perf stat: Expand events for each cgroup
Namhyung Kim
maintainerv3 cover perf stat: Expand events for each cgroup
(Resending with +LKML, sorry)
Hello,
When we profile cgroup events with perf stat, it's very annoying to specify events and cgroups on the command line as it requires the mapping between events and cgroups. (Note that perf record can use cgroup sampling but it's not usable for perf stat).
I guess most cases we just want to use a same set of events (N) for all cgroups (M), but we need to specify NxM events and NxM cgroups. This is not good especially when profiling large number of cgroups:
say M=200.
So I added --for-each-cgroup option to make it easy for that case. It will create NxM events from N events and M cgroups. One more upside is that it can handle metrics too.
For example, the following example measures IPC metric for 3 cgroups
$ cat perf-expand-cgrp.sh #!/bin/sh
METRIC=${1:-IPC}
CGROUP_DIR=/sys/fs/cgroup/perf_eventsudo mkdir $CGROUP_DIR/A $CGROUP_DIR/B $CGROUP_DIR/C
# add backgroupd workload for each cgroup echo $$ | sudo tee $CGROUP_DIR/A/cgroup.procs > /dev/null yes > /dev/null & echo $$ | sudo tee $CGROUP_DIR/B/cgroup.procs > /dev/null yes > /dev/null & echo $$ | sudo tee $CGROUP_DIR/C/cgroup.procs > /dev/null yes > /dev/null &
# run 'perf stat' in the root cgroup echo $$ | sudo tee $CGROUP_DIR/cgroup.procs > /dev/null perf stat -a -M $METRIC --for-each-cgroup A,B,C sleep 1
kill %1 %2 %3
sudo rmdir $CGROUP_DIR/A $CGROUP_DIR/B $CGROUP_DIR/C
$ ./perf-expand-cgrp.sh IPC
Performance counter stats for 'system wide':
11,284,850,010 inst_retired.any A # 2.71 IPC
4,157,915,982 cpu_clk_unhalted.thread A
11,342,188,640 inst_retired.any B # 2.72 IPC
4,173,014,732 cpu_clk_unhalted.thread B
11,135,863,604 inst_retired.any C # 2.67 IPC
4,171,375,184 cpu_clk_unhalted.thread C 1.011948803 seconds time elapsed
* Changes from v2:
- put relevant fields in evsel together (Jiri)
- add various error checks (Jiri)
- split cgroup open patch (Jiri)
* Changes from v1:
- rename the option to --for-each-cgroup (Jiri)
- copy evsel fields explicitly (Jiri)
- add libpfm4 test (Ian)
The code is available at 'perf/cgroup-multiply-v3' branch on
git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
Thanks
Namhyung
Namhyung Kim (5):
perf evsel: Add evsel__clone() function
perf stat: Add --for-each-cgroup option
perf tools: Copy metric events properly when expand cgroups perf tools: Allow creation of cgroup without open perf test: Add expand cgroup event test
tools/perf/builtin-stat.c | 21 ++- tools/perf/tests/Build | 1 + tools/perf/tests/builtin-test.c | 4 + tools/perf/tests/expand-cgroup.c | 241 +++++++++++++++++++++++++++++++ tools/perf/tests/tests.h | 1 + tools/perf/util/cgroup.c | 112 +++++++++++++- tools/perf/util/cgroup.h | 3 + tools/perf/util/evlist.c | 11 ++ tools/perf/util/evlist.h | 1 + tools/perf/util/evsel.c | 104 +++++++++++++ tools/perf/util/evsel.h | 93 +++++++----- tools/perf/util/metricgroup.c | 83 +++++++++++ tools/perf/util/metricgroup.h | 6 + tools/perf/util/stat.h | 1 + 14 files changed, 637 insertions(+), 45 deletions(-) create mode 100644 tools/perf/tests/expand-cgroup.c
Signature · 1 line
2.28.0.681.g6f77f65b4e-goog
Namhyung Kim
maintainerPATCH v3 1/5 perf evsel: Add evsel__clone() function
The evsel__clone() is to create an exactly same evsel from same attributes. The function assumes the given evsel is not configured yet so it cares fields set during event parsing. Those fields are now moved together as Jiri suggested. Note that metric events will be handled by later patch.
It will be used by perf stat to generate separate events for each cgroup.
--- tools/perf/util/evsel.c | 104 ++++++++++++++++++++++++++++++++++++++++ tools/perf/util/evsel.h | 93 ++++++++++++++++++++--------------- 2 files changed, 158 insertions(+), 39 deletions(-)
tools/perf/util/evsel.c +104 -0
@@ -331,6 +331,110 @@ struct evsel *evsel__new_cycles(bool precise) goto out; } +static int evsel__copy_config_terms (struct evsel *dst,struct evsel *src) +{ +struct evsel_config_term *pos, *tmp; + +list_for_each_entry (pos, &src->config_terms, list) { + tmp =malloc (sizeof (*tmp)); +if (tmp ==NULL ) +return -ENOMEM ; + + *tmp = *pos; +if (tmp->free_str) { + tmp->val.str =strdup (pos->val.str); +if (tmp->val.str ==NULL ) { +free (tmp); +return -ENOMEM ; + } + } +list_add_tail (&tmp->list, &dst->config_terms); + } +return 0 ; +} + +/** +* evsel__clone - create a new evsel copied from @orig +* @orig: original evsel +* +* The assumption is that @orig is not configured nor opened yet. +* So we only care about the attributes that can be set while it's parsed. +*/ +struct evsel *evsel__clone (struct evsel *orig) +{ +struct evsel *evsel; + +BUG_ON (orig->core.fd); +BUG_ON (orig->counts); +BUG_ON (orig->priv); +BUG_ON (orig->per_pkg_mask); + +/* cannot handle BPF objects for now */ +if (orig->bpf_obj) +return NULL ; + + evsel =evsel__new (&orig->core.attr); +if (evsel ==NULL ) +return NULL ; + + evsel->core.cpus =perf_cpu_map__get (orig->core.cpus); + evsel->core.own_cpus =perf_cpu_map__get (orig->core.own_cpus); + evsel->core.threads =perf_thread_map__get (orig->core.threads); + evsel->core.nr_members = orig->core.nr_members; + evsel->core.system_wide = orig->core.system_wide; + +if (orig->name) { + evsel->name =strdup (orig->name); +if (evsel->name ==NULL ) + goto out_err; + } +if (orig->group_name) { + evsel->group_name =strdup (orig->group_name); +if (evsel->group_name ==NULL ) + goto out_err; + } +if (orig->pmu_name) { + evsel->pmu_name =strdup (orig->pmu_name); +if (evsel->pmu_name ==NULL ) + goto out_err; + } +if (orig->filter) { + evsel->filter =strdup (orig->filter); +if (evsel->filter ==NULL ) + goto out_err; + } + evsel->cgrp =cgroup__get (orig->cgrp); + evsel->tp_format = orig->tp_format; + evsel->handler = orig->handler; + evsel->leader = orig->leader; + + evsel->max_events = orig->max_events; + evsel->tool_event = orig->tool_event; + evsel->unit = orig->unit; + evsel->scale = orig->scale; + evsel->snapshot = orig->snapshot; + evsel->per_pkg = orig->per_pkg; + evsel->percore = orig->percore; + evsel->precise_max = orig->precise_max; + evsel->use_uncore_alias = orig->use_uncore_alias; + evsel->is_libpfm_event = orig->is_libpfm_event; + + evsel->exclude_GH = orig->exclude_GH; + evsel->sample_read = orig->sample_read; + evsel->auto_merge_stats = orig->auto_merge_stats; + evsel->collect_stat = orig->collect_stat; + evsel->weak_group = orig->weak_group; + +if (evsel__copy_config_terms (evsel, orig) <0 ) + goto out_err; + +return evsel; + +out_err: +evsel__delete (evsel); +return NULL ; +} +/* * Returns pointer with encoded error via <linux/err.h> interface. */
tools/perf/util/evsel.h +54 -39
125 lines not on this page · raw message
Signature · 1 line
2.28.0.681.g6f77f65b4e-goog
Namhyung Kim
maintainerPATCH v3 2/5 perf stat: Add --for-each-cgroup option
In reply to Namhyung KimThe --for-each-cgroup option is a syntax sugar to monitor large number of cgroups easily. Current command line requires to list all the events and cgroups even if users want to monitor same events for each cgroup. This patch addresses that usage by copying given events for each cgroup on user's behalf.
For instance, if they want to monitor 6 events for 200 cgroups each they should write 1200 event names (with -e) AND 1200 cgroup names (with -G) on the command line. But with this change, they can just specify 6 events and 200 cgroups with a new option.
A simpler example below: It wants to measure 3 events for 2 cgroups ('A' and 'B'). The result is that total 6 events are counted like below.
$ ./perf stat -a -e cpu-clock,cycles,instructions --for-each-cgroup A,B sleep 1
Performance counter stats for 'system wide':
988.18 msec cpu-clock A # 0.987 CPUs utilized
3,153,761,702 cycles A # 3.200 GHz (100.00%)
8,067,769,847 instructions A # 2.57 insn per cycle (100.00%)
982.71 msec cpu-clock B # 0.982 CPUs utilized
3,136,093,298 cycles B # 3.182 GHz (99.99%)
8,109,619,327 instructions B # 2.58 insn per cycle (99.99%)1.001228054 seconds time elapsed
--- tools/perf/builtin-stat.c | 20 +++++++++- tools/perf/util/cgroup.c | 84 +++++++++++++++++++++++++++++++++++++++ tools/perf/util/cgroup.h | 1 + tools/perf/util/stat.h | 1 + 4 files changed, 105 insertions(+), 1 deletion(-)
tools/perf/builtin-stat.c +19 -1
@@ -1051,6 +1051,17 @@ static int parse_control_option(const struct option *opt,return evlist__parse_control (str, &config->ctl_fd, &config->ctl_fd_ack, &config->ctl_fd_close); } +static int parse_stat_cgroups (const struct option *opt, +const char *str,int unset) +{ +if (stat_config.cgroup_list) { +pr_err ("--cgroup and --for-each-cgroup cannot be used together\n" ); +return -1 ; + } + +return parse_cgroups (opt, str, unset); +} +static struct option stat_options[] = {OPT_BOOLEAN ('T' ,"transaction" , &transaction_run,"hardware transaction statistics" ),
@@ -1094,7 +1105,9 @@ static struct option stat_options[] = {OPT_STRING ('x' ,"field-separator" , &stat_config.csv_sep,"separator" ,"print counts with custom separator" ),OPT_CALLBACK ('G' ,"cgroup" , &evsel_list,"name" , -"monitor event in cgroup name only" , parse_cgroups), +"monitor event in cgroup name only" , parse_stat_cgroups), +OPT_STRING (0 ,"for-each-cgroup" , &stat_config.cgroup_list,"name" , +"expand events for each cgroup" ),OPT_STRING ('o' ,"output" , &output_name,"file" ,"output file name" ),OPT_BOOLEAN (0 ,"append" , &append_file,"append to the output file" ), OPT_INTEGER(0, "log-fd", &output_fd,
@@ -2234,6 +2247,11 @@ int cmd_stat(int argc, const char **argv)if (add_default_attributes ()) goto out; +if (stat_config.cgroup_list) { +if (evlist__expand_cgroup (evsel_list, stat_config.cgroup_list) <0 ) + goto out; + } +target__validate (&target);if ((stat_config.aggr_mode ==AGGR_THREAD ) && (target.system_wide))
tools/perf/util/cgroup.c +84 -0
102 lines not on this page · raw message
tools/perf/util/cgroup.h +1 -0
@@ -24,6 +24,7 @@ void cgroup__put(struct cgroup *cgroup);struct evlist ;struct cgroup *evlist__findnew_cgroup (struct evlist *evlist,const char *name); +int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups);void evlist__set_default_cgroup (struct evlist *evlist,struct cgroup *cgroup);
tools/perf/util/stat.h +1 -0
@@ -137,6 +137,7 @@ struct perf_stat_config {int ctl_fd;int ctl_fd_ack;bool ctl_fd_close; +const char *cgroup_list; };void perf_stat__set_big_num (int set);
Signature · 1 line
2.28.0.681.g6f77f65b4e-goog
Re: [PATCH 2/5] perf stat: Add --for-each-cgroup option
On Mon, Sep 21, 2020 at 06:46:07PM +0900, Namhyung Kim wrote:
SNIP
Quoted 45 lines
tools/perf/util/cgroup.c +45
-0+int evlist__expand_cgroup (struct evlist *evlist,const char *str) +{ +struct evlist *orig_list, *tmp_list; +struct evsel *pos, *evsel, *leader; +struct cgroup *cgrp =NULL ; +const char *p, *e, *eos = str +strlen (str); +int ret =-1 ; + +if (evlist->core.nr_entries ==0 ) { +fprintf (stderr,"must define events before cgroups\n" ); +return -EINVAL ; + } + + orig_list =evlist__new (); + tmp_list =evlist__new (); +if (orig_list ==NULL || tmp_list ==NULL ) { +fprintf (stderr,"memory allocation failed\n" ); +return -ENOMEM ; + } + +/* save original events and init evlist */ +perf_evlist__splice_list_tail (orig_list, &evlist->core.entries); + evlist->core.nr_entries =0 ; + +for (;;) { + p =strchr (str,',' ); + e = p ? p : eos; + +/* allow empty cgroups, i.e., skip */ +if (e - str) { +/* termination added */ +char *name =strndup (str, e - str); +if (!name) + goto out_err; + + cgrp =cgroup__new (name); +free (name); +if (cgrp ==NULL ) + goto out_err; + }else { + cgrp =NULL ; + } + + leader =NULL ; +evlist__for_each_entry (orig_list, pos) {
tools/perf/util/cgroup.c int parse_cgroups(const struct option *opt, const char *str, · line 255
Quoted 10 lines
tools/perf/util/cgroup.c +10
-0+ evsel =evsel__clone (pos); +if (evsel ==NULL ) + goto out_err; + +cgroup__put (evsel->cgrp); + evsel->cgrp =cgroup__get (cgrp); + +if (evsel__is_group_leader (pos)) + leader = evsel; + evsel->leader = leader;
hum, will this work if there's standalone event after group? like:
{cycles,instructions},cache-misses
cache-misses will get cycles as group leader no?
jirka
Quoted 6 lines
tools/perf/util/cgroup.c +6
-0+ +evlist__add (tmp_list, evsel); + } +/* cgroup__new() has a refcount, release it here */ +cgroup__put (cgrp); + nr_cgroups++;
SNIP
Namhyung Kim
maintainerRe: [PATCH 2/5] perf stat: Add --for-each-cgroup option
Quoted 63 lines · Jiri Olsa
On Mon, Sep 21, 2020 at 06:46:07PM +0900, Namhyung Kim wrote:
SNIP
Quoted 45 lines
tools/perf/util/cgroup.c +45
-0+int evlist__expand_cgroup (struct evlist *evlist,const char *str) +{ +struct evlist *orig_list, *tmp_list; +struct evsel *pos, *evsel, *leader; +struct cgroup *cgrp =NULL ; +const char *p, *e, *eos = str +strlen (str); +int ret =-1 ; + +if (evlist->core.nr_entries ==0 ) { +fprintf (stderr,"must define events before cgroups\n" ); +return -EINVAL ; + } + + orig_list =evlist__new (); + tmp_list =evlist__new (); +if (orig_list ==NULL || tmp_list ==NULL ) { +fprintf (stderr,"memory allocation failed\n" ); +return -ENOMEM ; + } + +/* save original events and init evlist */ +perf_evlist__splice_list_tail (orig_list, &evlist->core.entries); + evlist->core.nr_entries =0 ; + +for (;;) { + p =strchr (str,',' ); + e = p ? p : eos; + +/* allow empty cgroups, i.e., skip */ +if (e - str) { +/* termination added */ +char *name =strndup (str, e - str); +if (!name) + goto out_err; + + cgrp =cgroup__new (name); +free (name); +if (cgrp ==NULL ) + goto out_err; + }else { + cgrp =NULL ; + } + + leader =NULL ; +evlist__for_each_entry (orig_list, pos) {tools/perf/util/cgroup.c int parse_cgroups(const struct option *opt, const char *str, · line 255
Quoted 10 lines
tools/perf/util/cgroup.c +10
-0+ evsel =evsel__clone (pos); +if (evsel ==NULL ) + goto out_err; + +cgroup__put (evsel->cgrp); + evsel->cgrp =cgroup__get (cgrp); + +if (evsel__is_group_leader (pos)) + leader = evsel; + evsel->leader = leader;hum, will this work if there's standalone event after group? like:
Quoted 3 lines
{cycles,instructions},cache-misses
cache-misses will get cycles as group leader no?
AFAIK non-group events are treated as a leader of its own group. So evsel__is_group_leader() will return true for cache-misses.
Thanks
Namhyung
Quoted 10 lines
Quoted 7 lines
tools/perf/util/cgroup.c +6
-0+ +evlist__add (tmp_list, evsel); + } +/* cgroup__new() has a refcount, release it here */ +cgroup__put (cgrp); + nr_cgroups++;SNIP
Re: [PATCH 2/5] perf stat: Add --for-each-cgroup option
Quoted 64 lines · Namhyung Kim
Quoted 63 lines · Jiri Olsa
On Mon, Sep 21, 2020 at 06:46:07PM +0900, Namhyung Kim wrote:
SNIP
Quoted 45 lines
tools/perf/util/cgroup.c +45
-0+int evlist__expand_cgroup (struct evlist *evlist,const char *str) +{ +struct evlist *orig_list, *tmp_list; +struct evsel *pos, *evsel, *leader; +struct cgroup *cgrp =NULL ; +const char *p, *e, *eos = str +strlen (str); +int ret =-1 ; + +if (evlist->core.nr_entries ==0 ) { +fprintf (stderr,"must define events before cgroups\n" ); +return -EINVAL ; + } + + orig_list =evlist__new (); + tmp_list =evlist__new (); +if (orig_list ==NULL || tmp_list ==NULL ) { +fprintf (stderr,"memory allocation failed\n" ); +return -ENOMEM ; + } + +/* save original events and init evlist */ +perf_evlist__splice_list_tail (orig_list, &evlist->core.entries); + evlist->core.nr_entries =0 ; + +for (;;) { + p =strchr (str,',' ); + e = p ? p : eos; + +/* allow empty cgroups, i.e., skip */ +if (e - str) { +/* termination added */ +char *name =strndup (str, e - str); +if (!name) + goto out_err; + + cgrp =cgroup__new (name); +free (name); +if (cgrp ==NULL ) + goto out_err; + }else { + cgrp =NULL ; + } + + leader =NULL ; +evlist__for_each_entry (orig_list, pos) {tools/perf/util/cgroup.c int parse_cgroups(const struct option *opt, const char *str, · line 255
Quoted 10 lines
tools/perf/util/cgroup.c +10
-0+ evsel =evsel__clone (pos); +if (evsel ==NULL ) + goto out_err; + +cgroup__put (evsel->cgrp); + evsel->cgrp =cgroup__get (cgrp); + +if (evsel__is_group_leader (pos)) + leader = evsel; + evsel->leader = leader;hum, will this work if there's standalone event after group? like:
Quoted 6 lines
Quoted 3 lines
{cycles,instructions},cache-misses
cache-misses will get cycles as group leader no?
AFAIK non-group events are treated as a leader of its own group. So
evsel__is_group_leader()will return true for cache-misses.
right, then it's fine
thanks,
jirka
Re: [PATCH 2/5] perf stat: Add --for-each-cgroup option
In reply to Namhyung KimHi,
Quoted 12 lines · Namhyung Kim
The --for-each-cgroup option is a syntax sugar to monitor large number of cgroups easily. Current command line requires to list all the events and cgroups even if users want to monitor same events for each cgroup. This patch addresses that usage by copying given events for each cgroup on user's behalf.
For instance, if they want to monitor 6 events for 200 cgroups each they should write 1200 event names (with -e) AND 1200 cgroup names (with -G) on the command line. But with this change, they can just specify 6 events and 200 cgroups with a new option.
Quoted 6 lines
A simpler example below: It wants to measure 3 events for 2 cgroups ('A' and 'B'). The result is that total 6 events are counted like below.
$ ./perf stat -a -e cpu-clock,cycles,instructions --for-each-cgroup A,B sleep 1
You could also do it by keeping the -G option and providing --for-each-cgroup as a modifier
of the behavior of -G:
$ ./perf stat -a -e cpu-clock,cycles,instructions --for-each-cgroup -G A,B sleep 1
That way, you do not have to handle the case where both are used. And it makes transitioning to the new style simpler, i.e., the -G option remains, just need
to trim the number of cgroups to 200 in your example.
Just a suggestion.
Quoted 200 lines
Performance counter stats for 'system wide': 988.18 msec cpu-clock A # 0.987 CPUs utilized 3,153,761,702 cycles A # 3.200 GHz (100.00%) 8,067,769,847 instructions A # 2.57 insn per cycle (100.00%) 982.71 msec cpu-clock B # 0.982 CPUs utilized 3,136,093,298 cycles B # 3.182 GHz (99.99%) 8,109,619,327 instructions B # 2.58 insn per cycle (99.99%) 1.001228054 seconds time elapsedSigned-off-by Namhyung Kim <[email protected]>--- tools/perf/builtin-stat.c | 20 +++++++++- tools/perf/util/cgroup.c | 84 +++++++++++++++++++++++++++++++++++++++ tools/perf/util/cgroup.h | 1 + tools/perf/util/stat.h | 1 + 4 files changed, 105 insertions(+), 1 deletion(-)tools/perf/builtin-stat.c +19
-1@@ -1051,6 +1051,17 @@ static int parse_control_option(const struct option *opt,return evlist__parse_control (str, &config->ctl_fd, &config->ctl_fd_ack, &config->ctl_fd_close); } +static int parse_stat_cgroups (const struct option *opt, +const char *str,int unset) +{ +if (stat_config.cgroup_list) { +pr_err ("--cgroup and --for-each-cgroup cannot be used together\n" ); +return -1 ; + } + +return parse_cgroups (opt, str, unset); +} +static struct option stat_options[] = {OPT_BOOLEAN ('T' ,"transaction" , &transaction_run,"hardware transaction statistics" ),@@ -1094,7 +1105,9 @@ static struct option stat_options[] = {OPT_STRING ('x' ,"field-separator" , &stat_config.csv_sep,"separator" ,"print counts with custom separator" ),OPT_CALLBACK ('G' ,"cgroup" , &evsel_list,"name" , -"monitor event in cgroup name only" , parse_cgroups), +"monitor event in cgroup name only" , parse_stat_cgroups), +OPT_STRING (0 ,"for-each-cgroup" , &stat_config.cgroup_list,"name" , +"expand events for each cgroup" ),OPT_STRING ('o' ,"output" , &output_name,"file" ,"output file name" ),OPT_BOOLEAN (0 ,"append" , &append_file,"append to the output file" ), OPT_INTEGER(0, "log-fd", &output_fd,@@ -2234,6 +2247,11 @@ int cmd_stat(int argc, const char **argv)if (add_default_attributes ()) goto out; +if (stat_config.cgroup_list) { +if (evlist__expand_cgroup (evsel_list, stat_config.cgroup_list) <0 ) + goto out; + } +target__validate (&target);if ((stat_config.aggr_mode ==AGGR_THREAD ) && (target.system_wide))tools/perf/util/cgroup.c +84
-0@@ -12,6 +12,7 @@#include <api/fs/fs.h> int nr_cgroups; +bool multiply_cgroup;static int open_cgroup (const char *name) {@@ -156,6 +157,10 @@ int parse_cgroups(const struct option *opt, const char *str,return -1 ; } +/* delay processing cgroups after it sees all events */ +if (multiply_cgroup) +return 0 ; +for (;;) { p =strchr (str,',' ); e = p ? p : eos;@@ -193,6 +198,85 @@ int parse_cgroups(const struct option *opt, const char *str,return 0 ; } +int evlist__expand_cgroup (struct evlist *evlist,const char *str) +{ +struct evlist *orig_list, *tmp_list; +struct evsel *pos, *evsel, *leader; +struct cgroup *cgrp =NULL ; +const char *p, *e, *eos = str +strlen (str); +int ret =-1 ; + +if (evlist->core.nr_entries ==0 ) { +fprintf (stderr,"must define events before cgroups\n" ); +return -EINVAL ; + } + + orig_list =evlist__new (); + tmp_list =evlist__new (); +if (orig_list ==NULL || tmp_list ==NULL ) { +fprintf (stderr,"memory allocation failed\n" ); +return -ENOMEM ; + } + +/* save original events and init evlist */ +perf_evlist__splice_list_tail (orig_list, &evlist->core.entries); + evlist->core.nr_entries =0 ; + +for (;;) { + p =strchr (str,',' ); + e = p ? p : eos; + +/* allow empty cgroups, i.e., skip */ +if (e - str) { +/* termination added */ +char *name =strndup (str, e - str); +if (!name) + goto out_err; + + cgrp =cgroup__new (name); +free (name); +if (cgrp ==NULL ) + goto out_err; + }else { + cgrp =NULL ; + } + + leader =NULL ; +evlist__for_each_entry (orig_list, pos) { + evsel =evsel__clone (pos); +if (evsel ==NULL ) + goto out_err; + +cgroup__put (evsel->cgrp); + evsel->cgrp =cgroup__get (cgrp); + +if (evsel__is_group_leader (pos)) + leader = evsel; + evsel->leader = leader; + +evlist__add (tmp_list, evsel); + } +/* cgroup__new() has a refcount, release it here */ +cgroup__put (cgrp); + nr_cgroups++; + +perf_evlist__splice_list_tail (evlist, &tmp_list->core.entries); + tmp_list->core.nr_entries =0 ; + +if (!p) { + ret =0 ; +break ; + } + str = p+1 ; + } + +out_err: +evlist__delete (orig_list); +evlist__delete (tmp_list); + +return ret; +} +static struct cgroup *__cgroup__findnew (struct rb_root *root,uint64_t id,bool create,const char *path) {tools/perf/util/cgroup.h +1
-0@@ -24,6 +24,7 @@ void cgroup__put(struct cgroup *cgroup);struct evlist ;struct cgroup *evlist__findnew_cgroup (struct evlist *evlist,const char *name); +int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups);void evlist__set_default_cgroup (struct evlist *evlist,struct cgroup *cgroup);tools/perf/util/stat.h +1
-1@@ -137,6 +137,7 @@ struct perf_stat_config {int ctl_fd;int ctl_fd_ack;bool ctl_fd_close; +const char *cgroup_list; };void perf_stat__set_big_num (int set); --2.28.0.681.g6f77f65b4e-goog
Namhyung Kim
maintainerRe: [PATCH 2/5] perf stat: Add --for-each-cgroup option
Hi Stephane,
Quoted 29 lines · Stephane Eranian
Hi,
Quoted 12 lines · Namhyung Kim
The --for-each-cgroup option is a syntax sugar to monitor large number of cgroups easily. Current command line requires to list all the events and cgroups even if users want to monitor same events for each cgroup. This patch addresses that usage by copying given events for each cgroup on user's behalf.
For instance, if they want to monitor 6 events for 200 cgroups each they should write 1200 event names (with -e) AND 1200 cgroup names (with -G) on the command line. But with this change, they can just specify 6 events and 200 cgroups with a new option.
Quoted 6 lines
A simpler example below: It wants to measure 3 events for 2 cgroups ('A' and 'B'). The result is that total 6 events are counted like below.
$ ./perf stat -a -e cpu-clock,cycles,instructions --for-each-cgroup A,B sleep 1You could also do it by keeping the -G option and providing --for-each-cgroup as a modifier
of the behavior of -G:$ ./perf stat -a -e cpu-clock,cycles,instructions --for-each-cgroup -G A,B sleep 1
Quoted 6 lines
That way, you do not have to handle the case where both are used. And it makes transitioning to the new style simpler, i.e., the -G option remains, just need
to trim the number of cgroups to 200 in your example.Just a suggestion.
Thanks for the suggestion. Actually that's the approach I took in my v1 submission. And Jiri suggested the current way. Personally I'm fine with either.
Thanks
Namhyung
Namhyung Kim
maintainerRe: [PATCH 2/5] perf stat: Add --for-each-cgroup option
In reply to Namhyung KimQuoted 29 lines · Namhyung Kim
The --for-each-cgroup option is a syntax sugar to monitor large number of cgroups easily. Current command line requires to list all the events and cgroups even if users want to monitor same events for each cgroup. This patch addresses that usage by copying given events for each cgroup on user's behalf.
For instance, if they want to monitor 6 events for 200 cgroups each they should write 1200 event names (with -e) AND 1200 cgroup names (with -G) on the command line. But with this change, they can just specify 6 events and 200 cgroups with a new option.
A simpler example below: It wants to measure 3 events for 2 cgroups ('A' and 'B'). The result is that total 6 events are counted like below.
$ ./perf stat -a -e cpu-clock,cycles,instructions --for-each-cgroup A,B sleep 1Performance counter stats for 'system wide': 988.18 msec cpu-clock A # 0.987 CPUs utilized 3,153,761,702 cycles A # 3.200 GHz (100.00%) 8,067,769,847 instructions A # 2.57 insn per cycle (100.00%) 982.71 msec cpu-clock B # 0.982 CPUs utilized 3,136,093,298 cycles B # 3.182 GHz (99.99%) 8,109,619,327 instructions B # 2.58 insn per cycle (99.99%) 1.001228054 seconds time elapsed
Quoted 7 lines
Signed-off-by Namhyung Kim <[email protected]>--- tools/perf/builtin-stat.c | 20 +++++++++- tools/perf/util/cgroup.c | 84 +++++++++++++++++++++++++++++++++++++++ tools/perf/util/cgroup.h | 1 + tools/perf/util/stat.h | 1 + 4 files changed, 105 insertions(+), 1 deletion(-)
Oh, I've realized that I didn't update the man page.. I'll send v4 soon.
Thanks
Namhyung
Namhyung Kim
maintainerPATCH v3 3/5 perf tools: Copy metric events properly when expand cgroups
In reply to Namhyung KimThe metricgroup__copy_metric_events() is to handle metrics events when expanding event for cgroups. As the metric events keep pointers to evsel, it should be refreshed when events are cloned during the operation.
The perf_stat__collect_metric_expr() is also called in case an event has a metric directly.
During the copy, it references evsel by index as the evlist now has cloned evsels for the given cgroup.
--- tools/perf/builtin-stat.c | 3 +- tools/perf/util/cgroup.c | 15 ++++++- tools/perf/util/cgroup.h | 4 +- tools/perf/util/evlist.c | 11 +++++ tools/perf/util/evlist.h | 1 + tools/perf/util/metricgroup.c | 83 +++++++++++++++++++++++++++++++++++ tools/perf/util/metricgroup.h | 6 +++ 7 files changed, 120 insertions(+), 3 deletions(-)
tools/perf/builtin-stat.c +2 -1
@@ -2248,7 +2248,8 @@ int cmd_stat(int argc, const char **argv) goto out;if (stat_config.cgroup_list) { -if (evlist__expand_cgroup (evsel_list, stat_config.cgroup_list) <0 ) +if (evlist__expand_cgroup (evsel_list, stat_config.cgroup_list, + &stat_config.metric_events) <0 ) goto out; }
tools/perf/util/cgroup.c +14 -1
@@ -3,6 +3,9 @@#include "evsel.h" #include "cgroup.h" #include "evlist.h" +#include "rblist.h" +#include "metricgroup.h" +#include "stat.h" #include <linux/zalloc.h> #include <sys/types.h> #include <sys/stat.h>
@@ -198,10 +201,12 @@ int parse_cgroups(const struct option *opt, const char *str,return 0 ; } -int evlist__expand_cgroup (struct evlist *evlist,const char *str) +int evlist__expand_cgroup (struct evlist *evlist,const char *str, +struct rblist *metric_events) {struct evlist *orig_list, *tmp_list;struct evsel *pos, *evsel, *leader; +struct rblist orig_metric_events;struct cgroup *cgrp =NULL ;const char *p, *e, *eos = str +strlen (str);int ret =-1 ;
@@ -221,6 +226,8 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str)/* save original events and init evlist */ perf_evlist__splice_list_tail (orig_list, &evlist->core.entries); evlist->core.nr_entries =0 ; + orig_metric_events = *metric_events; +rblist__init (metric_events);for (;;) { p =strchr (str,',' );
@@ -260,6 +267,11 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str)cgroup__put (cgrp); nr_cgroups++; +perf_stat__collect_metric_expr (tmp_list); +if (metricgroup__copy_metric_events (tmp_list, cgrp, metric_events, + &orig_metric_events) <0 ) +break ; +perf_evlist__splice_list_tail (evlist, &tmp_list->core.entries); tmp_list->core.nr_entries =0 ;
@@ -273,6 +285,7 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str) out_err:evlist__delete (orig_list);evlist__delete (tmp_list); +rblist__exit (&orig_metric_events);return ret; }
tools/perf/util/cgroup.h +3 -1
@@ -22,9 +22,11 @@ struct cgroup *cgroup__get(struct cgroup *cgroup);void cgroup__put (struct cgroup *cgroup);struct evlist ; +struct rblist ;struct cgroup *evlist__findnew_cgroup (struct evlist *evlist,const char *name); -int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups); +int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups, +struct rblist *metric_events);void evlist__set_default_cgroup (struct evlist *evlist,struct cgroup *cgroup);
tools/perf/util/evlist.c +11 -0
@@ -1964,3 +1964,14 @@ int evlist__ctlfd_process(struct evlist *evlist, enum evlist_ctl_cmd *cmd)return err; } + +struct evsel *evlist__get_evsel (struct evlist *evlist,int idx) +{ +struct evsel *evsel; + +evlist__for_each_entry (evlist, evsel) { +if (evsel->idx == idx) +return evsel; + } +return NULL ; +}
tools/perf/util/evlist.h +1 -0
@@ -386,4 +386,5 @@ int evlist__ctlfd_ack(struct evlist *evlist);#define EVLIST_ENABLED_MSG "Events enabled\n"#define EVLIST_DISABLED_MSG "Events disabled\n" +struct evsel *evlist__get_evsel (struct evlist *evlist,int idx);#endif /* __PERF_EVLIST_H */
tools/perf/util/metricgroup.c +83 -0
92 lines not on this page · raw message
tools/perf/util/metricgroup.h +6 -0
@@ -7,11 +7,13 @@#include <stdbool.h> #include "pmu-events/pmu-events.h" +struct evlist ;struct evsel ;struct evlist ;struct option ;struct rblist ;struct pmu_events_map ; +struct cgroup ;struct metric_event {struct rb_node nd;
@@ -55,4 +57,8 @@ void metricgroup__print(bool metrics, bool groups, char *filter,bool metricgroup__has_metric (const char *metric);int arch_get_runtimeparam (struct pmu_event *pe__maybe_unused );void metricgroup__rblist_exit (struct rblist *metric_events); + +int metricgroup__copy_metric_events (struct evlist *evlist,struct cgroup *cgrp, +struct rblist *new_metric_events, +struct rblist *old_metric_events);#endif
Signature · 1 line
2.28.0.681.g6f77f65b4e-goog
Re: [PATCH 3/5] perf tools: Copy metric events properly when expand cgroups
On Mon, Sep 21, 2020 at 06:46:08PM +0900, Namhyung Kim wrote:
SNIP
Quoted 42 lines
tools/perf/util/cgroup.c +6
-0@@ -260,6 +267,11 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str)cgroup__put (cgrp); nr_cgroups++; +perf_stat__collect_metric_expr (tmp_list); +if (metricgroup__copy_metric_events (tmp_list, cgrp, metric_events, + &orig_metric_events) <0 ) +break ; +perf_evlist__splice_list_tail (evlist, &tmp_list->core.entries); tmp_list->core.nr_entries =0 ;@@ -273,6 +285,7 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str) out_err:evlist__delete (orig_list);evlist__delete (tmp_list); +rblist__exit (&orig_metric_events);return ret; }tools/perf/util/cgroup.h +3
-1@@ -22,9 +22,11 @@ struct cgroup *cgroup__get(struct cgroup *cgroup);void cgroup__put (struct cgroup *cgroup);struct evlist ; +struct rblist ;struct cgroup *evlist__findnew_cgroup (struct evlist *evlist,const char *name); -int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups); +int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups, +struct rblist *metric_events);void evlist__set_default_cgroup (struct evlist *evlist,struct cgroup *cgroup);tools/perf/util/evlist.c +0
-0
tools/perf/util/evlist.c int evlist__ctlfd_process(struct evlist *evlist, enum evlist_ctl_cmd *cmd) · line 1968
Quoted 5 lines
tools/perf/util/evlist.c +2
-0@@ -1964,3 +1964,14 @@ int evlist__ctlfd_process(struct evlist *evlist, enum evlist_ctl_cmd *cmd)return err; } + +struct evsel *evlist__get_evsel (struct evlist *evlist,int idx)
perhaps evlist__find_evsel is better name?
we have get/put names for functions changing refcount
jirka
Quoted 9 lines
tools/perf/util/evlist.c +9
-0+{ +struct evsel *evsel; + +evlist__for_each_entry (evlist, evsel) { +if (evsel->idx == idx) +return evsel; + } +return NULL ; +}
SNIP
Namhyung Kim
maintainerRe: [PATCH 3/5] perf tools: Copy metric events properly when expand cgroups
Quoted 48 lines · Jiri Olsa
On Mon, Sep 21, 2020 at 06:46:08PM +0900, Namhyung Kim wrote:
SNIP
Quoted 43 lines
tools/perf/util/cgroup.c +6
-0@@ -260,6 +267,11 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str)cgroup__put (cgrp); nr_cgroups++; +perf_stat__collect_metric_expr (tmp_list); +if (metricgroup__copy_metric_events (tmp_list, cgrp, metric_events, + &orig_metric_events) <0 ) +break ; +perf_evlist__splice_list_tail (evlist, &tmp_list->core.entries); tmp_list->core.nr_entries =0 ;@@ -273,6 +285,7 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str) out_err:evlist__delete (orig_list);evlist__delete (tmp_list); +rblist__exit (&orig_metric_events);return ret; }tools/perf/util/cgroup.h +3
-1@@ -22,9 +22,11 @@ struct cgroup *cgroup__get(struct cgroup *cgroup);void cgroup__put (struct cgroup *cgroup);struct evlist ; +struct rblist ;struct cgroup *evlist__findnew_cgroup (struct evlist *evlist,const char *name); -int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups); +int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups, +struct rblist *metric_events);void evlist__set_default_cgroup (struct evlist *evlist,struct cgroup *cgroup);tools/perf/util/evlist.c +0
-0@@ -1964,3 +1964,14 @@ int evlist__ctlfd_process(struct evlist *evlist, enum evlist_ctl_cmd *cmd)
Quoted 7 lines
tools/perf/util/evlist.c int evlist__ctlfd_process(struct evlist *evlist, enum evlist_ctl_cmd *cmd) · line 1968
Quoted 4 lines
tools/perf/util/evlist.c +2
-0return err; } + +struct evsel *evlist__get_evsel (struct evlist *evlist,int idx)perhaps
evlist__find_evselis better name?
we have get/put names for functions changing refcount
Looks better, will change.
Thanks
Namhyung
Quoted 13 lines
Quoted 10 lines
tools/perf/util/evlist.c +9
-0+{ +struct evsel *evsel; + +evlist__for_each_entry (evlist, evsel) { +if (evsel->idx == idx) +return evsel; + } +return NULL ; +}SNIP
Namhyung Kim
maintainerPATCH v3 4/5 perf tools: Allow creation of cgroup without open
In reply to Namhyung KimThis is a preparation for a test case of expanding events for multiple cgroups. Instead of using real system cgroup, the test will use fake cgroups so it needs a way to have them without a open file descriptor.
--- tools/perf/builtin-stat.c | 2 +- tools/perf/util/cgroup.c | 19 ++++++++++++------- tools/perf/util/cgroup.h | 2 +- 3 files changed, 14 insertions(+), 9 deletions(-)
tools/perf/builtin-stat.c +1 -1
@@ -2249,7 +2249,7 @@ int cmd_stat(int argc, const char **argv)if (stat_config.cgroup_list) {if (evlist__expand_cgroup (evsel_list, stat_config.cgroup_list, - &stat_config.metric_events) <0 ) + &stat_config.metric_events, true) <0 ) goto out; }
tools/perf/util/cgroup.c +12 -7
@@ -52,7 +52,7 @@ static struct cgroup *evlist__find_cgroup(struct evlist *evlist, const char *strreturn NULL ; } -static struct cgroup *cgroup__new (const char *name) +static struct cgroup *cgroup__new (const char *name,bool do_open) {struct cgroup *cgroup =zalloc (sizeof (*cgroup));
@@ -62,9 +62,14 @@ static struct cgroup *cgroup__new(const char *name) cgroup->name =strdup (name);if (!cgroup->name) goto out_err; - cgroup->fd =open_cgroup (name); -if (cgroup->fd ==-1 ) - goto out_free_name; + +if (do_open) { + cgroup->fd =open_cgroup (name); +if (cgroup->fd ==-1 ) + goto out_free_name; + }else { + cgroup->fd =-1 ; + } }return cgroup;
@@ -80,7 +85,7 @@ struct cgroup *evlist__findnew_cgroup(struct evlist *evlist, const char *name) {struct cgroup *cgroup =evlist__find_cgroup (evlist, name); -return cgroup ?:cgroup__new (name); +return cgroup ?:cgroup__new (name, true); }static int add_cgroup (struct evlist *evlist,const char *str)
@@ -202,7 +207,7 @@ int parse_cgroups(const struct option *opt, const char *str, }int evlist__expand_cgroup (struct evlist *evlist,const char *str, -struct rblist *metric_events) +struct rblist *metric_events,bool open_cgroup) {struct evlist *orig_list, *tmp_list;struct evsel *pos, *evsel, *leader;
@@ -240,7 +245,7 @@ int evlist__expand_cgroup(struct evlist *evlist, const char *str,if (!name) goto out_err; - cgrp =cgroup__new (name); + cgrp =cgroup__new (name, open_cgroup);free (name);if (cgrp ==NULL ) goto out_err;
tools/perf/util/cgroup.h +1 -1
@@ -26,7 +26,7 @@ struct rblist;struct cgroup *evlist__findnew_cgroup (struct evlist *evlist,const char *name);int evlist__expand_cgroup (struct evlist *evlist,const char *cgroups, -struct rblist *metric_events); +struct rblist *metric_events,bool open_cgroup);void evlist__set_default_cgroup (struct evlist *evlist,struct cgroup *cgroup);
Signature · 1 line
2.28.0.681.g6f77f65b4e-goog
Namhyung Kim
maintainerPATCH v3 5/5 perf test: Add expand cgroup event test
In reply to Namhyung KimIt'll expand given events for cgroups A, B and C.
$ ./perf test -v expansion
69: Event expansion for cgroups :
--- start ---
test child forked, pid 983140
metric expr 1 / IPC for CPI
metric expr instructions / cycles for IPC
found event instructions
found event cycles
adding {instructions,cycles}:W
copying metric event for cgroup 'A': instructions (idx=0)
copying metric event for cgroup 'B': instructions (idx=0)
copying metric event for cgroup 'C': instructions (idx=0)
test child finished with 0
---- end ----
Event expansion for cgroups: Ok--- tools/perf/tests/Build | 1 + tools/perf/tests/builtin-test.c | 4 + tools/perf/tests/expand-cgroup.c | 241 +++++++++++++++++++++++++++++++ tools/perf/tests/tests.h | 1 + 4 files changed, 247 insertions(+) create mode 100644 tools/perf/tests/expand-cgroup.c
tools/perf/tests/Build +1 -0
@@ -61,6 +61,7 @@ perf-y += demangle-java-test.o perf-y += pfm.o perf-y += parse-metric.o perf-y += pe-file-parsing.o +perf-y += expand-cgroup.o $(OUTPUT)tests/llvm-src-base.c: tests/bpf-script-example.c tests/Build $(call rule_mkdir)
tools/perf/tests/builtin-test.c +4 -0
@@ -345,6 +345,10 @@ static struct test generic_tests[] = { .desc = "PE file support", .func = test__pe_file_parsing, }, + { + .desc = "Event expansion for cgroups", + .func = test__expand_cgroup_events, + }, { .func = NULL, },
tools/perf/tests/expand-cgroup.c new file +241 -0
241 lines not on this page · raw message
tools/perf/tests/tests.h +1 -0
@@ -123,6 +123,7 @@ const char *test__pfm_subtest_get_desc(int subtest);int test__pfm_subtest_get_nr (void );int test__parse_metric (struct test *test,int subtest);int test__pe_file_parsing (struct test *test,int subtest); +int test__expand_cgroup_events (struct test *test,int subtest);bool test__bp_signal_is_supported (void );bool test__bp_account_is_supported (void );
Signature · 1 line
2.28.0.681.g6f77f65b4e-goog