fvp_common.sh: Use shutdown_on_eot=1 FVP option only for codecov on OpenCI
Because otherwise we're hitting cases when short-running, little-output
tests run too fast and FVP terminates, before LAVA can establish telnet
connection and capture the output.
Signed-off-by: Paul Sokolovsky <paul.sokolovsky@linaro.org>
Change-Id: I7243e241283ba5dcd794bd475b3aee1013b37202
diff --git a/model/fvp_common.sh b/model/fvp_common.sh
index f31111c..399ee47 100644
--- a/model/fvp_common.sh
+++ b/model/fvp_common.sh
@@ -113,12 +113,18 @@
fi
# TFTF: There are two scenarions where simulation should be shutdown
-# when a EOT (ASCII 4) char is transmitted: on local or Open CI runs.
-# For the latter case, shutdown is required so further commands parse
-# or transfer any produced files during execution, i.e. trace code
-# coverage logs
+# when a EOT (ASCII 4) char is transmitted: on local or some OpenCI
+# runs. In the latter case, shutdown is required for code coverage
+# runs, so that FVP trace log was properly dumped (and parsed by LAVA).
+# Don't use it for other OpenCI runs, as it may lead to race condition
+# with LAVA's capturing of FVP output.
if echo "$RUN_CONFIG" | grep -iq 'tftf'; then
- if ! is_arm_jenkins_env || upon "$local_ci"; then
+ is_arm_env=0
+ if is_arm_jenkins_env; then
+ is_arm_env=1
+ fi
+
+ if [ "$is_arm_env" == "0" -a "$COVERAGE_ON" == "1" ] || upon "$local_ci"; then
cat <<EOF >>"$model_param_file"
-C bp.pl011_uart0.shutdown_on_eot=1
EOF