Re: [PATCH v3 2/3] perf record: Propagate exit status of a command line workload

[Jiri Olsa <jolsa@redhat.com>]

On Thu, Apr 24, 2014 at 10:27:33PM +0900, Namhyung Kim wrote:

SNIP

> -		rec->bytes_written / 24);
> +out_child:
> +	if (forks) {
> +		int exit_status;
>  
> -	return 0;
> +		if (!child_finished)
> +			kill(rec->evlist->workload.pid, SIGTERM);

also while at it.. do we want to force SIGKILL in case we dont
get any response for SIGTERM? so we dont get record hanging like
for following program:

---
#include <signal.h>

static void sig_handler(int sig)
{
}

int main(int argc, char **argv)
{
        signal(SIGINT, sig_handler);
        signal(SIGTERM, sig_handler);

        while (1) {}

        return 0;
}
---

the change would go into separate patch of course,
something like in patch below

thanks,
jirka


---
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 526edf5..a973ba5 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -479,9 +479,20 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 out_child:
 	if (forks) {
 		int exit_status;
+		int zleep = 0;
+
+		while (!child_finished) {
+			if (!zleep)
+				kill(rec->evlist->workload.pid, SIGTERM);
+			if (zleep == 2000) {
+				pr_info("Child killed by SIGKILL.\n");
+				kill(rec->evlist->workload.pid, SIGKILL);
+				break;
+			}
 
-		if (!child_finished)
-			kill(rec->evlist->workload.pid, SIGTERM);
+			usleep(1000);
+			zleep++;
+		}
 
 		wait(&exit_status);