++++++
tools/perf/util/parse-events.c | 32 +++++++-
tools/perf/util/parse-events.h | 5 +-
tools/perf/util/parse-events.l | 2 +
tools/perf/util/parse-events.y | 93 ++++++++++++++++++---
tools/perf/util/python.c | 7 +-
14 files changed, 341 insertions(+), 81 deletions(-)
Adding 3 more tests for new event group syntax. Tests are executed
within the 'perf test parse' test suite.
Signed-off-by: Jiri Olsa <[email protected]>
---
tools/perf/util/parse-events-test.c | 152 +++++++++++++++++++++++++++++++++++
1 files changed, 152 insertions(+), 0 deletions(-)
diff --git a/tools/perf/util/parse-events-test.c b/tools/perf/util/parse-events-test.c
index 229af6d..66e6cc1 100644
--- a/tools/perf/util/parse-events-test.c
+++ b/tools/perf/util/parse-events-test.c
@@ -473,6 +473,146 @@ static int test__checkterms_simple(struct list_head *terms)
return 0;
}
+static int test__group1(struct perf_evlist *evlist)
+{
+ struct perf_evsel *evsel, *leader;
+
+ TEST_ASSERT_VAL("wrong number of entries", 2 == evlist->nr_entries);
+
+ /* instructions:k */
+ 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_INSTRUCTIONS == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
+ TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+ TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ /* cycles:upp */
+ 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_CPU_CYCLES == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
+ TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+ TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip == 2);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ return 0;
+}
+
+static int test__group2(struct perf_evlist *evlist)
+{
+ struct perf_evsel *evsel, *leader;
+
+ TEST_ASSERT_VAL("wrong number of entries", 3 == evlist->nr_entries);
+
+ /* faults + :u modifier */
+ evsel = leader = list_entry(evlist->entries.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 exclude_user", !evsel->attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
+ TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+ TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ /* cache-references + :u modifier */
+ 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_CACHE_REFERENCES == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
+ TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+ TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ /* cycles:k */
+ 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_CPU_CYCLES == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
+ TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+ TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ return 0;
+}
+
+static int test__group3(struct perf_evlist *evlist __used)
+{
+ struct perf_evsel *evsel, *leader;
+
+ TEST_ASSERT_VAL("wrong number of entries", 5 == evlist->nr_entries);
+
+ /* group1 syscalls:sys_enter_open:H */
+ evsel = leader = list_entry(evlist->entries.next,
+ struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", PERF_TYPE_TRACEPOINT == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong sample_type",
+ (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | PERF_SAMPLE_CPU) ==
+ evsel->attr.sample_type);
+ TEST_ASSERT_VAL("wrong sample_period", 1 == evsel->attr.sample_period);
+ TEST_ASSERT_VAL("wrong exclude guest", evsel->attr.exclude_guest);
+ TEST_ASSERT_VAL("wrong exclude host", !evsel->attr.exclude_host);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+ TEST_ASSERT_VAL("wrong group name",
+ !strcmp(leader->group_name, "group1"));
+
+ /* group1 cycles:kppp */
+ 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_CPU_CYCLES == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong exclude_user", evsel->attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", !evsel->attr.exclude_kernel);
+ TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+ TEST_ASSERT_VAL("wrong precise_ip", evsel->attr.precise_ip == 3);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+ TEST_ASSERT_VAL("wrong group name", !evsel->group_name);
+
+ /* group2 cycles + G modifier */
+ evsel = leader = 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_CPU_CYCLES == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong exclude guest", !evsel->attr.exclude_guest);
+ TEST_ASSERT_VAL("wrong exclude host", evsel->attr.exclude_host);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+ TEST_ASSERT_VAL("wrong group name",
+ !strcmp(leader->group_name, "group2"));
+
+ /* group2 1:3 + G modifier */
+ evsel = list_entry(evsel->node.next, struct perf_evsel, node);
+ TEST_ASSERT_VAL("wrong type", 1 == evsel->attr.type);
+ TEST_ASSERT_VAL("wrong config", 3 == evsel->attr.config);
+ TEST_ASSERT_VAL("wrong exclude guest", !evsel->attr.exclude_guest);
+ TEST_ASSERT_VAL("wrong exclude host", evsel->attr.exclude_host);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == leader);
+
+ /* instructions:u */
+ 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 exclude_user", !evsel->attr.exclude_user);
+ TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
+ TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+ TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
+ TEST_ASSERT_VAL("wrong leader", evsel->leader == NULL);
+
+ return 0;
+}
+
struct test__event_st {
const char *name;
__u32 type;
@@ -584,6 +724,18 @@ static struct test__event_st test__events[] = {
.name = "instructions:H",
.check = test__checkevent_exclude_guest_modifier,
},
+ [26] = {
+ .name = "{instructions:k,cycles:upp}",
+ .check = test__group1,
+ },
+ [27] = {
+ .name = "{faults:k,cache-references}:u,cycles:k",
+ .check = test__group2,
+ },
+ [28] = {
+ .name = "group1{syscalls:sys_enter_open:H,cycles:kppp},group2{cycles,1:3}:G,instructions:u",
+ .check = test__group3,
+ },
};
#define TEST__EVENTS_CNT (sizeof(test__events) / sizeof(struct test__event_st))
--
1.7.7.6
Adding scanner/parser bits to parse event groups.
The grammar for group is:
groups: groups ',' group | group
group: group_name '{' events '}' group_mod
group_name: name | empty
group_mod: ':' group_mods | empty
group_mods: event_mod
Currently it's possible to use standard event modifier as a
modifier for group. It spans over all events in the group
and overrides any event modifier settings.
It's necessary to use quoting ("'\) when specifying group on
command line, since {} characters are interpretted by most of
the shells.
It is now possible to specify groups in event syntax like:
'{cycles,faults}'
- anonymous group
'group1{cycles,faults}
- group with name 'group1'
'{cycles,faults}:k
- anonymous group with event modifier 'k'
'{cpu-clock,task-clock},{minor-faults,major-faults}'
- two anonymous groups
The grouping functionality itself is comming shortly.
Signed-off-by: Jiri Olsa <[email protected]>
---
tools/perf/util/parse-events.c | 14 +++++-
tools/perf/util/parse-events.h | 4 +-
tools/perf/util/parse-events.l | 2 +
tools/perf/util/parse-events.y | 93 ++++++++++++++++++++++++++++++++++------
4 files changed, 97 insertions(+), 16 deletions(-)
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 0cc27da..f62cbee 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -547,19 +547,29 @@ int parse_events_add_pmu(struct list_head **list, int *idx,
pmu_event_name(head_config));
}
+int parse_events__modifier_group(struct list_head *list __used,
+ char *event_mod __used)
+{
+ return 0;
+}
+
+void parse_events__group(char *name __used, struct list_head *list __used)
+{
+}
+
void parse_events_update_lists(struct list_head *list_event,
struct list_head *list_all)
{
/*
* Called for single event definition. Update the
- * 'all event' list, and reinit the 'signle event'
+ * 'all event' list, and reinit the 'single event'
* list, for next event definition.
*/
list_splice_tail(list_event, list_all);
free(list_event);
}
-int parse_events_modifier(struct list_head *list, char *str)
+int parse_events__modifier_event(struct list_head *list, char *str)
{
struct perf_evsel *evsel;
int exclude = 0, exclude_GH = 0;
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index ee9c218..c2f2ed9 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -79,7 +79,8 @@ int parse_events__term_str(struct parse_events__term **_term,
int parse_events__term_clone(struct parse_events__term **new,
struct parse_events__term *term);
void parse_events__free_terms(struct list_head *terms);
-int parse_events_modifier(struct list_head *list, char *str);
+int parse_events__modifier_event(struct list_head *list, char *str);
+int parse_events__modifier_group(struct list_head *list, char *event_mod);
int parse_events_add_tracepoint(struct list_head **list, int *idx,
char *sys, char *event);
int parse_events_add_numeric(struct list_head **list, int *idx,
@@ -91,6 +92,7 @@ int parse_events_add_breakpoint(struct list_head **list, int *idx,
void *ptr, char *type);
int parse_events_add_pmu(struct list_head **list, int *idx,
char *pmu , struct list_head *head_config);
+void parse_events__group(char *name, struct list_head *list);
void parse_events_update_lists(struct list_head *list_event,
struct list_head *list_all);
void parse_events_error(void *data, void *scanner, char const *msg);
diff --git a/tools/perf/util/parse-events.l b/tools/perf/util/parse-events.l
index 488362e..931bb6e 100644
--- a/tools/perf/util/parse-events.l
+++ b/tools/perf/util/parse-events.l
@@ -151,6 +151,8 @@ r{num_raw_hex} { return raw(yyscanner); }
- { return '-'; }
, { return ','; }
: { return ':'; }
+"{" { return '{'; }
+"}" { return '}'; }
= { return '='; }
<mem>{
diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
index 9525c45..2141c0c 100644
--- a/tools/perf/util/parse-events.y
+++ b/tools/perf/util/parse-events.y
@@ -30,7 +30,7 @@ do { \
%token PE_NAME
%token PE_MODIFIER_EVENT PE_MODIFIER_BP
%token PE_NAME_CACHE_TYPE PE_NAME_CACHE_OP_RESULT
-%token PE_PREFIX_MEM PE_PREFIX_RAW
+%token PE_PREFIX_MEM PE_PREFIX_RAW PE_PREFIX_GROUP
%token PE_ERROR
%type <num> PE_VALUE
%type <num> PE_VALUE_SYM
@@ -51,6 +51,11 @@ do { \
%type <head> event_legacy_numeric
%type <head> event_legacy_raw
%type <head> event_def
+%type <head> event
+%type <head> events
+%type <head> group_def
+%type <head> group
+%type <head> groups
%union
{
@@ -62,33 +67,95 @@ do { \
%%
start:
-PE_START_EVENTS events
+PE_START_EVENTS start_events
|
-PE_START_TERMS terms
+PE_START_TERMS start_terms
+
+start_events: groups
+{
+ struct parse_events_data__events *data = _data;
+
+ parse_events_update_lists($1, &data->list);
+}
+
+groups:
+groups ',' group
+{
+ struct list_head *list = $1;
+ struct list_head *group = $3;
+
+ parse_events_update_lists(group, list);
+ $$ = list;
+}
+|
+groups ',' event
+{
+ struct list_head *list = $1;
+ struct list_head *event = $3;
+
+ parse_events_update_lists(event, list);
+ $$ = list;
+}
+|
+group
+|
+event
+
+group:
+group_def ':' PE_MODIFIER_EVENT
+{
+ struct list_head *list = $1;
+
+ ABORT_ON(parse_events__modifier_group(list, $3));
+ $$ = list;
+}
+|
+group_def
+
+group_def:
+PE_NAME '{' events '}'
+{
+ struct list_head *list = $3;
+
+ parse_events__group($1, list);
+ $$ = list;
+}
+|
+'{' events '}'
+{
+ struct list_head *list = $2;
+
+ parse_events__group(NULL, list);
+ $$ = list;
+}
events:
-events ',' event | event
+events ',' event
+{
+ struct list_head *event = $3;
+ struct list_head *list = $1;
+
+ parse_events_update_lists(event, list);
+ $$ = list;
+}
+|
+event
event:
event_def PE_MODIFIER_EVENT
{
- struct parse_events_data__events *data = _data;
+ struct list_head *list = $1;
/*
* Apply modifier on all events added by single event definition
* (there could be more events added for multiple tracepoint
* definitions via '*?'.
*/
- ABORT_ON(parse_events_modifier($1, $2));
- parse_events_update_lists($1, &data->list);
+ ABORT_ON(parse_events__modifier_event(list, $2));
+ $$ = list;
}
|
event_def
-{
- struct parse_events_data__events *data = _data;
-
- parse_events_update_lists($1, &data->list);
-}
event_def: event_pmu |
event_legacy_symbol |
@@ -215,7 +282,7 @@ PE_RAW
$$ = list;
}
-terms: event_config
+start_terms: event_config
{
struct parse_events_data__terms *data = _data;
data->terms = $1;
--
1.7.7.6
This patch adds a functionality that allows to create event groups
based on the way they are specified on the command line. Adding
functionality to the '{}' group syntax introduced in earlier patch.
The current '--group/-g' option behaviour remains intact. If you
specify it for record/stat/top command, all the specified events
become members of a single group with the first event as a group
leader.
With the new '{}' group syntax you can create group like:
# perf record -e '{cycles,faults}' ls
resulting in single event group containing 'cycles' and 'faults'
events, with cycles event as group 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 '{cpu-clock,task-clock}' ls
# 2 groups (cpu-clock,task-clock) (minor-faults,major-faults)
perf record -e '{cpu-clock,task-clock},{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
perf record -e '{cpu-clock,task-clock,minor-faults,major-faults}' ls
# 2 groups (cpu-clock,task-clock) (minor-faults,major-faults)
perf record -e '{cpu-clock,task-clock} -e '{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
perf record -e '{cpu-clock,task-clock,minor-faults,major-faults,instructions}' ls
It's possible to use standard event modifier for a group, which spans
over all events in the group and overrides any event modifier settings,
for example:
# perf record -r '{faults:k,cache-references}:u'
resulting in ':u' modifier being used for both 'faults' and 'cache-references'
events, regardless of their modifier setup (':k' for faults event).
Signed-off-by: Jiri Olsa <[email protected]>
---
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 | 20 +++++++--------
tools/perf/util/evlist.h | 3 +-
tools/perf/util/evsel.c | 51 ++++++++++++++++++++++++++-------------
tools/perf/util/evsel.h | 11 ++++----
tools/perf/util/parse-events.c | 26 +++++++++++++++++---
tools/perf/util/parse-events.h | 1 +
tools/perf/util/python.c | 7 ++++-
11 files changed, 96 insertions(+), 69 deletions(-)
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index f5a6452..94d6b12 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -176,18 +176,18 @@ 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.
@@ -202,16 +202,13 @@ 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 861f0ae..23908a8 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -281,13 +281,9 @@ 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;
bool exclude_guest_missing = false;
int ret;
- if (group && evsel != first)
- group_fd = first->fd;
-
if (scale)
attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
PERF_FORMAT_TOTAL_TIME_RUNNING;
@@ -299,8 +295,7 @@ retry:
evsel->attr.exclude_guest = evsel->attr.exclude_host = 0;
if (perf_target__has_cpu(&target)) {
- ret = perf_evsel__open_per_cpu(evsel, evsel_list->cpus,
- group, group_fd);
+ ret = perf_evsel__open_per_cpu(evsel, evsel_list->cpus);
if (ret)
goto check_ret;
return 0;
@@ -311,8 +306,7 @@ retry:
attr->enable_on_exec = 1;
}
- ret = perf_evsel__open_per_thread(evsel, evsel_list->threads,
- group, group_fd);
+ ret = perf_evsel__open_per_thread(evsel, evsel_list->threads);
if (!ret)
return 0;
/* fall through */
@@ -483,6 +477,9 @@ 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 5ce3030..03568b8 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));
@@ -739,7 +739,7 @@ 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 e3cab5f..c5cc2ab 100644
--- a/tools/perf/builtin-top.c
+++ b/tools/perf/builtin-top.c
@@ -875,17 +875,14 @@ 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;
@@ -916,8 +913,7 @@ 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 7400fb3..5fa9257 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -108,6 +108,12 @@ void perf_evlist__splice_list_tail(struct perf_evlist *evlist,
evlist->nr_entries += nr_entries;
}
+void perf_evlist__group(struct perf_evlist *evlist)
+{
+ if (evlist->nr_entries)
+ parse_events__group_leader(&evlist->entries);
+}
+
int perf_evlist__add_default(struct perf_evlist *evlist)
{
struct perf_event_attr attr = {
@@ -757,21 +763,13 @@ 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 989bee9..d842931 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -82,7 +82,7 @@ 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);
@@ -126,4 +126,5 @@ void perf_evlist__splice_list_tail(struct perf_evlist *evlist,
struct list_head *list,
int nr_entries);
+void 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 876f639..7e8cc31 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -18,7 +18,6 @@
#include "../../include/linux/perf_event.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)
{
@@ -451,6 +450,7 @@ void perf_evsel__delete(struct perf_evsel *evsel)
{
perf_evsel__exit(evsel);
close_cgroup(evsel->cgrp);
+ free(evsel->group_name);
free(evsel->name);
free(evsel);
}
@@ -526,9 +526,28 @@ 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;
+ int fd;
+
+ if (!leader)
+ return -1;
+
+ /*
+ * Leader must be already processed/open,
+ * if not it's a bug.
+ */
+ BUG_ON(!leader->fd);
+
+ fd = FD(leader, cpu, thread);
+ BUG_ON(fd == -1);
+
+ return fd;
+}
+
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;
@@ -544,13 +563,15 @@ 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],
@@ -560,8 +581,9 @@ 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);
}
}
@@ -605,8 +627,7 @@ 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 */
@@ -616,23 +637,19 @@ 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 67cc503..b39f3db 100644
--- a/tools/perf/util/evsel.h
+++ b/tools/perf/util/evsel.h
@@ -66,6 +66,8 @@ struct perf_evsel {
void *data;
} handler;
bool supported;
+ struct perf_evsel *leader;
+ char *group_name;
};
struct cpu_map;
@@ -105,14 +107,11 @@ 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.c b/tools/perf/util/parse-events.c
index f62cbee..40ad468 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -547,14 +547,32 @@ int parse_events_add_pmu(struct list_head **list, int *idx,
pmu_event_name(head_config));
}
-int parse_events__modifier_group(struct list_head *list __used,
- char *event_mod __used)
+struct perf_evsel *parse_events__group_leader(struct list_head *list)
{
- return 0;
+ struct perf_evsel *evsel, *leader;
+
+ leader = list_entry(list->next, struct perf_evsel, node);
+ leader->leader = NULL;
+
+ list_for_each_entry(evsel, list, node)
+ if (evsel != leader)
+ evsel->leader = leader;
+
+ return leader;
}
-void parse_events__group(char *name __used, struct list_head *list __used)
+int parse_events__modifier_group(struct list_head *list,
+ char *event_mod)
{
+ return parse_events__modifier_event(list, event_mod);
+}
+
+void parse_events__group(char *name, struct list_head *list)
+{
+ struct perf_evsel *leader;
+
+ leader = parse_events__group_leader(list);
+ leader->group_name = name ? strdup(name) : NULL;
}
void parse_events_update_lists(struct list_head *list_event,
diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h
index c2f2ed9..fdbc72a 100644
--- a/tools/perf/util/parse-events.h
+++ b/tools/perf/util/parse-events.h
@@ -92,6 +92,7 @@ int parse_events_add_breakpoint(struct list_head **list, int *idx,
void *ptr, char *type);
int parse_events_add_pmu(struct list_head **list, int *idx,
char *pmu , struct list_head *head_config);
+struct perf_evsel *parse_events__group_leader(struct list_head *list);
void parse_events__group(char *name, struct list_head *list);
void parse_events_update_lists(struct list_head *list_event,
struct list_head *list_all);
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
Em Fri, Jun 29, 2012 at 11:08:26AM +0200, Jiri Olsa escreveu:
> This patch adds a functionality that allows to create event groups
> based on the way they are specified on the command line. Adding
> functionality to the '{}' group syntax introduced in earlier patch.
>
> The current '--group/-g' option behaviour remains intact. If you
> specify it for record/stat/top command, all the specified events
> become members of a single group with the first event as a group
> leader.
>
> With the new '{}' group syntax you can create group like:
> # perf record -e '{cycles,faults}' ls
All this cset comment makes for great tools/perf/Documentation material,
right? Can you try and send a patch to add this to the 'record' man
page?
On top of these, that I'm merging/testing now.
- Arnaldo
On Fri, Jun 29, 2012 at 01:47:36PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Fri, Jun 29, 2012 at 11:08:26AM +0200, Jiri Olsa escreveu:
> > This patch adds a functionality that allows to create event groups
> > based on the way they are specified on the command line. Adding
> > functionality to the '{}' group syntax introduced in earlier patch.
> >
> > The current '--group/-g' option behaviour remains intact. If you
> > specify it for record/stat/top command, all the specified events
> > become members of a single group with the first event as a group
> > leader.
> >
> > With the new '{}' group syntax you can create group like:
> > # perf record -e '{cycles,faults}' ls
>
> All this cset comment makes for great tools/perf/Documentation material,
> right? Can you try and send a patch to add this to the 'record' man
> page?
ok, I'll send doc patch with the next patchset
jirka