From 8aacb573cd2c4a2f8160d99ff100ad0aa5e7859d Mon Sep 17 00:00:00 2001
From: Carl Love <cel@us.ibm.com>
Date: Thu, 25 Jul 2019 10:24:16 -0400
Subject: [PATCH] Only start the application if the perf events setup was
 successful

Changes the order of starting the application and performance events.

Given this change we have a new issue.  The issue is the routine
start_counting() calls fork, creating app_PID process.

The parent then tries to setup the performance events, then if the
performance events were setup correctly, app_PID is then told to start
before exiting.  If the performance counter setup fails, the app_PID is
left running.  The app_PID is never told to start the workload, which is
correct but we don't record the fact that app_PID is running.  The
error path then fails to kill app_PID in routine main(), in file
oprofile-git/pe_counting, at about lie 909 because the if statement

 if (startApp && app_started && (run_result != APP_ABNORMAL_END)) {

is false because app_started is False.

The fix, I believe, is to set app_started to True in the parent code if
the fork was successful.  With this fix, there is no orphan processes
left after ocount exits.
---
 pe_counting/ocount.cpp | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/pe_counting/ocount.cpp b/pe_counting/ocount.cpp
index 77177176..2470745d 100644
--- a/pe_counting/ocount.cpp
+++ b/pe_counting/ocount.cpp
@@ -242,6 +242,10 @@ bool start_counting(void)
 
 	// parent
 	int startup;
+	if ( app_PID != -1)
+		// app_PID child process created successfully
+		app_started = true;
+
 	if (startApp) {
 		if (read(app_ready_pipe[0], &startup, sizeof(startup)) == -1) {
 			perror("Internal error on app_ready_pipe");
@@ -297,7 +301,6 @@ bool start_counting(void)
 			perror("Internal error on start_app_pipe");
 			return false;
 		}
-		app_started = true;
 	}
 
 	return ret;
-- 
2.21.0