perf_events: add cgroup support to perf tool (v8)

index fcd29e8..b0c60bf 100644
@@ -23,6 +23,7 @@
 #include "util/session.h"
 #include "util/symbol.h"
 #include "util/cpumap.h"
+#include "util/cgroup.h"
 
 #include <unistd.h>
 #include <sched.h>

@@ -234,9 +235,10 @@ static void create_counter(struct perf_evsel *evsel, int cpu)
 	char *filter = evsel->filter;
 	struct perf_event_attr *attr = &evsel->attr;
 	struct perf_header_attr *h_attr;
+	unsigned long flags = 0;
 	int track = !evsel->idx; /* only the first counter needs these */
 	int thread_index;
-	int ret;
+	int ret, pid = -1;
 	struct {
 		u64 count;
 		u64 time_enabled;

@@ -313,6 +315,11 @@ static void create_counter(struct perf_evsel *evsel, int cpu)
 		attr->wakeup_events = 1;
 	}
 
+	if (evsel->cgrp) {
+		flags = PERF_FLAG_PID_CGROUP;
+		pid = evsel->cgrp->fd;
+	}
+
 	attr->mmap		= track;
 	attr->comm		= track;
 	attr->inherit		= !no_inherit;

@@ -325,7 +332,10 @@ retry_sample_id:
 
 	for (thread_index = 0; thread_index < threads->nr; thread_index++) {
 try_again:
-		FD(evsel, nr_cpu, thread_index) = sys_perf_event_open(attr, threads->map[thread_index], cpu, group_fd, 0);
+		if (!evsel->cgrp)
+			pid = threads->map[thread_index];
+
+		FD(evsel, nr_cpu, thread_index) = sys_perf_event_open(attr, pid, cpu, group_fd, flags);
 
 		if (FD(evsel, nr_cpu, thread_index) < 0) {
 			int err = errno;

@@ -884,6 +894,9 @@ const struct option record_options[] = {
 		    "do not update the buildid cache"),
 	OPT_BOOLEAN('B', "no-buildid", &no_buildid,
 		    "do not collect buildids in perf.data"),
+	OPT_CALLBACK('G', "cgroup", NULL, "name",
+		     "monitor event in cgroup name only",
+		     parse_cgroups),
 	OPT_END()
 };
 

@@ -908,6 +921,12 @@ int cmd_record(int argc, const char **argv, const char *prefix __used)
 		write_mode = WRITE_FORCE;
 	}
 
+	if (nr_cgroups && !system_wide) {
+		fprintf(stderr, "cgroup monitoring only available in"
+			" system-wide mode\n");
+		usage_with_options(record_usage, record_options);
+	}
+
 	symbol__init();
 
 	if (no_buildid_cache || no_buildid)

104 lines not on this page · raw message

161 lines not on this page · raw message

new file mode 100644
index 0000000..89acd6d
@@ -0,0 +1,17 @@
+#ifndef __CGROUP_H__
+#define __CGROUP_H__
+
+struct option;
+
+struct cgroup_sel {
+	char *name;
+	int fd;
+	int refcnt;
+};
+
+
+extern int nr_cgroups; /* number of explicit cgroups defined */
+extern void close_cgroup(struct cgroup_sel *cgrp);
+extern int parse_cgroups(const struct option *opt, const char *str, int unset);
+
+#endif /* __CGROUP_H__ */

index f5cfed6..65cab64 100644
@@ -53,6 +53,7 @@ void perf_evsel__delete(struct perf_evsel *evsel)
 {
 	assert(list_empty(&evsel->node));
 	xyarray__delete(evsel->fd);
+	close_cgroup(evsel->cgrp);
 	free(evsel);
 }
 

@@ -130,17 +131,28 @@ int __perf_evsel__read(struct perf_evsel *evsel,
 static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus,
 			      struct thread_map *threads)
 {
-	int cpu, thread;
+	unsigned long flags = 0;
+	int cpu, thread, pid = -1;
 
 	if (evsel->fd == NULL &&
 	    perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0)
 		return -1;
 
+	if (evsel->cgrp) {
+		flags = PERF_FLAG_PID_CGROUP;
+		pid = evsel->cgrp->fd;
+	}
 	for (cpu = 0; cpu < cpus->nr; cpu++) {
 		for (thread = 0; thread < threads->nr; thread++) {
+
+			if (!evsel->cgrp)
+				pid = threads->map[thread];
+
 			FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr,
-								     threads->map[thread],
-								     cpus->map[cpu], -1, 0);
+								     pid,
+								     cpus->map[cpu],
+								     -1,
+								     flags);
 			if (FD(evsel, cpu, thread) < 0)
 				goto out_close;
 		}

index b2d755f..2e2beb1 100644
@@ -6,6 +6,7 @@
 #include "../../../include/linux/perf_event.h"
 #include "types.h"
 #include "xyarray.h"
+#include "cgroup.h"
  
 struct perf_counts_values {
 	union {

@@ -32,6 +33,7 @@ struct perf_evsel {
 	struct perf_counts	*counts;
 	int			idx;
 	void			*priv;
+	struct cgroup_sel	*cgrp;
 };
 
 struct cpu_map;