[RFC PATCH 2/2] perf_events: add support for per-cpu per-cgroup monitoring (v4)

[Stephane Eranian]

This perf tool patch adds the ability to filter monitoring based on container
groups (cgroups) for both perf stat and perf record. The cgroup to monitor are
passed via a new -G option followed by a list of cgroups.

The cgroup filesystem has to be mounted. The tool will find it automatically,
open the right file and pass the descriptor to perf_events.

In this fourth version, we fix the case where no event is specified yet -G
is used. In that case, the cgroup constraint is applied to all events.

Example:
$ perf stat -B -a -e cycles:u,cycles:u,cycles:u -G test1,,test2 -- sleep 1
 Performance counter stats for 'sleep 1':

      2,368,667,414  cycles                   test1
      2,369,661,459  cycles                  
      <not counted>  cycles                   test2

        1.001856890  seconds time elapsed

Signed-off-by: Stephane Eranian <eranian@xxxxxxxxxx>

---

diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index 3ee27dc..0f9b8c8 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -116,6 +116,12 @@ Do not update the builid cache. This saves some overhead in situations
 where the information in the perf.data file (which includes buildids)
 is sufficient.
 
+-G name::
+--cgroup name::
+monitor only in the container called "name". This option is available only in per-cpu
+mode. The cgroup filesystem must be mounted. All threads belonging to container "name"
+are monitored when they run on the monitored CPUs.
+
 SEE ALSO
 --------
 linkperf:perf-stat[1], linkperf:perf-list[1]
diff --git a/tools/perf/Documentation/perf-stat.txt b/tools/perf/Documentation/perf-stat.txt
index 4b3a2d4..4115f77 100644
--- a/tools/perf/Documentation/perf-stat.txt
+++ b/tools/perf/Documentation/perf-stat.txt
@@ -53,6 +53,11 @@ comma-sperated list with no space: 0,1. Ranges of CPUs are specified with -: 0-2
 In per-thread mode, this option is ignored. The -a option is still necessary
 to activate system-wide monitoring. Default is to count on all CPUs.
 
+-G name::
+--cgroup name::
+monitor only in the container called "name". This option is available only in per-cpu
+mode. The cgroup filesystem must be mounted. All threads belonging to container "name"
+are monitored when they run on the monitored CPUs.
 EXAMPLES
 --------
 
diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index d1db0f6..7b27c49 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -417,6 +417,7 @@ LIB_H += util/probe-finder.h
 LIB_H += util/probe-event.h
 LIB_H += util/pstack.h
 LIB_H += util/cpumap.h
+LIB_H += util/cgroup.h
 
 LIB_OBJS += $(OUTPUT)util/abspath.o
 LIB_OBJS += $(OUTPUT)util/alias.o
@@ -464,6 +465,7 @@ LIB_OBJS += $(OUTPUT)util/hist.o
 LIB_OBJS += $(OUTPUT)util/probe-event.o
 LIB_OBJS += $(OUTPUT)util/util.o
 LIB_OBJS += $(OUTPUT)util/cpumap.o
+LIB_OBJS += $(OUTPUT)util/cgroup.o
 
 BUILTIN_OBJS += $(OUTPUT)builtin-annotate.o
 
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index ff77b80..c1b330a 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -22,6 +22,7 @@
 #include "util/session.h"
 #include "util/symbol.h"
 #include "util/cpumap.h"
+#include "util/cgroup.h"
 
 #include <unistd.h>
 #include <sched.h>
@@ -229,6 +230,8 @@ static void create_counter(int counter, int cpu)
 	char *filter = filters[counter];
 	struct perf_event_attr *attr = attrs + counter;
 	struct perf_header_attr *h_attr;
+	unsigned long flags = 0;
+	int pid;
 	int track = !counter; /* only the first counter needs these */
 	int thread_index;
 	int ret;
@@ -286,6 +289,9 @@ static void create_counter(int counter, int cpu)
 		attr->sample_type	|= PERF_SAMPLE_CPU;
 	}
 
+	if (cgroups[counter])
+		flags = PERF_FLAG_PID_CGROUP;
+
 	attr->mmap		= track;
 	attr->comm		= track;
 	attr->inherit		= !no_inherit;
@@ -296,8 +302,13 @@ static void create_counter(int counter, int cpu)
 
 	for (thread_index = 0; thread_index < thread_num; thread_index++) {
 try_again:
+		if (cgroups[counter])
+			pid = cgroups_fd[counter];
+		else
+			pid = all_tids[thread_index];
+
 		fd[nr_cpu][counter][thread_index] = sys_perf_event_open(attr,
-				all_tids[thread_index], cpu, group_fd, 0);
+				pid, cpu, group_fd, flags);
 
 		if (fd[nr_cpu][counter][thread_index] < 0) {
 			int err = errno;
@@ -828,6 +839,9 @@ static const struct option options[] = {
 		    "don't sample"),
 	OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid,
 		    "do not update the buildid cache"),
+	OPT_CALLBACK('G', "cgroup", NULL, "name",
+		     "monitor in cgroup name only",
+		     parse_cgroups),
 	OPT_END()
 };
 
@@ -851,6 +865,9 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
 		write_mode = WRITE_FORCE;
 	}
 
+	if (nr_cgroups && !system_wide)
+		usage_with_options(record_usage, options);
+
 	symbol__init();
 	if (no_buildid)
 		disable_buildid_cache();
@@ -861,6 +878,9 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
 		attrs[0].config = PERF_COUNT_HW_CPU_CYCLES;
 	}
 
+	if (open_cgroups())
+		usage_with_options(record_usage, options);
+
 	if (target_pid != -1) {
 		target_tid = target_pid;
 		thread_num = find_all_tid(target_pid, &all_tids);
@@ -870,6 +890,7 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
 			usage_with_options(record_usage, options);
 		}
 	} else {
+		err = -ENOMEM;
 		all_tids=malloc(sizeof(pid_t));
 		if (!all_tids)
 			goto out_symbol_exit;
@@ -921,5 +942,6 @@ out_free_fd:
 	all_tids = NULL;
 out_symbol_exit:
 	symbol__exit();
+	close_cgroups();
 	return err;
 }
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index a6b4d44..9bde425 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -47,6 +47,7 @@
 #include "util/header.h"
 #include "util/cpumap.h"
 #include "util/thread.h"
+#include "util/cgroup.h"
 
 #include <sys/prctl.h>
 #include <math.h>
@@ -151,6 +152,8 @@ struct stats			runtime_branches_stats;
 static int create_perf_stat_counter(int counter)
 {
 	struct perf_event_attr *attr = attrs + counter;
+	unsigned long flags = 0;
+	int pid = -1;
 	int thread;
 	int ncreated = 0;
 
@@ -161,9 +164,13 @@ static int create_perf_stat_counter(int counter)
 	if (system_wide) {
 		int cpu;
 
+		if (cgroups[counter]) {
+			flags = PERF_FLAG_PID_CGROUP;
+			pid = cgroups_fd[counter];
+		}
 		for (cpu = 0; cpu < nr_cpus; cpu++) {
 			fd[cpu][counter][0] = sys_perf_event_open(attr,
-					-1, cpumap[cpu], -1, 0);
+					pid, cpumap[cpu], -1, flags);
 			if (fd[cpu][counter][0] < 0)
 				pr_debug(ERR_PERF_OPEN, counter,
 					 fd[cpu][counter][0], strerror(errno));
@@ -398,6 +405,9 @@ static void abs_printout(int counter, double avg)
 	else
 		fprintf(stderr, " %18.0f  %-24s", avg, event_name(counter));
 
+	if (cgroups[counter])
+		fprintf(stderr, " %s", cgroups[counter]);
+
 	if (MATCH_EVENT(HARDWARE, HW_INSTRUCTIONS, counter)) {
 		total = avg_stats(&runtime_cycles_stats);
 
@@ -433,8 +443,13 @@ static void print_counter(int counter)
 	int scaled = event_scaled[counter];
 
 	if (scaled == -1) {
-		fprintf(stderr, " %18s  %-24s\n",
+		fprintf(stderr, " %18s  %-24s",
 			"<not counted>", event_name(counter));
+
+		if (cgroups[counter])
+			fprintf(stderr, " %s", cgroups[counter]);
+
+		fprintf(stderr, "\n");
 		return;
 	}
 
@@ -454,7 +469,6 @@ static void print_counter(int counter)
 		fprintf(stderr, "  (scaled from %.2f%%)",
 				100 * avg_running / avg_enabled);
 	}
-
 	fprintf(stderr, "\n");
 }
 
@@ -545,6 +559,9 @@ static const struct option options[] = {
 		    "print large numbers with thousands\' separators"),
 	OPT_STRING('C', "cpu", &cpu_list, "cpu",
 		    "list of cpus to monitor in system-wide"),
+	OPT_CALLBACK('G', "cgroup", NULL, "name",
+		     "monitor in cgroup name only",
+		     parse_cgroups),
 	OPT_END()
 };
 
@@ -562,10 +579,23 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used)
 	if (run_count <= 0)
 		usage_with_options(stat_usage, options);
 
+	if (nr_cgroups && !system_wide)
+		usage_with_options(stat_usage, options);
+
 	/* Set attrs and nr_counters if no event is selected and !null_run */
 	if (!null_run && !nr_counters) {
 		memcpy(attrs, default_attrs, sizeof(default_attrs));
 		nr_counters = ARRAY_SIZE(default_attrs);
+		if (nr_cgroups == 1) {
+			for (i = 1; i < nr_counters; i++) {
+				cgroups[i] = strdup(cgroups[0]);
+				if (!cgroups[i]) {
+					close_cgroups();
+					return -ENOMEM;
+				}
+				nr_cgroups++;
+			}
+		}
 	}
 
 	if (system_wide)
@@ -612,6 +642,9 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used)
 	signal(SIGALRM, skip_signal);
 	signal(SIGABRT, skip_signal);
 
+	if (open_cgroups())
+		usage_with_options(stat_usage, options);
+
 	status = 0;
 	for (run_idx = 0; run_idx < run_count; run_idx++) {
 		if (run_count != 1 && verbose)
@@ -622,5 +655,7 @@ int cmd_stat(int argc, const char **argv, const char *prefix __used)
 	if (status != -1)
 		print_stat(argc, argv);
 
+	close_cgroups();
+
 	return status;
 }
--
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/