perf trace: Add support for tracing workload given by command line
authorNamhyung Kim <namhyung.kim@lge.com>
Fri, 5 Oct 2012 05:02:16 +0000 (14:02 +0900)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 5 Oct 2012 15:51:27 +0000 (12:51 -0300)
Now perf trace is able to trace specified workload by forking it like
perf record does.  And also finish the tracing if the workload quits or
gets SIGINT.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349413336-26936-5-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-trace.c

index da1183f..4e9320b 100644 (file)
@@ -52,6 +52,13 @@ struct trace {
        struct perf_record_opts opts;
 };
 
+static bool done = false;
+
+static void sig_handler(int sig __maybe_unused)
+{
+       done = true;
+}
+
 static int trace__read_syscall_info(struct trace *trace, int id)
 {
        char tp_name[128];
@@ -189,11 +196,12 @@ static int trace__sys_exit(struct trace *trace, struct perf_evsel *evsel,
        return 0;
 }
 
-static int trace__run(struct trace *trace)
+static int trace__run(struct trace *trace, int argc, const char **argv)
 {
        struct perf_evlist *evlist = perf_evlist__new(NULL, NULL);
        struct perf_evsel *evsel;
        int err = -1, i, nr_events = 0, before;
+       const bool forks = argc > 0;
 
        if (evlist == NULL) {
                printf("Not enough memory to run!\n");
@@ -214,6 +222,17 @@ static int trace__run(struct trace *trace)
 
        perf_evlist__config_attrs(evlist, &trace->opts);
 
+       signal(SIGCHLD, sig_handler);
+       signal(SIGINT, sig_handler);
+
+       if (forks) {
+               err = perf_evlist__prepare_workload(evlist, &trace->opts, argv);
+               if (err < 0) {
+                       printf("Couldn't run the workload!\n");
+                       goto out_delete_evlist;
+               }
+       }
+
        err = perf_evlist__open(evlist);
        if (err < 0) {
                printf("Couldn't create the events: %s\n", strerror(errno));
@@ -227,6 +246,10 @@ static int trace__run(struct trace *trace)
        }
 
        perf_evlist__enable(evlist);
+
+       if (forks)
+               perf_evlist__start_workload(evlist);
+
 again:
        before = nr_events;
 
@@ -272,8 +295,15 @@ again:
                }
        }
 
-       if (nr_events == before)
+       if (nr_events == before) {
+               if (done)
+                       goto out_delete_evlist;
+
                poll(evlist->pollfd, evlist->nr_fds, -1);
+       }
+
+       if (done)
+               perf_evlist__disable(evlist);
 
        goto again;
 
@@ -286,7 +316,8 @@ out:
 int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
 {
        const char * const trace_usage[] = {
-               "perf trace [<options>]",
+               "perf trace [<options>] [<command>]",
+               "perf trace [<options>] -- <command> [<options>]",
                NULL
        };
        struct trace trace = {
@@ -326,8 +357,6 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
        char bf[BUFSIZ];
 
        argc = parse_options(argc, argv, trace_options, trace_usage, 0);
-       if (argc)
-               usage_with_options(trace_usage, trace_options);
 
        err = perf_target__validate(&trace.opts.target);
        if (err) {
@@ -343,8 +372,8 @@ int cmd_trace(int argc, const char **argv, const char *prefix __maybe_unused)
                return err;
        }
 
-       if (perf_target__none(&trace.opts.target))
+       if (!argc && perf_target__none(&trace.opts.target))
                trace.opts.target.system_wide = true;
 
-       return trace__run(&trace);
+       return trace__run(&trace, argc, argv);
 }