perf stat: Don't show counter information when workload fails
authorArnaldo Carvalho de Melo <acme@redhat.com>
Sat, 28 Dec 2013 18:45:08 +0000 (15:45 -0300)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 13 Jan 2014 13:06:21 +0000 (10:06 -0300)
When starting a workload 'stat' wasn't using prepare_workload evlist
method's signal based exec() error reporting mechanism.

Use it so that the we don't report 'not counted' counters.

Before:

  [acme@zoo linux]$ perf stat dfadsfa
  dfadsfa: No such file or directory

   Performance counter stats for 'dfadsfa':

       <not counted>      task-clock
       <not counted>      context-switches
       <not counted>      cpu-migrations
       <not counted>      page-faults
       <not counted>      cycles
       <not counted>      stalled-cycles-frontend
     <not supported>      stalled-cycles-backend
       <not counted>      instructions
       <not counted>      branches
       <not counted>      branch-misses

         0.001831462 seconds time elapsed

  [acme@zoo linux]$

After:

  [acme@zoo linux]$ perf stat dfadsfa
  dfadsfa: No such file or directory
  [acme@zoo linux]$

Reported-by: David Ahern <dsahern@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-5yui3bv7e3hitxucnjsn6z8q@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-stat.c

index 106a5e5b7842434a659a0e597019cdbb78ae7bbf..1c76c7a66f78de2a0e9f4de6093c2ac9a1a16152 100644 (file)
@@ -509,6 +509,18 @@ static void handle_initial_delay(void)
        }
 }
 
+static volatile bool workload_exec_failed;
+
+/*
+ * perf_evlist__prepare_workload will send a SIGUSR1
+ * if the fork fails, since we asked by setting its
+ * want_signal to true.
+ */
+static void workload_exec_failed_signal(int signo __maybe_unused)
+{
+       workload_exec_failed = true;
+}
+
 static int __run_perf_stat(int argc, const char **argv)
 {
        char msg[512];
@@ -529,7 +541,7 @@ static int __run_perf_stat(int argc, const char **argv)
 
        if (forks) {
                if (perf_evlist__prepare_workload(evsel_list, &target, argv,
-                                                 false, false) < 0) {
+                                                 false, true) < 0) {
                        perror("failed to prepare workload");
                        return -1;
                }
@@ -584,6 +596,14 @@ static int __run_perf_stat(int argc, const char **argv)
        clock_gettime(CLOCK_MONOTONIC, &ref_time);
 
        if (forks) {
+               /*
+                * perf_evlist__prepare_workload will, after we call
+                * perf_evlist__start_Workload, send a SIGUSR1 if the exec call
+                * fails, that we will catch in workload_signal to flip
+                * workload_exec_failed.
+                */
+               signal(SIGUSR1, workload_exec_failed_signal);
+
                perf_evlist__start_workload(evsel_list);
                handle_initial_delay();
 
@@ -594,6 +614,10 @@ static int __run_perf_stat(int argc, const char **argv)
                        }
                }
                wait(&status);
+
+               if (workload_exec_failed)
+                       return -1;
+
                if (WIFSIGNALED(status))
                        psignal(WTERMSIG(status), argv[0]);
        } else {