[igt-dev] [PATCH i-g-t 1/2] runner: Inject a message when killing test to taints

[Petri Latvala]

Normally runner injecting a message to the test's stdout/stderr logs
has a race condition; The test outputs have special lines (subtest
starting/ending) and accidentally injecting stuff in between would
cause funky results.

When we're killing a test because the kernel got tainted, we know
already that we're not getting a subtest ending line and we can
inject, if we make sure we have newlines printed before and after the
injection.

Having a message in the stdout of the test will aid automatic bug
filtering.

Signed-off-by: Petri Latvala <petri.latvala at intel.com>
Cc: Arkadiusz Hiler <arkadiusz.hiler at intel.com>
---
 runner/executor.c | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

diff --git a/runner/executor.c b/runner/executor.c
index 06390262..7bb2b14c 100644
--- a/runner/executor.c
+++ b/runner/executor.c
@@ -1049,9 +1049,27 @@ static int monitor_output(pid_t child,
 				 * the result an incomplete we avoid
 				 * journaling a timeout here.
 				 */
-				if (is_tainted(taints))
+				if (is_tainted(taints)) {
 					exitline = EXECUTOR_EXIT;
 
+					/*
+					 * Also inject a message to
+					 * the test's stdout. As we're
+					 * shooting for an incomplete
+					 * anyway, we don't need to
+					 * care if we're not between
+					 * full lines from stdout. We
+					 * do need to make sure we
+					 * have newlines on both ends
+					 * of this injection though.
+					 */
+					dprintf(outputs[_F_OUT],
+						"\nrunner: This test was killed due to a kernel taint (0x%lx).\n",
+						taints);
+					if (settings->sync)
+						fdatasync(outputs[_F_OUT]);
+				}
+
 				dprintf(outputs[_F_JOURNAL], "%s%d (%.3fs)\n",
 					exitline,
 					status, time);
-- 
2.20.1