The POWER_CYCLE fallback added to reboot() flushes monitor output at the
wrong time. In the untimed reboot path, flushing immediately after
start_monitor() can consume the first output from the new boot before
monitor() begins reading it. In the timed path, flushing after POWER_CYCLE
can eat the "Linux version" banner or REBOOT_SUCCESS_LINE from the new
kernel.

That makes ktest miss the boot it is waiting for and can trigger an
unnecessary second power cycle.

Start the monitor before POWER_CYCLE so the reference counting stays
balanced, but only flush when reboot() was asked to wait for a timed
reboot. Perform that flush before issuing POWER_CYCLE so it drains stale
output from the old kernel instead of consuming the next boot.

Signed-off-by: Ricardo B. Marlière <rbm@suse.com>
---
 tools/testing/ktest/ktest.pl | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/tools/testing/ktest/ktest.pl b/tools/testing/ktest/ktest.pl
index 9d6f50045dbd..bd2e2311884c 100755
--- a/tools/testing/ktest/ktest.pl
+++ b/tools/testing/ktest/ktest.pl
@@ -1499,12 +1499,13 @@ sub reboot {
     }
 
     if ($powercycle) {
-	run_command "$power_cycle";
-
 	start_monitor;
-	# flush out current monitor
-	# May contain the reboot success line
-	wait_for_monitor 1;
+	if (defined($time)) {
+		# Flush stale console output from the old kernel before power-cycling.
+		wait_for_monitor 1;
+	}
+
+	run_command "$power_cycle";
 
     } else {
 	# Make sure everything has been written to disk

-- 
2.53.0